IPP 5.1: AMD vs. Intel

[ligoban]

I've got another question. After I tested Intel's IPP Speech coding
samples on AMD Sempron 1.8GHz and on the Intel Celeron 2.54GHz (960MB
memory in both cases) I've got an interesting result: AMD is on average
for 80% faster then Intel's CPU. What is the magic? Is this okay?
Thank you.

[Vladimir_Dudnik]

Hi,

actually it is interesting. Could you please verify that in both cases your application used optimized code? The simplies way to do that just to print out information from ippGetLibraryVersion() call.

It will not be OK if you get the same "interessting" result on the latest Intel Core 2 Duo architecture.

BTW, do you use IPP in DLLs or in static libraries? Have compiled sample with Intel compiler, or some other?

Regards,
Vladimir

[ligoban]

I used in both cases the same code. Also I used DLLs. Sources were compiled using MS VisualStudio 2006 in both cases too. So I created the same circumstances.

I 'll check info about optimized code with ippGetLibraryVersion() later.

Thanks.

[jnoring]

The Intel Celeron was not one of Intel's more memorable processors; by any metric, it was a dog compared to the competition. A lot of it had to do with the fact that it was a P4 with three quarters of itsL2 cache disabled, and the P4 was notoriously dependant on having ample L2 available. It doesn't surprise me that the Sempron eats it for breakfast.

That being said, if you swapped in a new Core 2 Duo, there is no doubt it my mind that the tables would be (very) turned. It'd be really interesting to see how an A64 4800+ X2 stacked up against, say, a Core 2 Duo E6600 though. Do you have benchmarking code you can share?

[jnoring]

...a bit more info:

http://www.xbitlabs.com/articles/cpu/print/sempron.html

...if you specified what chip models it was, rather than frequency (i.e. is it a Celeron D, which is based on the Prescott core, or is it a Celeron based on some of the previous Northwood cores?), then it would further explain it. If you're comparing one of the new Sempron chips based on the Paris core to one of the old Celeron chips that use the Northwood core, then the results aren't surprising. (note the benchmarks contain a lone 2.8 GHz Celeron based on the Northwood core that consistently gets whooped by just about everything. If yours is 250 MHz slower, then it's just that much worse)

[ligoban]

What do you mean saying "benchmarking code"? I used Intel's sources I downloaded from Intel's site. It is open code. By the way, the number of Cores is not interesting for me. I see that I can can get better result on the cheaper CPU. So, thats okay.

In any case, if somebody will got new information about IPP competitions - it can be interesting (I think) not only for me.

[Vladimir_Dudnik]

I think it is possible to find information about benefits provided by the latest Intel u-architectures, Intel Core 2 Duo in many places in the Net, isn't it?

Vladimir

[Vyacheslav_Baranniko]

Hi

Basically, no wonder if you will get the similar comparison result for AMD Sempron vs Intel Celeron performancemeasurementon any IPP domain library and samples.

So, I would suggest you to try IPP onIntel Core Solo/Duo processors.:)

Slava

IPP, Speech coding