Installation issues

How can I make PerformanceTest portable on a USB drive?

PerformanceTest can be setup to be portable and run from a USB flash drive. To do this you first need to install PerformanceTest in the normal fashion to a fixed hard drive. Once installed you can then also make an additional portable copy.

From within PerformanceTest use the menu option "File"->"Install PerformanceTest to a USB drive".

From the "Install PerformanceTest to a USB drive" window, you need to specify:

1. The USB drive and directory you want to install PerformanceTest to. For example, "F:\PerformanceTest". PerformanceTest will create the directory if it does not exist.
2. The type of installation. If you have a license key, then select Licensed, otherwise select Evaluation for a trial period.
3. If you selected a "Licensed" installation type, then enter the Username and Key.

When you select install, PerformanceTest will create the directory on the USB drive (e.g. F:\PerformanceTest), copy all of the files from the PerformanceTest directory (e.g. C:\Program Files\PerformanceTest) to the USB drive (e.g. F:\PerformanceTest) and install the license information onto the USB drive.

During install I get the error, The drive or UNC share you selected does not exist?

Error: The drive or UNC share you selected does not exist

The full error message displayed by the installer is:
"The drive or UNC share you selected does not exist or is not accessible. Please select another"

This error can happen when the user previously performed an installation of the PerformanceTest software to a removable drive (such as a USB drive) then removed the drive. Subsequent installations of the same software on the same machine attempt to update the software on the same drive and same folder (there is no prompt for an installation folder). Because the drive is not connected, an error results.

The solution is to uninstall PerformanceTest first from the Window's "Programs and features" window. An alternative solution is to use the /DIR="<path>" command line option with the installer.

After the software is uninstalled (even with the USB drive not connected) the installer will once again prompt the user for an installation folder.

This error does not occur if you follow our recomended procedure for installing PerformanceTest to a USB drive.

How Can I Setup PerformanceTest to be portable on a USB drive?

PerformanceTest can be setup to be portable and run from a USB flash drive. To do this you first need to install PerformanceTest in the normal fashion to a fixed hard drive. Once installed you can then also make an additional portable copy.

From within PerformanceTest use the menu option "File"->"Install PerformanceTest to a USB drive".

From the "Install PerformanceTest to a USB drive" window, you need to specify:

1. The USB drive and directory you want to install PerformanceTest to. For example, "F:\PerformanceTest". PerformanceTest will create the directory if it does not exist.
2. The type of installation. If you have a license key, then select Licensed, otherwise select Evaluation for a trial period.
3. If you selected a "Licensed" installation type, then enter the Username and Key.

When you select install, PerformanceTest will create the directory on the USB drive (e.g. F:\PerformanceTest), copy all of the files from the PerformanceTest directory (e.g. C:\Program Files\PerformanceTest) to the USB drive (e.g. F:\PerformanceTest) and install the license information onto the USB drive.

General Troubleshooting and Support

PerformanceTest crashes when launched, possible causes and solutions.

There are various reasons PerformanceTest might not launch correctly

I have Riva Tuner V6 installed

RivaTuner is a 3rd party graphic card software utility. It does frame rate display and video capture. Riva Tuner is also called Afterburner and Riva Tuner Statistics Server (RTSS). Unfortunately it is also buggy and can cause some 3D applications to crash.

The observed crash occurs in Win10 in d3d9.dll (version 10.0.15063), which Riva presumably hooks into. Other older versions of d3d9.dll are OK it seems.

Crash dump information is
EXCEPTION_RECORD: (.exr -1)
ExceptionAddress: 00007ffe2d52fd0c (d3d9!CSwapChain::Reset)
ExceptionCode: c0000005 (Access violation)

d3d9!CSwapChain::Reset
d3d9!CBaseDevice::ResetMain+0x921
d3d9!CBaseDevice::Reset+0xcc
RTSSHooks64+0x195c0

Another crash dump seen caused by Riva Tuner
ExceptionAddress: 00007ff96cb9c416 (d3d9!CBaseDevice::GetBackBuffer+0x00000000000000c6))
ExceptionCode: c0000005 (Access violation)

STACK_TEXT:
00000000`00000000 00000000`00000000 memory_corruption!d3d9+0x0

Uninstalling RivaTuner / Afterburner can fix the problem. Copying a old version of d3d9.dll (e.g. version 10.0.14393) into the PerformanceTest install folder can also fix the problem.

Update: After contacting the developers of Riva, they have released a fix in their V7 Riva Tuner release. The change from their release notes are, "Hook engine is now using alternate shorter x64 hook trampoline to provide compatibility with Windows 10 Creators Update Direct3D9 runtimes." So this should also fix the problem.

Anti-virus software

Some Anti-Virus/Internet Security configurations have been known in the past to prevent PerformanceTest from running without displaying any message letting the user know what has happened (so no crash, and no error message). Additionally PerformanceTest uses the Internet to download baselines from the web, you may need to unblock HTTP(port 80) for PerformanceTest in your firewall for this functionality to work. This is a rare issue associated with false postives in the anti-virus software and hasn't been seen for the last few years.

No 3D support

PerformanceTest used 3D graphics. But in some cases a system might have no support for 3D graphics. Possible reasons include not having the correct video card device driver installed, using remote desktop software or running in a virtual machine. PerformanceTest makes an attempt to detect the lack of 3D support and antomatically switch to a 2D interface. But you can manually force this behaviour with /NO3D command line flag at start-up if desired.

Failure to collect system information

At startup PerformanceTest attempts to collect information about the system. The CPU type, the RAM type, temperature information and lots of other details. Rarely this process can fail. The following command line flags can be used to disable individual sections of the system information gathering, for faster startup or for bypassing sections that may have a problem.

/DontGatherGraphics
/DontGatherUSB
/DontGatherDisk
/DontGatherSMART
/DontGatherMemory
/DontGatherMemorySPD
/DontGatherWMI
/DontGatherSMBIOS
/DontGatherTemperature

If you do experience this problem like this, we would like to know about it (maybe we can fix it).

Debugging other issues

Please see next item How to collect a debug log to help us with debugging. for details on helping us collect a debug log to trace this problem.

How to collect a debug log to help us with debugging.

Overview

We have had a number of reports of PerformanceTest not starting correctly. The startup process seems to freeze either permanently and in some cases it may crash. We have done extensive testing on 100's of different computers, in various configurations. We have found that the main reason for this is corrupt SMBIOS (System Management BIOS) in the PC.

We have added two command line options to PerformanceTest to help with this startup problem.

DEBUGMODE: Creates a debug log file that can be sent to us for investigation;

SAFEMODE: Skips the loading of SMBIOS system information on startup.

This page describes the process needed to create a startup log file from PerformanceTest and starting PerformanceTest in Safe mode.

Debug mode

Placing the "DEBUGMODE" text on the PerformanceTest command line will result in two log files being created.

In older  versions of Windows you can edit the command line by finding PerformanceTest in the start menu then right clicking to edit the shortcut properties.

In Windows 10 you can right click on the PerformanceTest icon in the start menu, choose "Open File Location" and the right click on the PerformanceTest shortcut to open the properties. After adding the DEBUGMODE parameter you will likely see a warning that administrator permissions are required so click the "continue" button.

Log file

The log file is created in the PerformanceTest executable directory or My Documents. For PerformanceTest V6.0 (and earlier) this is, C:\Program Files\PerformanceTest. For PerformanceTest V6.1 (and later) this is, C:\Documents and Setting\<username>\My Documents\Passmark\Performance Test.

The name of the first log file is "PerfTestLog.txt" and the name of the second log file is "SysInfoLog.txt"

These are the files you should send us. Once you have created these files you should mail them to us at help@passmark.com . A startup log file can help us locate the fault in the software.

Example PerfTestLog.txt file

Safe mode

Placing the "SAFEMODE" text on the PerformanceTest command line will result in PerformanceTest starting without trying to obtain SMBIOS information. This may help with problems starting PerformanceTest, due to corrupt SMBIOS or SMBIOS not configured by the OEM. This command line parameter is only compatible with DEBUGMODE.
eg. PT.exe SAFEMODE

Are Windows 95, 98, ME, or 2000 supported?

