CPU mark composite average 45%+ lower than others

[College Smart]

I've got a Core 2 Quad Q9400 at 2.66Ghz, stock settings (Asus P5N-D) on W7 Pro 64bit. After running the CPU tests and comparing it to 3 other Q9400s, I've noticed a huge difference in performance.

Results:
Integer Math, 506.6 vs. baselines of 1683+ (230%+ lower)
Floating Point math, 1991.7+ vs. baselines of 2589.3 (30%+ lower)
CPU Physics, 70.1 vs. baselines of 211.1+ (200%+ lower)
String Sorting, 1034.6 vs. baselines of 1700+ (65%+ lower)

I'm familiar with computer hardware, but I'm no pro. Everything appears to be properly installed. The heat sync is properly set and my core temperatures idle at 34C. I recently sent my motherboard in to Asus of diagnostics and they claimed that it passed all tests.

I should also note: modern games such as Bad Company 2 and CoD: Black Ops seem to perform very poorly somewhat regardless of my video settings, almost as if there is a bottleneck somewhere in my system. I know it's not the GPU because I've tried it out in other PCs with similar specs and they're getting double my FPS.

System specs:
Core 2 Quad Q9400 @ 2.66Ghz
Asus P5N-D
4gb DDR2 KVR RAM
Radeon HD 4770
Windows 7 Pro 64bit

[wonderwrench]

Make sure you are using the 64 bit version of PerformanceTest and not 32 bit.
Also make sure PT is using all 4 cores. Edit, Preferences and set processes to 4.

Bill

[College Smart]

Ah, that's what I get for downloading anything from CNET. With the 64bit version and the program configured to run on all 4 cores my performance is on par with other baselines, however CPU physics and String Sorting are still around 230% and 70% lower, respectively.

I read somewhere that low physics results could be due to your GPU. I wasn't able to find anything within in the CCC to turn off physics processing as suggested to users with NVIDIA cards.

[wonderwrench]

Ah, that's what I get for downloading anything from CNET. With the 64bit version and the program configured to run on all 4 cores my performance is on par with other baselines, however CPU physics and String Sorting are still around 230% and 70% lower, respectively.

I read somewhere that low physics results could be due to your GPU. I wasn't able to find anything within in the CCC to turn off physics processing as suggested to users with NVIDIA cards.

Are you using the latest version build 1019. If you are using build 1015,1016 or 1017 the string sorting results will be low. The physics test in PT is run only on the CPU.

Bill

[College Smart]

Yeah, it's 1019.

[David (PassMark)]

We are looking at using hardware accelerated physics for the next major PT release. But at the moment it is CPU based using the Tokamak Physics Engine.

Can you post your new CPU results with the 64bit software.

[College Smart]

CPU - Integer Math: 1650.7
CPU - Floating Point Math: 2545.2
CPU - Find Prime Numbers: 925.1
CPU - SSE: 9.3
CPU - Compression: 5621.5
CPU - Encryption: 16.4
CPU - Physics: 67.0
CPU - String Sorting: 1028.0
CPU Mark: 3848.8
PassMark Rating: 1655.0

[David (PassMark)]

While you need to be careful not to compare your results to over clocked results, the numbers on the last 2 test are well down on what they should be.

I looked at ten other Q9400 Win7 64bit machines and none of them showed this behaviour.

Can you check the CPU temperature again, but do it while the CPU is under full load and has had a chance for the heat to build up. Say at the end of the CPU tests.

[College Smart]

Under load I seldom break 49C. Average of 46 during tests. I checked to make sure the heat sync was seated properly and it looks good.

For the first two year I owned the machine I used the stock heat sync, which was quite insufficient. My load temperatures were into the 70s, idling around 50 if I remember correctly. Everything I've read online indicates it probably didn't sustain any damage from those temperatures, though.

[David (PassMark)]

Maybe it is RAM or bus speed problem. These last 2 tests do tend to use more RAM than the other tests.

Just speculation, but maybe tests that run inside the CPU and the CPU's cache are fast and the tests that make heavier use of main RAM are slow.

This might also explain the poor performance in games.

Check your BIOS settings for the bus speed and RAM timings.

In the advanced memory test run a test using these settings,
- Read
- Memory speed per block size
- QWORD

Then look at the graph.

On my machine (i5 750, 6GB RAM) I get a peak speed of ~10,000MB/sec for cached data, then it drops right of to 3300MB/sec for main RAM.
http://www.passmark.com/images/forums/ramspeed.png

[College Smart]

I see. My write speed is fairly atrocious at a score of 846. It's 4, 1GB sticks of Kingston Value RAM.

