Intel vs. AMD quad-core. Who is the best?
first a short info on what my concern is.
I work at a company developing CFD Software (Computational Fluid Dynamics) and in order to give our customers the best advice on which computer system they should invest for best performance with our software I would need your help.
The computers our customers would buy are usually CAD workstations. So standard would be a Quadro FX GPU and between 4 and 8 GB RAM but recently more often 16 or even 32 GB RAM due to the CFD usage.
Since our software (like every CFD software) is capable of parallel processing (MPI-technique) it is best to use as many processors as possible within on system to solve the calculations as fast as possible.
Sometimes I get really confused on all the different processor types and how to compare them best. I found your site and thought I'll give it a shot and put some questions into this forum.
1. I recently read an article about HP was doing a benchmark to advice their customer for the best system to their needs. They said Intel is better in single core performance (the single cores are faster) but AMD is better when using several cores do to their HT-Technologie (no Bottleneck in memory adressing and storing) and therefore AMD scales better than Intel. Can you confirm that?
2. I haven't found the new AMD Phenom on the list, do you have any numbers about this processor type?
3. Does anyone know if there are already motherboards for 2 or 4 Processors for Intel and AMD that have also a high maximum of RAM (32 GB or more). They are often hard to find if not in a full system from Dell, HP or any other vendor.
Thank you for your help!
Just as background to other readers.
HT = HyperTransport (from AMD) and not HyerThreading from Intel. HyperTransport is the AMD technology for connecting the CPU to the RAM.
The Phenom does appear in the graphs. But it's performance isn't spectacular.
The current Intel chips like the Xeon 5482 and Core2 X9770 offer (in our opinion) better performance in both single and multithreaded applications than the Phenom range.
If anything this advantage will be extended even further with the new Intel Nehalam chips (which have similar HT technology). Unless AMD pull a rabit out of the hat before year end.
However AMD probalby does still have a minor advantage if your CPUs are idle while waiting for new data to arrive from main RAM. i.e. the CPUs never reach 100% utilization when using your CFD applications becuase there is nothing available to process. As the memory throughput in multithreaded applications can be better on the new AMD systems (until Nehalam arrives).
Forget about question 3, I found some.
Sorry, I found the Phenom, I wanted to ask for a dual CPU of the Phenom.
Also there are no newer Opterons of the Barcelona model.
Where do you get these benchmark results? Is it from customers of your software or do you test different systems yourself?
Maybe there is a bigger difference between the Phenom and the Barcelona. In the article I read, they said that the Barcelona reaches almost 4 times the performance with all 4 cores compared to only one core and has therefore an almost linear scaling and with a second CPU he reached about 6.5 times the performance. The Intel Xeon Harpertown reached about 3.5 times than with a single core and with a second CPU (now 8 cores) Intel reaches only little more than 4 times the performance.
The Intel core is faster when comparing the single core but is scaling not so good within one CPU and even worse with two CPUs.
Is there a release date for the Nehalam kown?
Thank you for your help.
As far as I am aware there are no dual CPU Phenom systems. I think they want to keep the dual / quad CPU market for the Opterons. This allows them to keep a price preimum on the Opterons.
The benchmark results come from our own test systems and from results submitted by users.
Wikipedia is saying late 2008 for Nehalem
Originally Posted by passmark
NO, the dram controller has it own means to talk to memory aka DD2 or DD3 current protocols.
HyperTransport is used to commicate between processors and other IO IC's.
If you really want to sell something, re-write your code to use "CUDA" for the math to be run on the GPU, then you won't really care which CPU it is, other than the CPU's need to be fast enough to keep the GPU filled with work.
Yes you are correct. HT replaced the traditional front side bus and the memory controler is in the CPU. But the point remains that all else being equal the AMD chips with HT and built in memory controller should have faster RAM access.