From Version 4.0 onwards, PerformanceTest no longer supports Windows 95 and Windows NT4. From V6.1 build 1002 (May 2007) onwards, support for Windows 98 and ME was also dropped. Windows 2000 support was dropped in V8.0, as well as support for XP installs without service pack 3 installed.

Some older, unsupported versions of the software are available for download at the bottom of the download page. If you are using Windows 2000 or Windows 98 but don't have DirectX 9, it can be downloaded from the Microsoft web site.

Are Windows 8 & 10 supported?

From version 8.0 of PerformanceTest Windows Server 2012, Windows 8 and Windows 10 are supported

From version 7.0 of PerformanceTest, Windows server 2008 64bit and Windows 7 64bit are also supported.

From version 6.0 of PerformanceTest, PerformanceTest supports 64bit operating systems. Including Windows XP 64bit and Vista 64bit.

All 32bit versions of XP, Vista and Windows 7 are also supported.

I don't understand the results. What do all these numbers mean?

What Do All These Numbers Mean? The short version.

If you don't have a lot of computer knowledge interpreting the PerformanceTest results can be confusing. There are however a few simple concepts that can help you.

• The bigger the number the faster the computer.
• The results are not a percentage figure. They are relative figures.
• A computer with a CPU result of 4000 can process roughly twice as much data as a computer with a result of 2000.
• The overall Passmark Rating is not an average or a sum of the sub scores. This rating is limited by the weakest component in the system. So even if you had an extremely really fast CPU but only average memory and HDD then your overall score would still only be average.

The longer version...

PerformanceTest executes a collection of different tests on your computer to test different aspects of it's performance. There is a suite of tests for the CPU, Disk, Memory, 3D graphics and 2D graphics. For each suite there is a "Mark" value. For example the CPUmark. These mark values are then combined into a single overall score called the PassMark rating.

The CPUmark value is a measure of the CPU's performance. The PassMark rating is a measure of the entire system's performance. If you want to understand how all the individual scores are combined into Mark values you can find the PerformanceTest formula documented here.

You can find a chart of all the CPUMark values on the CPUbenchmark.net web site. A the time of writing 20,000+ was a high CPUMark while 8000 was more typical for a newish machine.

The Mark values are good for a quick assessment of the hardware's performance. However people use computers in different ways with different software. While an attempt was made by us, the developer, to write benchmark code that resembled real life code used in real applications, it is impossible for any benchmark software to exactly reproduce any particular individual's usage patterns. Some computers are used for gaming, some for web servers, some for office tasks. So you need to apply some common sense when interpreting the results. For example the 3DMark value isn't particularly relevant to an office worker.

For more details of the benchmark tests performed, see the Help file included with the PerformanceTest software and the CPU test description page, Graphics test description page, Disk test description page and RAM test description page

Double the score, double the performance?

If the Mark value is doubled, does this mean double the performance?

The vague wishy washy answer is: Yes it does, some of the time, at least for a limited set of circumstances.

Taking the CPUMark as an example. The CPUMark score is mostly made up of benchmark algorithms that A) execute almost exclusively on the CPU and B) Fully uses the all the CPUs cores that are available. There isn't any point, for example, having a CPU benchmark whose result is linked to the speed of the hard disk. In more technical terms the CPU benchmark is CPU bound. However many real world applications are not CPU bound. They spend some of their time waiting for the hard drive to read a file, some of their time receiving data from the Internet, some of their time updating the display, etc. Also many real world applications are not very well "threaded" and only run on one CPU core. So for these applications you won't see double the performance from a doubling in the CPUMark.

For poorly threaded applications that run on a single CPU core, it makes more sense to look at the single threaded benchmark chart. as this will give a much more realistic indication of performance compared to the main CPU benchmark charts.

Also the CPU test has a small dependence on the RAM speed, so at least for the faster CPUs, better RAM can make the CPU look slightly faster. Likewise the 3D, 2D & RAM tests have some dependencies on the CPU speed. So upgrading to a new video card which in theory is double the power, might not give the desired results if the performance is being bottlenecked by a slow CPU.

The calculated CPU clock speed is not correct. Why?

Why is my Calculated CPU Clock Speed Incorrect?

There are a few reasons why your calculated CPU clock speed may differ from what is expected:

1. CPUs made for portable computers can decrease their clock speed to lower power usage. For example, a Pentium III 1000 Mobile CPU may be measured as having a clock speed of 730Mhz at the time the measurement is done.
2. Newer CPUs also have a feature where they may over-clock some of their cores when others aren't in use.
3. The speed rating values that AMD assign to their Athlon CPU's do not correspond to their clock speed. For example, a Athlon 2000+ runs at a clock speed of 1660Mhz. This is also the case for the older Cyrix 150+ & 200+ CPUs

For reasons 1 and 2 we are constantly endeavouring to better detect when this is the case and use other methods to find the correct speed. We believe we can do this in most case with recent versions of PerformanceTest.

My video card name is not displayed correctly or appears strange.

In some cases the video card name may not be retrieved correctly or appear similair to "%iHSWGT15D%". We have found this is usually the result of an out of date video card driver. Check the manufactures website for the most recent version of the driver.

Using the default Micorosft supplied driver can also lead to an incorrect video card name, and will severely lower the test score for the video card. Again downloading the most recent driver available should fix this problem.

How can I test mapped network drives in the advanced disk test?

Due the way Windows User Account Control (UAC) works programs like PerformanceTest, that run as elevated administrator, are unable to access mapped network drives that were created by other users.

The suggested workaround is to edit/create the "EnableLinkedConnections" registry value at "HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows\CurrentVersion\Policies\System" and set it to a value of 1.

There is more information available here.

Your benchmark vs model average score differences

When viewing and comparing your benchmark result on the "Your score vs the same model" screen there are a few things to note.

This score is an average of nearly all the results in the database, including systems that contain more than one CPU (eg multi cpu capable Xeon systems) and ones that are overclocked. So on multi cpu capable systems (eg Xeon) with only one CPU the result will look a little slower and on multi CPU systems the scores will look closer to the average.

It uses a different script to the result that are displayed at http://www.cpubenchmark.net/ and generally the average result displayed in PerformanceTest will be higher than the results used in the online charts as not as many results are filtered out.

Test Performance and Optimization

How can I make my PC run faster?

We receive many questions from users about poor PC performance, and whether there is anything a user can do to improve the speed of their machine. We have written this article as a quick, troubleshooting checklist to assist you in locating causes of and finding solutions for poor PC performance. While this article does not directly attempt to address the other common question we receive, "Why is my PC Crashing?", we believe that some of these solutions could assist users with crash issues.

Problem: Spyware, Viruses and Other Malicious Software

Solution: Viruses, spyware and malware running on your PC will consume CPU time, RAM and network bandwidth. We recommend that you get a good antivirus and anti-spyware scanning program. In some cases of bad infections (e.g. multiple viruses), it might be quicker, safer and more effective to reformat the hard disk and re-install the complete operating system.

Problem: Disk Fragmentation

Solution: Files can be written and read from the disk faster if they are stored in a continuous sequence of bytes on the disk. However, as files on the hard disk are updated, deleted and created, the file clusters become fragmented which makes disk access slower. Defragmenting your hard disk every couple of months will help keep your system running at optimal performance. Most operating systems have a built-in disk defragmenter. Some operating systems will automatically defragment during machine idle.

Note that for Solid State Hard disks this is not relevant.

Problem: Slow 3D Graphics

Solution: Make sure you have the latest copy of the device driver for your video card and the most recent version of DirectX. Video device drivers are complicated pieces of software and as a result, it is very common that they are buggy or unreliable. If you have the latest device driver and are still encountering problems, consider rolling back to a previous version as it may be more stable.

Also check you video hardware acceleration is turned on from the Windows, "display properties" dialog and in DirectX (use Dxdiag.exe for this).

Problem: Slow 3D graphics (Onboard)

Solution: All onboard graphics solutions are slow when compared to non-integrated solutions. The onboard chipsets may have performance-driven names like "Express Chipset" or "Extreme Chipset" but they are not designed for high performance; they are designed for an economical price. If you are a gamer, you may want to consider upgrading to a mid-tier or better current graphics card for optimal performance.