I ran a CPU and RAM test, then compared it to a baseline that used the same processor/clock speed and DDR2 ram. My write speed was 130% lower and my physics and string sorting tests were 257% and 189% lower, respectively.

[College Smart]

http://img574.imageshack.us/img574/3488/56797089.jpg

And there are the test results.

[David (PassMark)]

The memory read speed graph above looks OK. (an average of 10,275MB/sec is fine)
Can you do the same graph for Write, if you think writing is slow.

The write speed should be within maybe 15% of the read speeds.

[College Smart]

http://img220.imageshack.us/img220/1302/42303003.jpg

Here are my write results; they look about 25% lower.

[David (PassMark)]

The values when writing data to L1 and L2 cache seem OK (the first 2 steps in the graph). But you have a big step down when it comes to main RAM. The Q9400 CPU has 6MB of L2 cache. Which is why you see the step down at this block size.

From the 2 graphs above, at the 8MB block size level, your read speed is 3000MB/sec. But write speed is ~600MB/sec. Which is way more than 25% lower. (if you export the data, you can get the exact values).

For comparision here are 3 results. All write speeds at the 8MB block level with 64bits.

Core 2 E8500 - 2603 MB/Sec
Core 2 E9650 - 2810 MB/Sec
AMD Phenom II B55 - 3628 MB/sec

So the interesting thing now is working out what is causing the problem. Start by looking at the BIOS settings. Check timings, bus speed, dual channel RAM setup, try pulling out half the RAM at a time, or swapping the RAM if you have some spare. What MB do you have? Is the BIOS up to date?

[College Smart]

It's an ASUS P5N-D. My BIOS is not up to date, but I don't see any notes related to RAM in the last 3 or so updates I don't have. I'll check out the timing and such on my ram, though I believe it is all set to stock.

I've tried swapping ram out 1 stick at a time to no avail. I actually have 3 sticks of Kingston Value that came from the manufacturer and 1 Crucial stick I got on Newegg. I'll try it again and do more thorough testing.

[David (PassMark)]

Try with 2 matching sticks at a time in the same coloured sockets on the MB, otherwise you won't get the benefit of dual channel. Is this all DDR2 800 RAM or better? I note that the Qualified Vendors List (QVL) for RAM on the ASUS site claims that some RAM doesn't support 4 sticks for two pairs of dual channel RAM (and using sticks from different vendors isn't the best for dual channel performance).

[College Smart]

Okay, I've made some odd discoveries.

Using CPU-Z, I got these results for timing with all 4 sticks inserted:
http://img837.imageshack.us/img837/5466/87332277.jpg

Looks like it was adjusting to the odd RAM, which was 5-5-5-15. Also note that it says single channel rather than dual.

Here's what it looked like with only 2 sticks in the same colored DIMMs.
http://img225.imageshack.us/img225/7123/58644969.jpg

Now it says dual channel. After looking at my MB user guide, it appears that my KVR does work in dual channel if you have either 1 or 2 sets of matching RAM.

Test results with this 2GB config were about the same.
http://img266.imageshack.us/img266/8798/33213626.jpg

Oddly enough, when I ran Black Ops, I was getting 60 FPS on average (up from about 35). Adding the third KVR stick which put it into single mode put me back down to 35 average. I'm really not sure how that works.

A friend of mine suggested that I manually adjust the tRAS in my BIOS to something like 15, as it appears quite high.

[wonderwrench]

Can you post screen shots of the SPD tab so we can see the specs for each module. BTW it does seem strange that the tRAS is at 31. I see you are using the latest version of CPU-Z. I wonder if the newer versions report incorrect info on Nvidia based boards? It looks like the board is down clocking the memory from 800 to 667 when more than 2 modules are installed. This is not uncommon but not preferred.

Bill

[College Smart]

Here are the screens from the SPD tab. There are 3 sticks of the Kingston RAM and 1 stick of the G-Skill RAM (I previously mistook this for Crucial brand memory). Clearly they're different timings. Perhaps this is causing it to say single channel?

Would it be possible and/or safe to manually set the timing in my BIOS so they're all the same? And if so, which timing would you recommend?

Kingston:
http://img541.imageshack.us/img541/7175/25560964.jpg

G-Skill:
http://img87.imageshack.us/img87/4109/47410083.jpg

[David (PassMark)]

I was getting 60 FPS on average (up from about 35). Adding the third KVR stick which put it into single mode put me back down to 35 average. I'm really not sure how that worksWhen running with just the 2 sticks in dual channel mode, did it also fix up the low PerformanceTest benchmark results?

