Why less L2 cache size in IA-32 based core in Intel EP80579 ?

[DPath]

Dear All Hi,

I have a query, which is an elementary one but please give your views on the same.

Intel EP80579 CPU core IA-32 based is having less size of L2 cache i.e 256 KB as compared to previous versions for example Intel Celeron M 373 having 1 MB of L2 cache. On comparison both are having same size of L1 cache. But what is the reason of reducing the cache size in the new versions of CPU's?

Thanking You,

Regards,

Deepak

[FMcNu1]

Hi Deepak and welcome to the Intel® Embedded Community.

I can't say exactly why the cache size is different but let me point out that the EP80579 is a specialized System-on-Chip (SoC) designed and optimized for communications, storage, and other special embedded applications. Celeron is a general purpose PC processor. Therefore comparing the cache specifications of these two very different device classes is really not relevant. If you want to compare "apples to apples", I suggest you look at specs for Intel® Core™ or Intel® Xeon®.

I hope this helps.

Felix

J. Felix McNulty

Community Moderator (Intel contractor)

[DPath]

Thanks Felix and its a pleasure being a part of the community!

I understand and agree with the point that Intel EP80579 is specialized SoC meant for embedded applications, Celeron is a general purpose CPU and hence no comparisons.

My concern with a cache size is because if I am running an Embedded Linux operating system on the Intel EP80579 and Celeron M 373 will I be able to achieve better performance in the former one in terms of data transfer and processing as compared to the latter or not? or

Having a low size L2 cache can effect the performance when transferring data sizes like 128 KB or higher in Intel EP80579 as compared to Intel Celeron M 373 processor?

Is it advicable to run an operating system on Intel EP80579 or it should support specific embedded applications running on it?

I know alot of questions but surely the community is very helpful and I wish to contribute soon, Thanks in advance!!!!

Regards,

Deepak Pathania

[FMcNu1]

Hi again, Deepak,

Let's see if we can find someone who can answer your question about the cache.

In the meantime, I do believe the standard EP80578 usage model is to run an OS. This part is supported by the mainstream OS / RTOS vendors.

Felix

[DPath]

Hello Felix,

I believe community will soon answer

Out of all the question, one question still is the most important one for such a strong processor of 1.2 GHz why one wishes to use such low size cache (256 KB, 2-way set associative) in Intel EP80579?

Thanks and Regards

Deepak

[FMcNu1 · Accepted solution]

Hi

Here is some information I obtained from the EP80579 design team that might answer your question about the cache size.

The EP80579's IA core came from a particular SKU of Pentium M. They chose a model with a smaller cache as a design tradeoff to reduce the footprint of the core portion of the SoC. As you know the SoC integrates functionality in addition to the processor core, including GbE MACs, and these additional functions expand the overall footprint. The team acknowledges your point that compared with Celeron M, the smaller cache may impact performance is certain systems. However they believe that for the telecom applications targeted, EP80579 still provides good performance.

If you haven't aready see it, below is a link to the Intel® Embedded Design Center (EDC) site, where performance information like L3 Forwarding and IPSec measurements for EP80579 can be found. All of the EP80579 specs can also be found here.

https://www-ssl.intel.com/p/s/en_US/embedded/hwsw/hardware/ep80579/overview https://www-ssl.intel.com/p/s/en_US/embedded/hwsw/hardware/ep80579/overview

If you have more specific technical questions, I suggest you check out the Intel® e-Help desk. e-Help is staffed by Intel representatives who support select Intel embedded platforms including EP80579.

e-Help is here: http://embedded.communities.intel.com/community/en/intel_e-help_desk http://embedded.communities.intel.com/community/en/intel_e-help_desk

To access e-Help, you need to be a Privileged member of the Intel® Embedded Community. If you are not already a Privileged member, you can request an upgrade to your community account here:

https://edc.intel.com/My-Account.aspx https://edc.intel.com/My-Account.aspx

I hope this helps.

Felix

[DPath]

HI Felix,

Thanks very much!! Your last post was really very helpful. I had a quite an interesting time knowing about the Intel EP80579 SoC.

While working on the same I have encountered some observations. I have a query regarding the CPU performance again :

I used the Intel suggested LMBench Tool for benchmarking and comparing, the performance of Intel EP80579 IA based CPU core and Intel Celeron M 373.

The size of the L2 cache differ in both of the processors Intel EP80579 is having lower (256 KB) cache than Intel Celeron M 373 (1 MB) and

also as acknowledged in the last post that lower cache was tradeoff for reducing overall footprint of the SoC.

Hence, in order to measure their performances, I disabled the cache's in both of the processors, by the setting CD bit 31 as mentioned in the

datasheets. Taking out the possibility which could hamper the performance of the Embedded SoC CPU (Intel EP80579).

Now, with cache disabled, since Intel EP80579 is 1.2 GHz, 533 MHz FSB, 1 GB DDR2@800MHz should give better performance than

Intel Celeron M 373 1.0 GHz, 400 MHz FSB, 1 GB DDR@333MHz.

But I notice while measuring Memory Read Latency, Memory Bandwidth(read, write, copy etc.) still Intel Celeron M 373 is better than Intel EP80579

IA based Core. I am still going through the data sheets to find out the exact reason for lower performance.

Please suggest what is the probable reason or which part on the datapath from CPU to memory except cache could be causing this issue?

Is there anything more parameters than cache size parameter, while moving a general prupose processor to an embedded processor?

Thanks in Advance! your suggestions and replies are very helpful.

With Warmest Regards,

Deepak

[FMcNu1]

Hi Deepak,

It sounds like you are digging deeply into comparing the EP80579 SoC against a general purpose CPU. I think it would be very helpful if you have a relationship with Intel technical team who can help you analyze your application to recommend the best solution. I have the feeling that such analysis exceeds what you might achieve from the community discussion.

Do you have any contact with local Intel representatives? If not, you can initiate an Intel contact here: http://edc.intel.com/Get-Help/ http://edc.intel.com/Get-Help/

I hope this helps,

Felix