Problem: 2D graphics are slow. 3D graphics are also slow in windowed mode, but seem okay in full screen mode

Solution: We have seen some Windows skinning applicatons (such as Windows Blinds from Stardock) have a dramatically negative effect on the 2D and 3D video performance. The full-screen 3D results are not affected as the normal Windows interface is hidden when an application is in full-screen mode. Uninstalling or disabling Windows Blinds will fix this performance problem.

Problem: The full capacity of my CPUs / CPU cores are not being used

Solution: If you are running PerformanceTest then check the number of processes in the Edit / Preferences window matches the number of CPU's and cores that you have in the machine. So as an example, having two Dual core CPUs mean you want four processes to max out the CPUs. If you still appear to be missing a CPU, check BIOS settings In some cases, it might be possible to disable CPUs or Cores from BIOS.

Problem: CPU is slow all the time

Solution: You might be running your CPU in low power mode, which reduces the clock speed to save electricity. Check your BIOS settings for this and other incorrect settings (such as accidental under-clocking).

Problem: Laptop computer is slow all the time

Solution: If you are running on battery power, you might find things speed up by connecting the PC to mains power.

Problem: CPU is too hot

Solution: Your CPU might be overheating and then throttling down to a slower speed. Check the CPU temperature and if it is too high, switch off your machine and clean out any dust in the case, ensuring that the fan outlets are not otherwise obstructed. If the temperature of your CPU is still too high, check that your computer fans are operational and running at the right speed. You may wish to consider purchasing extra case fans to help ventilate your machine.

Problem: Your PC is full of junk

Solution: Over time, people tend to install more and more software. There has been a trend by software developers over recent years to have their own applications set to automatically startup when the PC is turned on. The more stuff you have running, the less RAM you have and the longer it will take for your PC will take to start up. The solution is to uninstall any applications you don't need and disable unnecessary applications from running at startup by removing their shortcuts from the StartUp folder.

Problem: Your Operating System is running stuff that you don't need

Solution: Windows contains dozens of software services, which are pieces of software that run when the operating system starts. Most of these form crucial parts of the operating system, however, some services are not used by a majority of users and can be turned off for a small CPU/RAM usage saving. The detail of what services do what is beyond the scope of this article, but there are many articles on the internet which describe how to disable uncritical Windows services.

Problem: PC is slow, applications frequently lock up and the disk light is on frequently

Solution: Your machine may be low on RAM. This may be as a result of too many applications running at once, or because you don't have enough physical RAM. Try shutting down all applications and rebooting your machine. If you encounter this problem frequently, you may wish to consider upgrading and adding more RAM as it is relatively inexpensive.

Problem: PC is slow

Solution: If your PC is more than five years old, it will never run Windows 8, Windows 7, or Windows Vista very well. Consider a hardware upgrade.

Problem: Slow Disk Access and Errors from the Disk

Solution: Check the S.M.A.R.T. status of the drive to see if the drive itself thinks everything is OK. You can also use DiskCheckup to verify the health and reliability of the hard disk drive - DiskCheckup can give you a heads up in case it is time to replace your hard drive.

You may also have some corrupted files. Windows comes with a utility called Scandisk which scans for file corruption and bad sectors.

Problem: Slow disk access

Solution: Check you have the correct motherboard device drivers for your motherboard. If you are just using the default device drivers, you might not be running the hard drive in the faster possible mode (PIO mode vs Bus mastering Ultra DMA mode). Also check you are using the best cable (80 conductor cable vs 40 if using IDE drives).

Problem: RAM access is slow

Solution: Check the BIOS RAM timings. Select automatic unless you really know what you are doing. Check you have enough RAM to avoid swapping out to disk all the time. Check the RAM is in the correct slots and matched pairs if you have a dual channel motherboard.

Problem: 3D graphics results don't get above 60 frames / sec

Solution: If the frame rate for the 3D test always sits at approximately 60 frames / sec then your frame rate might be limited to your monitors refresh rate. There is an option in PerformanceTest to limit the frame rate to the refresh rate. (Edit / Preferences). But there are additional options in some device drivers to control this and further overrides in some versions of DirectX. As stated in the Wikipedia article "...For that reason it is not uncommon to limit the frame rate to the refresh rate of the monitor in a process called vertical synchronization."

Problem: CD / DVD Score is low in PerformanceTest

Solution: The media selected for the test has a big effect on the score. Media with a large number of small files will give lower scores than media with a small number of large files. We suggest you use our test CD/DVD for consistent results. DVD media is often better than CD media as well.

Problem: Windows 7 only reports 2 cores for a Core i7 9xx CPU

Solution: The Core i7 CPU should have 4 core and 8 threads (when hyperthreading is turned on in BIOS). We have seen some rare instances where the wrong number of cores is identified by Windows, 2 cores instead of 4. This might happen if you upgraded the CPU from a 2 core unit to a 4 core unit. The solution is to reset the Core count via the Microsoft msconfig utility, under the advanced tab.

Problem: Disk results are slightly lower than expected

Solution #1: One of the (many) reasons you might get a slight decrease in disk speed is the BIOS disk controller mode. Typically BIOS options are IDE, SATA, RAID & AHCI. IDE is the most compatible, but often slower than the other options. If your hardware and O/S support SATA & AHCI, then select these options.

AHCI stands for "Advance Host Controller Interface". It offers a collection of features not available on the old Parallel ATA controllers such as hot plugging and NCQ command queuing.

Warning: turning this on without the right drivers (especially on Windows XP can result in crashes and a BSOD). Also see this Microsoft knowledge base article.

AHCI performance gains vary from 0% to 20% depending on the test method, the hardware & the O/S.

Solution #2: Also check you have the latest BIOS firmware for your motherboard.

Solution #3: Try different SATA ports on your motherboard. Some motherboard come with 2 disk controller chips. Normally there is one controller included with the motherboard chip set, but there might be a second set of ports to support RAID or other advanced functions.

Example:

This was taken from this old post. A Samsung 1TB drive (HD103SJ) scored 620 in the PerformanceTest DiskMark. While the average DiskMark for this drive was 829. The rest of the configuration was, Intel i7 920 @ 3.36ghz / 6gb / Win7 64 / Gigabyte GA-EX58-UD3R MB.

Initial score: 620
Switching cable from the Gigabyte to Intel SATA port increased score to: 784
Switching to AHCI on Intel port increased score to: 819

An overall 32% performance increase. CPU usage was also lowered at the same time. So a double benefit.

Problem: 3D graphics results are lower than expected. Especially the "Medium 3D" test (in V7 of PerformanceTest).

Solution: If the settings in the video card's device driver are manually set to maximum quality and maximum anti-aliasing, then this can significantly decrease the frame rate compared the leaving the setting on the default values.

Problem: CPU Results are lower than expected compared to other baselines with the same CPU.

Solution: The 64bit release of PerformanceTest will return higher results than the 32bit release. The reasons for 64bit being faster are examined here. And there is a chart here. So for the highest CPU benchmark results use a 64bit operating system and the native 64bit release of PerformanceTest. Note that the 32bit release of PerformanceTest will run on 64bit operating systems, but give lower results, than the native 64bit release, which is a separate download. Starting with PerformanceTest V8, both releases are installed and the correct edition will be ran when testing.

Especially effected by this is the integer maths results on Intel CPUs. The integer maths test does a combination of 32bit and 64bit arithmetic. A 64bit CPU on a 64bit O/S running native 64bit code doing 64bit maths is a lot faster than a 32bit configuration doing 64bit maths. This was shown 4 years ago in a comparison we did between 32bit and 64bit performance.

But what seems to have changed is that recent Intel CPUs have improved 64bit performance more than 32bit performance, compared to older CPUs.

So as an example, in the years between the Pentium 4 and Core i7 chips, 32bit maths improved 7 fold in speed, while 64bit maths improved 12 fold.

So now the difference between 32bit and 64bit is very pronounced on the Integer maths test on newer Intel CPUs. The difference also exists for AMD chips, but it is not as stark.