There really shouldn't have been such a large difference in frame rates (nor in the CPU benchmark test) between using dual channel or not. Maybe the 3rd stick is just bad in some way. The memory test charts above only test out to 512MB of RAM, so abysmal performance in the 3rd stick might be missed. An interesting test might be to just boot the machine with the suspect 1GB of RAM.

[College Smart]

When running with just the 2 sticks in dual channel mode, did it also fix up the low PerformanceTest benchmark results?

There really shouldn't have been such a large difference in frame rates (nor in the CPU benchmark test) between using dual channel or not. Maybe the 3rd stick is just bad in some way. The memory test charts above only test out to 512MB of RAM, so abysmal performance in the 3rd stick might be missed. An interesting test might be to just boot the machine with the suspect 1GB of RAM.

Yeah, it was weird because the tests didn't show any improvement. I tried testing each stick on its own and they all got the same results. I also tried using different paired combinations of the KVR ram to see if FPS would change and it stayed the same. I didn't check my FPS, but when running the G.Skill stick with one KVR stick, it was in single channel and my test results were identical to the others.

[wonderwrench]

Running 3 modules even matching ones will hamper performance some what because dual channel will be lost or at least hampered. You can try setting the timings manually. Use the timings from the stick with the bigger numbers. Start with the memory at 333 mhz/DDR 667 5-5-5-15-20 timings as shown in the SPD tab going from top to bottom. If that works you can try 400 Mhz/DDR800 6-6-6-18-24 timings. If both work DDR 800 may or my not be faster than DDR 667. This is because DDR 667 runs at 1 to 1 FSB/DRAM ratio. DDR 800 runs 5 to 6 FSB/DRAM ratio. There is a penalty when running the front side bus and memory at different speeds aka asynchronously.

You may need to increase the last number (tRAS) to get one or both manual settings stable. Your board sets tRAS to 31 on auto. I recommend downloading Memtest86+ and booting from it to test stability. This way you can't hose your OS if its not stable. If you have a floppy drive download the floppy version. If not get the ISO version and burn it to a CDR. Windows 7 pro has built in support for burning ISO files, just double click the ISO to bring up "windows disk image burner".

Boot from the disk you made to make sure it works. You may need to change the boot order in your bios to boot from the disk you made aka cdrom or floppy. Memtest will run all by its self with no intervention if you successfully booted from the disk. Esc will exit Memtest and reboot your PC. Once you have the boot disk working try setting the memory manually from the bios, save and exit. Once again boot from the Memtest boot disk. Let it run for several passes, it takes a while. Memtest will show memory speed and timings plus memory performance. If memtest shows no errors it should be safe to boot into windows and see what PT says.
http://www.memtest.org/

Bill

[College Smart]

Okay, I tried all 4 sticks at 5-5-5-15. Memtest checked out fine so I booted Windows. CPU-Z detected my entered timings just fine but said it was still single channel. I ran Performance Test and got the same results as before. Then, I tried 6-6-6-18 with just 2 sticks of KVR. Memtest checked out, CPU-Z showed my timings, but Performance Test still gave me the same results.

I'm wondering if maybe I should pick up a couple gigs of higher quality ram and sell what I've got. If my games run fine on 2 gigs I don't mind so much, and some quality RAM might alleviate this problem.

[David (PassMark)]

Would be good if you could borrow some to test before spending the money. While there is a fair chance it is RAM related it isn't 100% sure. Could still be BIOS or MB.

I did some more research and found a few other baselines (with the same MB) that have the same low results. For example Baseline numbers BL285667 and BL247088. In both cases they seem to have unmatched RAM and lower FSB speeds.

[College Smart]

Would be good if you could borrow some to test before spending the money. While there is a fair chance it is RAM related it isn't 100% sure. Could still be BIOS or MB.

I did some more research and found a few other baselines (with the same MB) that have the same low results. For example Baseline numbers BL285667 and BL247088. In both cases they seem to have unmatched RAM and lower FSB speeds.

Good point. I'm almost positive a friend of mine uses a DDR2 RAM, so I'll have to check with him.

[wonderwrench]

Yes testing different memory without spending money upfront is a very good idea. Its possible this problem is related to the Nvidia chipset so it would be nice to know there is a fix before spending money on ram.