Upgrading to 64bit software and operating system should give you much better results.

Problem: The individual CPU tests, "String Sorting" & "Physics" results in PerfornmanceTest are a bit low, but other CPU results aren't too bad. Memory test and 3D test results are also a bit low.

Solution: This can be caused by bad RAM configuration. The strings and physics test use more RAM than the other CPU tests and so the impact s more pronounced on these tests. Check the BIOS RAM timings. Select automatic unless you really know what you are doing. Check you have enough RAM to avoid swapping out to disk all the time. Check the RAM is in the correct slots and matched pairs if you have a dual (or tri) channel motherboard. Check the RAM is of the correct speed for your motherboard.

Problem: The 2D results in PerfornmanceTest are low in Windows 7 compared to XP.

Solution: Changes in Windows 7 have heavily impacted the kind of artificial benchmarks that PT does. The real world impact of these changes may be beneficial in some cases and detrimental in others, but the average windows user probably won't notice much difference. In some circumstances, XP also had better GDI acceleration than Win7. You can find a discussion of the 2D performance issues in Win7 here. Turning off some of the Windows effects and Aero can help improve the benchmark results. (Right click the desktop and click Personalization. Click Window Color.)

Problem: The incorrect number of CPUs or Cores are being reported by PerformanceTest.

Solution: If PerformanceTest doesn't detect the correct number of CPUs in the system or the correct number of cores, then it is likely that the CPU(s) will not be fully loaded during the CPU benchmark, and you'll get poor CPU performance results. If you are running PerformanceTest then check then check what is reported on the "System" and verify the CPU and core count is correct for your CPU. The only case we are aware of it not being correct (as of Mar 2011) is when there is a problem in the BIOS configuration.

In some BIOS setup screens you can find a value called something like, "Max CPUID Value Limit" or "Maximum CPUID Input Value BIOS" or "Limit Maximum CPUID to 3", etc...

This CPUID limit setting needs to be disabled

if you are running XP or later.

The technical reason for this is as follows:

CPUID is a low level machine code command that can be executed on the CPU to gather information about the CPU. This information includes details like the make and model of the CPU, the features it supports, cache configuration and the number of cores available. New CPUs make much more information available about themselves than old CPUs.

Full details for Intel CPUs can be found here, http://www.intel.com/assets/pdf/appnote/241618.pdf

When Windows boots it queries the CPU to see what level of information is supported (what numbers can be set in the CPU's EAX register). Old operating systems like Windows 98 didn't support a value greater than 3. So to support new CPUs with these old operating system the BIOS developers included this option.

Some programs like PerformanceTest also use the same CPUID instruction to detect the available CPUs and Cores in the machine. So having this value set wrong can result in wrong system information being reported for newer CPUs and thus lead to performance problems and other strange behavior.

As an example of what might be (incorrectly) reported in PerformanceTest, if this BIOS setting in incorrectly enabled. A Intel Core i7 950 might be reported as having 1 core instead of 4 (or 8 with hyper threading).

Problem: CPU is slow all the time.

Solution: You might be running your CPU in low power mode (which will reduce the clock speed). Check your BIOS settings for this and other incorrect settings (like accidental under-clocking). Also check the power settings in the Windows control panel.

If you are running the ASUS Cool n Quiet feature, consider turning it off in BIOS. According to ASUS this feature can automatically tune CPU voltage and frequency. However running cool and quiet does not equal maximum performance. Maximum performance is often hot and noisy.

Problem: Laptop computer is slow all the time.

Solution: If you are running on battery power you might find things speed up by connecting the PC to mains power. If your PC is in Power Saver Mode you might get at 30 to 40% improvement in the CPU benchmark after changing to High Performance mode. See the "Power options" in the Windows control panel to change this setting.

Problem: CPU results are low. But they were OK when Windows was first installed and the CPU has more than 1 core.

Solution: After checking power settings and for overheating (see above). Then also check the Windows boot settings. The machine might have been 'tweaked' to run on a reduced number of cores.

Run the configuration utility built into Windows called Msconfig.exe, from the Windows start menu. In the Boot tab select Advanced Options. Then uncheck the "Number of processors" check box. This will allow Windows to autodetect the correct number of CPUs/Cores in the machine.

Problem: Your high end SSD drive doesn't get past around 280MB/sec, but according to the specs should be able to run much faster.

Solution: Make sure you have connected the SSD drive to a SATA 3 port (6Gbits/sec) and not the slower SATA 2 port (which is only 3 GBit/sec). Note that older motherboards won't support SATA 3. After the overheads are taken into account, 3 GBit/sec corresponds to a transfer speed of around 280MBytes/sec

Also check you are in AHCI mode as mentioned above.

Problem: 3D test results are low. Especially for the 'complex' test. Frame rates in games might also be slow.

Solution: Make sure your video card is plugged into the correct PCI-E slot on your motherboard. Some MB have various PCI-E slots that run at different speeds (lane counts, x1, x2, x8, x16).

For PCI-E the lane count is automatically negotiated during device initialization, and can be restricted by either endpoint. For example, a single-lane PCIe (×1) card can be inserted into a multi-lane slot (×4, ×8, etc.), and the initialization cycle auto-negotiates the highest mutually supported lane count. The link can dynamically down-configure the link to use fewer lanes, thus providing some measure of failure tolerance in the presence of bad or unreliable lanes

Bad lanes or dirty connectors on your video card or motherboard can result in some of these lanes being disabled.

In the CPU-Z software there is a "link width" display. So check that out and see what you are now running at. For example it should be x16 if that is what your video card and MB support.

For most applications the difference in bandwidth between x8 and x16 isn't important, as x8 is enough most of the time. Being stuck at x1 or x2 however can be very noticeable.

Possible solutions include BIOS updates and re-seating the video card.

Problem: Integer maths test and Prime number test give very low results. About 80% lower than expected. All other tests are give the expected benchmark results and you have an AMD Fusion Llano CPU.

Solution: We believe this is related to a CPU bug.

At the time of writing (March 2012) this seems to effect about half of the CPUs when run in our CPU benchmark.

The bug itself can cause the CPU to hang or behave unpredictably while doing division operations. The workaround AMD suggested is to patch the CPU, which reduces the CPU performance to avoid the hang. The patch should be applied in BIOS. But only half the motherboard manufacturers have applied it from what we can see (as of April 2012).

AMD Fusion Llano CPUs include the A4-3300, A6-3500, A6-3600, A6-3650, A8-3800, A8-3850, Athlon II X4 641, AMD Athlon II X4 651 and a few others.

This problem will be reduced (slightly more hidden) in PerformanceTest version 8 where we will do less division.

For more details, see this post, http://www.passmark.com/forum/showthread.php?t=3656

Problem: The DirectX 9 complex test gives low 3D frame rate results (~30% down) compared to similar machines.

Solution: The Morphological filtering (also know as MLAA) setting might be turned on the video card's device driver control panel. This issue was seen with HD 6850 video card, but probably effects most cards.

Morphological filtering is a technique that applies full screen anti-aliasing, which can improve the smoothness and quality of rendered images.

Problem: 2D results are low compared to other some other baselines. (Or put another way, some machine's have 2D results that are too high).

Solution: We have seen a small number of baselines with very high 2D results in V7 of PerformanceTest. Roughly 10x faster than normal. e.g. Solid vector scores of 22.0 instead of 2.1

It turns out that running PerformanceTest across some remote desktop solutions gives higher than expected 2D performance. This isn't a real performance increase, the video driver is instead throwing away some requests to update the screen, making it seem to the application that updates are running faster than they actually are.

Problem: CPU and video card performance are low. In particular the 2D video performance is very low.

Solution: You are running some 3rd party software save power and lower CO2 emissions. The one we know about that causes a problem is the ASUS 'EPU-4 Engine' software. But there might be others as well. The EPU-4 software seems to under clock the CPU, making your PC run slower so that it takes longer to complete any task you give it. Considering all new CPUs automatically throttle themselves depending on load, it isn't clear to us why you would want this software installed. The ASUS EPU software is also part of what is grandiosely named 'ASUS AI Suite 3'

Problem: CPU, 2D & RAM results are low, as is the CPU's clock speed (in GHz).

Solution: Your motherboard's BIOS might not have correctly detected the CPU you are using. This can happen if the CPU is not on the list of supported CPUs for the motherboard. Which is often a result of a new model CPU being used on an older model motherboard.

An example of this is the Gigabyte GA-78LMT-S2P motherboard not correctly detecting the clock speeds for the newer AMD FX-6200 CPU in the latest F3 BIOS version. PerformanceTest reports the clock speed as 2.81 GHz [Turbo: 2.81 GHz] when it should be 3.8Ghz [4.1Ghz turbo]. So you are effectively loosing about 30% of the machines performance because the BIOS isn't up to date (thanks Gigabyte).

The solution, if newer BIOS isn't available for your motherboard, it to manually correct the clocks in speeds in BIOS, in the same manner as if you are overclocking the CPU.

Things Which Generally Don't Improve System Performance At All

Here is a list of stuff that probably won't help your PC run faster (despite what you might read elsewhere from people with vested interests):

• Fixing registry problems: Removing a few things from the registry rarely has any impact on PC speed. The whole registry 'cleaning' industry greatly exaggerates the benefits of doing this.
• Using 'memory optimizers': At best, these memory optimizers do close to nothing. At worst - they degrade system performance and are a straight-out software scam, preying on ignorant consumers.

Why do my results from PerformanceTest V9 differ from my results from previous releases?

There were several changes to the 3D tests and some minor changes to the CPU tests. Because of this, results created with version 8 will be slightly different to version 9.

The weighting of the 3D tests was changed due to the removal of the DX9 simple test and the addition of the DX12 test. The DX9 (complex) test is now weighted lower than the other tests (approx. 25% less). Systems that cannot run the DX12 test will receive a 20% penalty to the overall 3D mark score.

There is a 6.5% scale down of the CPU Mark from PT9 so the results are closer to PT8 results, to compensate for the increased scores in the Extended Instructions and Physics test changes. So the PT8 CPUMark should be broadly comparable to the PT9 CPUMark. Same for the overall PassMark rating. Still, when possible, for improved accuracy, compare PT9 results to PT9 results and PT8 results to PT8 results.

Baseline filess from version 7 will have their values scaled up and down when loaded into PerformanceTest version 9 to approximate the results that may have been obtained with version 8.

What kind of data does the disk test read/write.

The standard read test creates a file of 512MB, all other standard disk test create files 200MB in size. The file is filled with data that has a randomness of about 20% before the test begins. The randomness is seeded and is the same every time.

• All reads and writes are done in 32kb blocks
• Any data written once the tests commence is also roughly 20% random and seeded in the same manner.
• If the drive has NTFS compression turned on, PerformanceTest will still try to use an uncompressed file for the standard tests. If it can't create a file in a uncomprssed state, the test will fail.

The results for my hard disk aren't what I expected.

Factors Which May Affect Disk Test Results

There are many factors that affect the disk test results in a Windows environment. Here are the main ones:

• Cluster size: Larger clusters in general means better performance for large files.
  See Appendix 1 for typical default values. Reformatting your drive with larger clusters will give better results in our benchmark, but will also waste disk space. See Appendix 2 for an easy way to check your current cluster size.
• Fragmentation: If the disk is fragmented and almost full, this can badly affect performance. Windows operating systems include a utility for defragmenting the disk, search for "defrag" in the Windows online help for more details.
• The position of the test file on the disk (inner cylinder or outer cylinder) can also affect the performance. The only way to avoid this problem is to only test newly formatted disks with no partitions. In traditional drives, the fastest (outer) part of the drive, can be up to twice as fast as inner cylinder. In general hard drives will allocate partitions from the outside in.
• The type of file system being used: FAT, FAT32 or NTFS.
• The type of operating system: Window9x, NT, 2000, XP, Vista, Server2003, Server2008.
• The disk controller (PATA, SATA or SCSI) and what mode it is running in: ATA-33,66,100, RAID0, 1, SATA 3Gbit/s or 6Gbit/s, etc.
• If the drive is connected directly to the motherboard, or via USB, Firewire, etc..
• The amount of memory that Windows has currently allocated to the disk cache. This can and will vary from one run to the next.
• The amount of onboard cache the drive has.
• Anti-virus software scanning files read / written to the disk.
• If the disk has compression turned on, you may see very slow writes but much faster cached reads.
• For SSD drives, the amount of use a drive has recieved and if the drive and O/S you are using support the TRIM command.
• The firmware version in the drive itself. This can have a significant impact on newer SSD drives.
• The block size used in the benchmark test (this can be adjusted in the Advanced Disk test window in PerformanceTest). Many newer SSD drives only seem to perform at their peak when a very large block size is used.

PerformanceTest attempts to measure the non-cached disk speed. In some circumstances, e.g. when the compression driver is in use, the operating system ignores the request not to cache data. This can lead to strange looking results. Typical disk speeds are in the range of 50MB/sec to 100MB/sec and numbers of above 300MB/sec probably indicate that the cache is interfering with the result.

My Seek Read/Write Results Still Seem Too Low

This test creates a large test file on the disk under test. The file is then read randomly; a seek is performed to move the file pointer to a random position in the file, a 16KB block is read or written then another seek is performed. The amount of data actually transferred is highly dependent on the disk seek time.

Appendix 1 - Default Hard Disk Cluster Sizes

Default Cluster Sizes for Volumes with Windows 2000 and XP File Systems

Appendix 2 - Checking your cluster size

Versions of PerformanceTest later than 3.2 display the cluster size of the disk. In earlier versions of the software you'll need to check it yourself. Find or make a file on your hard disk with just a few bytes in it. Then while in Explorer right click on the file and select properties. Compare the values for 'Size' and 'Size on disk'. You can see that in the smallfile.txt example (left) that the disk is used extremely inefficiently (0.015% efficiency in fact). This is because this disk has been formatted with 32KB clusters. If the disk was formatted with 4KB clusters (top right) the results would have been very different (0.12% efficiency, a 8 fold improvement).

2D Test Issues

2D results in Windows 7 seem poor compared to previous versions of Windows, why is this?

2D results in Windows 7 seem poor compared to previous versions of Windows, why is this?

Most of the 2D benchmark test results in PerformanceTest haven't changed much in Windows 7 compared to Vista with the exception of the solid and transparent vector tests, which have shown a significant reduction in speed compared to Vista. The reason for this seems to be an intentional decision made by Microsoft that has significantly impacted the direct window rendering techniques used in these tests (known as GDI+).

According to information from Microsoft they have restructured the internals of the Windows graphical subsystem to make more efficient use of RAM on the videocard and less use of main memory. The consequence of this seems to be that when a large number of windows are in use much less system RAM is required (a good thing), but there is a side effect of negative performance  impact when data needs to be read back from video memory. This negative impact can be seen in the results of the 2D vector tests in PerformanceTest.

Further, in Win7 the locking of operating system resources in done at a finer level. So more locks are used, but they are held for a shorter period. Microsoft describe this as, "GDI concurrency and lock refactoring to enable more efficient multi-threaded access to GDI". For an individual single threaded application this can also result in a slight performance decrease. But when multiple applications or threads are all updating their windows at the same time, the user interface will appear to be more responsive (as one application won't block screen updates being done in another application nearly as often).

Different Windows applications use different methods of drawing to the screen. Some applications render directly to the window, which is the same method used in the vector tests in PerformanceTest (GDI+ to a hWnd). These application will likely suffer a performance hit in Win7. Other applications render to a buffer in main memory and then use a method known as bitblt to push the final image to the screen buffer (GDI+ to a DIB section then bitblt). This method is known as double buffering. Most large commercial applications (such as Word and Photoshop) use this later method and will not suffer much if any performance degradation in Win7. There is a discussion of this on Microsoft's graphics performance blog.

It has been reported elsewhere that the WDDM V1.1 video card drivers in Windows 7 have additional hardware acceleration for GDI compared to WDDM 1.0 used in Vista, and so should be faster. But out of the 100s of GDI functions only 6 functions are accelerated in hardware in Win7. These are, BitBlt, ColorFill, StretchBlt, AlphaBlend, Cleartype fonts & TransparentBlt. The other functions are executed on the CPU. It should be noted that Windows XP had better 2D hardware acceleration (but less eye candy) than Vista. Arguably in some circumstances, XP also had better GDI acceleration than Win7.