BTW did you take note of the memory performance differences in memtest? Oh and the big thing is getting matching modules. I have used KVR in several builds and it has worked flawlessly for me. Money being no object I like Corsair and Crucial best.

Bill

[wonderwrench]

If you end up buying new ram go for two 2 gig modules if you want 4 gigs total. If your in the US Newegg has a 4 gig kit of Corsair for $80 (http://www.newegg.com/Product/Product.aspx?Item=N82E16820145241) with free shipping. $70 after $10 rebate. I use this exact memory in two of my three PC's and find its rock solid.

Bill

[College Smart]

Okay, a friend lent me 2, 2GB sticks of Buffalo premium RAM. I got some interesting but disappointing results.

First, the Performance Test results were a better, but overall still underperformed compared to other machines with identical CPU/Motherboards (not taking into consideration over-clocked processors).

Here are the CPU results:

CPU - Integer Math: 1249.9
CPU - Floating Point Math: 1919.4
CPU - Find Prime Numbers: 700.0
CPU - SSE: 7.1
CPU - Compression: 4345.5
CPU - Encryption: 12.4
CPU - Physics: 137.3
CPU - String Sorting: 1749.0
CPU Mark: 3158.2
PassMark Rating: 1358.0

So it's better than before, yet still lower than comparable machines.

Here's the graph for write results:
http://img600.imageshack.us/img600/7066/85702313.jpg

Again, these look better than before, but perhaps not as good as they could be?

And the Memory and SPD tabs in CPU-Z:
http://img20.imageshack.us/img20/5296/80130135.jpg
http://img694.imageshack.us/img694/9436/50039296.jpg

So I haven't really changed the timing on the RAM. The sticker on it says CL5 so perhaps I could change the tRAS to 15?

Also, my FPS is identical to using just 2 sticks of my KVR.

[David (PassMark)]

So it's better than before...Indeed. Previous results, with mismatched RAM in single channel,
CPU - Physics: 67.0
CPU - String Sorting: 1028.0

Results with new matched RAM
CPU - Physics: 137.3 (105% improvement)
CPU - String Sorting: 1749.0 (70% improvment)

Plus the advanced memory test also reports better write speeds in main ram (at the 8MB block size). Maybe double what it was.

So this is a victory !

But I totally agree that something else is going on as well. All your CPU results, across the board, are now down a bit on where they should be. It also doesn't explain some of the other stuff you noted above.

Are you defintely running on 4 cores. Also check the edit / preferences window in PerformanceTest. The PT result almost look like you only have 3 cores available. Each result is now almost exactly 75% of what I would expect the value to be. Where has the 25% gone?

[wonderwrench]

Okay, a friend lent me 2, 2GB sticks of Buffalo premium RAM. I got some interesting but disappointing results.

First, the Performance Test results were a better, but overall still underperformed compared to other machines with identical CPU/Motherboards (not taking into consideration over-clocked processors).

Here are the CPU results:

CPU - Integer Math: 1249.9
CPU - Floating Point Math: 1919.4
CPU - Find Prime Numbers: 700.0
CPU - SSE: 7.1
CPU - Compression: 4345.5
CPU - Encryption: 12.4
CPU - Physics: 137.3
CPU - String Sorting: 1749.0
CPU Mark: 3158.2
PassMark Rating: 1358.0

So it's better than before, yet still lower than comparable machines.

Here's the graph for write results:


Again, these look better than before, but perhaps not as good as they could be?

And the Memory and SPD tabs in CPU-Z:
http://img20.imageshack.us/img20/5296/80130135.jpg


So I haven't really changed the timing on the RAM. The sticker on it says CL5 so perhaps I could change the tRAS to 15?

Also, my FPS is identical to using just 2 sticks of my KVR.

About the ram timings, CL = CAS latency. The CAS spec is 5 and CAS shows to be at 5. The tRAS spec is 18 and it shows as being at 18 so there is nothing that needs changing.

Bill

[College Smart]

But I totally agree that something else is going on as well. All your CPU results, across the board, are now down a bit on where they should be. It also doesn't explain some of the other stuff you noted above.

Are you defintely running on 4 cores. Also check the edit / preferences window in PerformanceTest. The PT result almost look like you only have 3 cores available. Each result is now almost exactly 75% of what I would expect the value to be. Where has the 25% gone?

Yeah, I checked it. Number of processes = 4. I also see activity on all of the core monitors in the processes tab. I checked to make sure the heat sync was attached properly and my processor was clocked at the regular speed and it all checked out.