The 2D benchmark tests in PerformanceTest are single threaded and don't use huge amounts of RAM. So the potential improvements in Win7 are not reflected in the benchmark results. But the negative side effects can be seen. In the real world the Win7 interface as a whole should feel more responsive (especially in low RAM situations) even if some individual applications might run slower.

In PerformanceTest 8 a Direct2D test was added. Direct2D is a new 2d graphics API for Windows 7 (and Vista with updates). The addition of this test should show an area where Windows 7 performs well and narrow the apparent gap in test results between the different version of Windows.

Other factors should also be considered when comparing 2D performance between different versions of Windows. These are,

1. Different video card device drivers versions can impact performance.
2. CPU performance also has a big impact on 2D rendering.
3. Different versions of Windows have different visual effects. XP can look rather plain compared to Vista, but the plain look can be faster to render on the screen.

2D results are extremely high when running via remote desktop?

We have found that when PerformanceTest is run on a remote machine via remote desktop and then the remote desktop window is closed or minimized while running the 2D tests, the 2D test scores can become extremely high.

The problem is that closing or minimizing the remote desktop window effectively removes the display from the remote machine, leaving the 2D tests with nowhere to render to. This leaves the 2D tests doing nothing, and doing it very fast, resulting in an extremely high score at the completion of the test.

Newer versions of PT have some checking to try and detect this issue however it isn't perfect and this issue can still arise in certain cases.

3D Test Issues

Does PerformanceTest support NVIDIA Optimus?

If you are using a laptop with NVIDIA Optimus enabled, for example those with GeForce GT 420M, 425M, 435M, 445M, 520M, 525M, 540M, 550M and 555M video cards, optimus may not correctly switch to the high performance video card when running PerformanceTest. Consequently the 3D score might not be as high as it should be as it may have been using the power saving and slower on board graphics.

You can change the NVIDIA settings to select the correct video card when performanceTest is started using these steps;

• Open the NVIDIA control panel
• Under "Select a task" open "3D settings" and select the "Manage 3D settings" option
• On the right hand side click the "Program settings tab"
• Under "Select a program" click "Add" and then browse to the PerformanceTest executable. The default location is C:\Program Files\PerformanceTest\. These steps will need to be repeated for the following files in this folder;
  • PerformanceTest32.exe or PerformanceTest64.exe depending on your Windows version, if you are unsure, adding both will cause no harm.
  • PT-D3D11Test.exe
  • PT-D3D12Test.exe
  • BitonicSort.exe
  • Fluid3D.exe
  • Mandel.exe
  • QJulia4D.exe
  • oclParticles.exe
• Under the "Select the preferred graphics" section choose "High-performance NVIDIA processor" and click Apply

I receive this message when trying to run the 3D tests "Nothing will be rendered."

The full error message is as follows

"Warning: Nothing will be rendered. The reference rendering device was selected, but your computer only has a reduced-functionality reference device installed. Please ensure that the most recent device driver for your graphics adapter is correctly installed. "

This message basically means that PerformanceTest could not find a video card that supports the features it requires. Possible reasons for this include.

1. Lack of properly installed video card drivers. Update your drivers to solve this issue.
2. Lack of 3D DirectX 9 capable video card. This is very rare these days but may be an issue on very old systems or some server set-ups.
3. Lack of supported display resolution. PerformanceTest 3D tests need a minimum supported resolution of 1024x768.
4. Other applications interfering with the display. One such instance we have encountered was "Pivot software from Portrait Displays" being used to switch the monitor into portrait view. The problem may have been that in portrait mode the horizontal display resolution was below 1024 pixels.
5. If you are using a Nvidia Quadro card it has been reported that they may not work properly when using Windows Remote Desktop Connection (or another viewer using the RDP protocol). Using another remote viewer and protocol, like VNC, may solve this problem.

Why do my 3D graphics results vary greatly between runs?

On a small number of graphics cards we have noted that the 3D performance can be greatly impacted by whether the mouse cursor is on the 3D Graphics PerformanceTest Window. We believe that this is a graphics card device driver issue for windowed DirectX 3D. For now, please just ensure that the focus (mouse pointer) is set on the 3D test window.

Unable to run the 3D test on an Nvidia Quadro.

We often have reports that some (or all) of the 3D tests will often not run or crash on systems with an Nvidia Quadro installed. Commonly this will be due to a driver issue and updating the driver will allow the test to run. As Nvidia release several version of the driver (performance, partner certified, auto cad, 3DS etc) it may also help to try different versions, eg using the performance driver instead of an AutoCAD optimised one.

It has also been reported that they may not work properly when using Windows Remote Desktop Connection (or another viewer using the RDP protocol). Using another remote viewer and protocol, like VNC, may solve this problem.

Running the advanced DX11 test on an Intel graphics card and get "Test aborted due to attempted resolution switch" error.

We have seen this behaviour on Intel HD 4400 cards when trying to run the test at 3200 x 1800 with anti aliasing turned on. With anti aliasing off the test will execute without an issue.

The cause appears to be the settings in the "Intel HD Graphics Control Panel" that override the AA settings chosen by PerformanceTest. Choosing the "Quality" setting or "custom settings" and "Use Application settings" should see the test run with AA settings of 2x and 4x.

Crash during 3D test on laptop when using docking station

We have recently begun to see several crashes during the 3D tests on laptops (mostly with Intel integrated graphics) when they are connected to a docking station, with external monitors connected to the dock.

In these cases removing the laptop from the docking station allows the tests to complete without issue.

I'm seeing white banding in 'DirectX 9 - Complex' test at very low resolutions

In some cases when the 'DirectX9 - Complex' test is run at very low resolutions (512x384 and lower) a flickering white band may be observed at the bottom of the test.

So far we have only seen it happen at very low resolutions and it does not appear if any level of anti-aliasing or V-Sync is selected. It appears to be a rendering issue of the 'DirectX9 - Complex' test and not related to a driver issue.

Update: Another cause seems to be running the EVGA PRECISION XOC software in the background. This tool is capable of displaying an overlay window, over the top of the normal 3D scene. This seems to cause glitches in various software.

When running the DX11 test I get a "Test aborted due to attempted resolution switch" error message.

The error message will look like this;

This error means that PerformanceTest was interrupted after the test has started by a resolution change notification. We have seen this occur in two circumstances;

1. Google Hangouts is installed and the mini window is displayed on the current primary monitor (the same monitor the DX11 test will be executed on). This error message can be resolved by closing Google Hangouts or moving it on to another monitor
2. The intel device driver notification when the monitor resolution is changed to a non-native (optimal) resolution. The Intel device driver displays a warning message in a balloon in the system tray which causes the resolution to switch back and forth. To resolve this error disable the balloon notifications in the intel device driver;
   • Right click the Intel system tray icon
   • Choose Graphics options -> Balloon notifications -> Optimal resolution notification
   • Select the disable option

When running the DX11 test I get a "DirectX 11 measured result is too high to be possible and will be discarded." error message.

This error message will occur if the frame rate for the DX11 test is detected as being too high to be valid. This has been reported to occur in some instance when running video capture software (eg Nvidia shadowplay, AMD gaming evolved, fraps etc). If you see this error message please try disabling the video capture software before trying to run the test again.

The Directx 12 test doesn't run & I get a white screen or crash instead.

Blank screens, crashes and system freezes

Only Windows 10 and above supports DirectX 12. So this test will never work on machines with older versions of Windows.

Older video cards don't support DirectX12. So check the specs of your video card if the test doesn't run at all.

If you attempt to launch the DX12 test and you get a white screen and or freezing of the test, then this can be caused by either

1. The EVGA PRECISION XOC 3rd party software application. This tool is capable of displaying an overlay window (OSD), over the top of the normal 3D scene. This seems to cause glitches in various software and can cause the DirectX12 test to crash. Uninstalling the EVGA PRECISION XOC software can fix the problem.
2. The ASUS AISuite3 3rd party software application. This tool is capable of displaying an on screen display (OSD), called the OSD Mini Bar, over the top of the normal 3D scene. This can cause the machine to lockup / freeze. Disabling the mini bar might be sufficient to fix the problem, otherwise uninstalling the ASUS software can fix the problem.

Additional information

If debug logging is turned on, then you might see the following errors in the log.

ERROR: HRESULT (0x887a0005)
DEBUG: Render(...) Failed Code: -2005270523

or this

ERROR: HRESULT (0x887a0005) Device removed reason (0x887a002b)

DirectX error code: 0x887a0005 means DXGI_ERROR_DEVICE_REMOVED, which Microsoft describe as, "The video card has been physically removed from the system, or a driver upgrade for the video card has occurred".

It doesn't really make sense that EVGA's Precision Xoc software can cause this device removal event, but it does. Also it seems to not effect 32bit systems. We only saw the problem with 64bit 3D code.

EVGA PRECISION XOC can cause a crash when running the Dirext12 test and can produce a crash dump similair to this;

BUGCHECK_STR: APPLICATION_FAULT_SOFTWARE_NX_FAULT_NULL_INVALID_POINTER_EXECUTE
STACK_TEXT:
0x0
dxgi!CDXGIFactory::GetListOfNotPresentingFullscreenSwapChains
dxgi!FullscreenWatchdogThreadWorker


OpenCL DLL not found error message and OpenCL test will not run.

This error message will occur if the OpenCL.dll file is not found on the system. Updating the video card drivers to the latest available should resolve this error message in most cases.

Network Issues

Does the advanced network test support 10-gigabit network cards?

Does the advanced network test support 10-gigabit network cards?

Yes it does support testing of 10Gbit/s NICs. But there are often bottle necks that prevent you getting close to the maximum speed. See FAQ below for details.

One solution to avoid bottle necks is by running multiple instances of the test at the same time. This can be done by launching two copies of PerformanceTest and running the Advanced Netowrk Test on two different IP ports. For example ports 1040 and 1041.

What are expected speeds for a 1Gb network card

On a 1Gb Ethernet link you should get around 900 to 950Mbit/sec, in ideal conditions. There are lots of good reasons why you might not see this level of performance however.

Example LAN bandwidth graph

This was on a 1Gb Ethernet link on a local area network.

What are possible performance bottle necks for network speeds

Here are some of the reasons you might be seeing the bandwidth you expect on a local area network (LAN).

• Existing background load on the network or the machines being used for testing
• 100Mb/sec was negotiated by the card / switch instead of 1Gb/s. Maybe due to errors on the link or excessive cable length
• CPU might not be quick enough. Note that single core performance is important for the network test. Multiple cores don't help when you have only a single instance of the test running.
• PCIe bus connecting the network card to the CPU might be too slow
• Cabling problems. e.g. using CAT5 instead of CAT6
• Wrong device driver is loaded up for the network card (NIC)
• Poor TCP/IP settings. For example jumbo Ethernet frames (MTU), PCI burst transfer sizes.
• If the NIC supports processing offload. (called TCP offload engine or TOE)
• Routers / switches / Firewalls are not setup correctly or were never designed for the required speed.
• The connection is high latency due to the machines being far apart, or routing problems causing the wrong route to be selected.
• Low level physical errors on the link, data corruption, electrical issues, re-transmisions, EMI.
• If you are using WiFi, then radio interference, congestion, low signal levels & reflections, can all cause problems.

Tools, like WireShark & NetStat, can help in getting to the bottom of networking problems & to monitor link errors.

Why is the network test so slow on my (old) laptop?

On older laptops network speeds are usually limited by the PCMCIA bus speed. You may have a 100 MBps network card, but the PCMCIA bus will not allow transmission rates much above 10 MBps.

How Can I Get the Network Test to Run Through a Firewall?

You may have to set up a rule allowing access on the port used by the network test. If you don't have direct control over the firewall, you'll have to contact your network administrator. The port number used by the test can be found in the network test dialog. In WIndows 10, you should be prompted automatically to allow PerformanceTest through the firewall at the start of the test. 3rd party firewalls might need to be manually configured.

Why is the Client Machine Giving Me the Error Message "Connect Failed"?

The most likely causes are:

• You forgot to click on the "Go" button on the server
• In the PerformanceTest Advanced network test window (on the client/sending PC), the IP address entered is incorrect or the port numbers on the client and server PC's do not match. The same port number must be used on both PC's. The IP address of the server PC, must be entered on the client PC.
• A firewall (eg. Microsoft XP firewall or Norton Internet Security Firewall) is turned on for one or both of the PC's LAN connections and this is blocking a connection. Solution: modify the firewall rules to allow this address, port and protocol through the firewall or switch the firewall off (if appropriate).
• The network cards are not working. You should check in device manager that your Network cards are shown as "This device is working properly".

Note: To test connectivity between the Client PC and the Server PC try the following: Open a "Command" window on the Client side PC by selecting "Start", "Run", type in "Command". Once you have the "Command" window open, type "ping <IP-Destination>", substituting <IP-Destination> for the IP address of the Server side PC.

Why Does The Client Machine Stop Receiving Data?

When the Client computer graph shows that nearly the entire bandwidth is utilized for the full test duration of the send session and the receiver shows that after a number of seconds into the test a drop from nearly 100% utilization to 0, the most likely cause is that a router between the PC's is dropping UDP packets.

UDP is not rate adaptive (basically packets are pushed out as quickly the PC can push them out with no consideration whether the packet was received), unlike TCP. A router between the PC's could become congested and hence start dropping the UDP packets, and remain congested until the end of the test (This is of course more likely if the router is a lower end router or there are other users on the network). Depending on your router, you may be able to get UDP statistics on the 'in' and the 'out' ports to confirm this. You could also try connecting the 2 PC's directly (depending on what you are trying to test).

What are expected network speeds and common bottlenecks.

Does the advanced network test support 10-gigabit network cards?

Yes it does support testing of 10Gbit/s NICs. But there are often bottle necks that prevent you getting close to the maximum speed. See FAQ below for details.

One solution to avoid bottle necks is by running multiple instances of the test at the same time. This can be done by launching two copies of PerformanceTest and running the Advanced Netowrk Test on two different IP ports. For example ports 1040 and 1041.

What are expected speeds for a 1Gb network card

On a 1Gb Ethernet link you should get around 900 to 950Mbit/sec, in ideal conditions. There are lots of good reasons why you might not see this level of performance however.

Example LAN bandwidth graph

This was on a 1Gb Ethernet link on a local area network.

What are possible performance bottle necks for network speeds

Here are some of the reasons you might be seeing the bandwidth you expect on a local area network (LAN).

• Existing background load on the network or the machines being used for testing
• 100Mb/sec was negotiated by the card / switch instead of 1Gb/s. Maybe due to errors on the link or excessive cable length
• CPU might not be quick enough. Note that single core performance is important for the network test. Multiple cores don't help when you have only a single instance of the test running.
• PCIe bus connecting the network card to the CPU might be too slow
• Cabling problems. e.g. using CAT5 instead of CAT6
• Wrong device driver is loaded up for the network card (NIC)
• Poor TCP/IP settings. For example jumbo Ethernet frames (MTU), PCI burst transfer sizes.
• If the NIC supports processing offload. (called TCP offload engine or TOE)
• Routers / switches / Firewalls are not setup correctly or were never designed for the required speed.
• The connection is high latency due to the machines being far apart, or routing problems causing the wrong route to be selected.
• Low level physical errors on the link, data corruption, electrical issues, re-transmisions, EMI.
• If you are using WiFi, then radio interference, congestion, low signal levels & reflections, can all cause problems.

Tools, like WireShark & NetStat, can help in getting to the bottom of networking problems & to monitor link errors.

Why is the network test so slow on my (old) laptop?

On older laptops network speeds are usually limited by the PCMCIA bus speed. You may have a 100 MBps network card, but the PCMCIA bus will not allow transmission rates much above 10 MBps.

How Can I Get the Network Test to Run Through a Firewall?

You may have to set up a rule allowing access on the port used by the network test. If you don't have direct control over the firewall, you'll have to contact your network administrator. The port number used by the test can be found in the network test dialog. In WIndows 10, you should be prompted automatically to allow PerformanceTest through the firewall at the start of the test. 3rd party firewalls might need to be manually configured.

Why is the Client Machine Giving Me the Error Message "Connect Failed"?

The most likely causes are:

• You forgot to click on the "Go" button on the server
• In the PerformanceTest Advanced network test window (on the client/sending PC), the IP address entered is incorrect or the port numbers on the client and server PC's do not match. The same port number must be used on both PC's. The IP address of the server PC, must be entered on the client PC.
• A firewall (eg. Microsoft XP firewall or Norton Internet Security Firewall) is turned on for one or both of the PC's LAN connections and this is blocking a connection. Solution: modify the firewall rules to allow this address, port and protocol through the firewall or switch the firewall off (if appropriate).
• The network cards are not working. You should check in device manager that your Network cards are shown as "This device is working properly".

Note: To test connectivity between the Client PC and the Server PC try the following: Open a "Command" window on the Client side PC by selecting "Start", "Run", type in "Command". Once you have the "Command" window open, type "ping <IP-Destination>", substituting <IP-Destination> for the IP address of the Server side PC.

Why Does The Client Machine Stop Receiving Data?

When the Client computer graph shows that nearly the entire bandwidth is utilized for the full test duration of the send session and the receiver shows that after a number of seconds into the test a drop from nearly 100% utilization to 0, the most likely cause is that a router between the PC's is dropping UDP packets.

UDP is not rate adaptive (basically packets are pushed out as quickly the PC can push them out with no consideration whether the packet was received), unlike TCP. A router between the PC's could become congested and hence start dropping the UDP packets, and remain congested until the end of the test (This is of course more likely if the router is a lower end router or there are other users on the network). Depending on your router, you may be able to get UDP statistics on the 'in' and the 'out' ports to confirm this. You could also try connecting the 2 PC's directly (depending on what you are trying to test).

Why is the network test so slow on my laptop?

On older laptops network speeds are usually limited by the PCMCIA bus speed. You may have a 100 MBps network card, but the PCMCIA bus will not allow transmission rates much above 10 MBps.

How can I get the network test to run through a firewall?

You may have to set up a rule allowing access on the port used by the network test. If you don't have direct control over the firewall, you'll have to contact your network administrator. The port number used by the test can be found in the network test dialog. In WIndows 10, you should be prompted automatically to allow PerformanceTest through the firewall at the start of the test. 3rd party firewalls might need to be manually configured.

Why is the client machine is giving me the error message "Connect Failed"?

• You forgot to click on the "Go" button on the server
• In the PerformanceTest Advanced network test window (on the client/sending PC), the IP address entered is incorrect or the port numbers on the client and server PC's do not match. The same port number must be used on both PC's. The IP address of the server PC, must be entered on the client PC.
• A firewall (eg. Microsoft XP firewall or Norton Internet Security Firewall) is turned on for one or both of the PC's LAN connections and this is blocking a connection. Solution: modify the firewall rules to allow this address, port and protocol through the firewall or switch the firewall off (if appropriate).
• The network cards are not working. You should check in device manager that your Network cards are shown as "This device is working properly".

Note: To test connectivity between the Client PC and the Server PC try the following: Open a "Command" window on the Client side PC by selecting "Start", "Run", type in "Command". Once you have the "Command" window open, type "ping <IP-Destination>", substituting <IP-Destination> for the IP address of the Server side PC.

Why is the client machine not receiving data?

When the Client computer graph shows that nearly the entire bandwidth is utilized for the full test duration of the send session and the receiver shows that after a number of seconds into the test a drop from nearly 100% utilization to 0, the most likely cause is that a router between the PC's is dropping UDP packets.

UDP is not rate adaptive (basically packets are pushed out as quickly the PC can push them out with no consideration whether the packet was received), unlike TCP. A router between the PC's could become congested and hence start dropping the UDP packets, and remain congested until the end of the test (This is of course more likely if the router is a lower end router or there are other users on the network). Depending on your router, you may be able to get UDP statistics on the 'in' and the 'out' ports to confirm this. You could also try connecting the 2 PC's directly (depending on what you are trying to test).

Why can't I select a network drive for disk testing?

It is possible to test mapped network drives using either the standard disk test (by selecting the drive from preferences) or from the advanced disk test.

However, on Windows Vista and later there is an issue relating to the User Access Control (UAC) and mapped network drives. Mapped network drives are not visible when running  as elevated administrator, which PerformanceTest is.

There is a workaround available that makes network drives available with elevated admin, and by extension available for testing in PT.

1. Click Start, type regedit in the Start Search box, and then press ENTER.
2. Locate and then right-click the following registry subkey:
   HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows\CurrentVersion\Policies\System
3. Point to New, and then click DWORD Value.
4. Type EnableLinkedConnections, and then press ENTER.
5. Right-click EnableLinkedConnections, and then click Modify.
6. In the Value data box, type 1, and then click OK.
7. Exit Registry Editor, and then restart the computer.

A more complete description of this issue and the workaround as explained by Microsoft can be found here http://support.microsoft.com/kb/937624

Older Issues

These issues are no longer relevant to modern systems or up to date versions of PerformanceTest.

I get a crash with this message: Access violation at 0x7c91084b (tried to write to 0x00040FFC)

This is caused by a bug in the Windows operating system. The crash address might also be slightly different, for example, 0x7C9108B. In particular, dual core CPUs such as the Athlon 4400 X2 and Intel Duo chips provoke the problem. Microsoft's description of the problem can be found in their knowledge base article, but in our opinion, it's a much more serious bug than the article describes.

Quoting from the Microsoft web site:
"Without these updates, computers that are equipped with these power management-capable, mobile, dual-core processors may experience decreased performance or unexpected behavior"

They say the solution is to download a hot fix, but at the moment (April 2006) Microsoft don't make it available for download. You need to E-mail or phone them first, which is far from helpful. Further discussion and references to alternative download links can be found in our forums.

Since PerformanceTest release v6 (build: 1012), we have worked around the problem and a host fix from Microsoft should no longer be required.

I have a ATI Radeon video card (with Catalyst video drivers). Why is PerformanceTest locking up during the 2D video tests?

PerformanceTest will lock up during the 2D video tests where you have an ATI Radeon video card (with Catalyst drivers) and have Windows Themes active for Windows XP. These conditions expose a bug in the ATI video drivers. This problem was seen in driver version 7.74 (6118), but was corrected in later versions. A workaround is to upgrade your video card drivers, or to disable Windows Themes.

I have dual monitors. Why are my 2D Performance Tests Results slower than expected?

Using extended desktops with dual monitors affect the 2D graphics test performance significantly; try disabling extended desktops and running the 2D graphics tests again.

Update: We have found that this problem seems to have mostly disappeared with the release of newer video cards.

Some 3D tests for PerformanceTest V7 use DirectX 9. So what happens in Windows Vista which is shipped with DirectX10?

In addition to the standard Windows Vista DirectX 10 installation, PerformanceTest requires a DirectX9 component.

We distribute PerformanceTest with the DirectX9 component that is required. PerformanceTest uses DirectX9 in order to be compatible with Windows 2000, and Windows XP and older video cards. Microsoft does not include all the DirectX9 interfaces with DirectX10, but they do provide a component for developers like us to install.

Normally this component will be automatically installed along side PT, however if for some reason it isn't then when you run PerformanceTest you just need to select "Yes" when the following message is displayed.

"PerformanceTest requires a DirectX 9.0c component from the October 2006 release of DirectX.
This is required even if you have DirectX 10. This is required for the 3D tests.
To install this component please select Yes.
This may take a minute.
Please note that this will not modify the current DirectX installation, but simply add additional DirectX functions.
To ignore and not display this message in future, please select No.
To ignore this message this time, please select Cancel."

Alternatively, you can run the DXSETUP.exe file that is included in the PerformanceTest install directory to install the DirectX9 component required. This will not overwrite, alter or affect your DirectX10 installation, but instead will add an extra interface into DirectX for DirectX 9.