Total # of threads in the Intel Data Center GPU Max 1100

[black__edgar]

Hi all,

According to the information given in:

https://github.com/oneapi-src/oneAPI-samples/blob/master/DirectProgramming/C%2B%2BSYCL/Jupyter/gpu-optimization-sycl-training/02_Thread_Mapping_and_Occupancy/02_Thread_Mapping_and_Occupancy.ipynb

the total number of threads of an Intel Data Center GPU Max 1100 is 3584 (56 Xe-cores * 8 Vector engines  * 8 threads per vector engine).

However, on the following web page:

https://www.techpowerup.com/gpu-specs/data-center-gpu-max-1100.c4066

they report 7168 cores.

Is the information listed on the techpowerup.com page incorrect?

or

is there an additional correspondence between threads and core I don't know about?

Thanks.

+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Additional context:

I have been testing the performance (speed of execution) of a sparse matrix-vector multiplication algorithm (implemented in sycl) using the Intel Data Center GPU Max 1100 GPU and comparing it with the performance of an equivalent algorithm (implemented in CUDA) using the NVIDIA Tesla V100 GPU.

According to

https://www.techpowerup.com/gpu-specs/tesla-v100-pcie-16-gb.c2957

the NVIDIA Tesla V100 GPU has 5120 cores.

What I am finding is that the performance of the NVIDIA Tesla V100 GPU is consistently higher than the performance of the Intel Data Center GPU Max 1100 GPU.

I found these results hard to explain if the Intel Data Center GPU Max 1100 GPU had 7168 cores as listed on the thechpoerup.com page, but it is easier to justify if the number of cores (vector engines?) were 3584.

Unfortunately, I don't have access to the NVIDIA Tesla V100 GPUs, so I cannot implement and run my SYCL core on this GPU for a direct comparison.

Thanks again.

[Subhashish]

Hello black_edgar,

Greetings from Intel!

Hope you are doing well.

We have reviewed your request for obtaining information regarding the number of threads in the Intel DC GPU Max 1100. We would like to know why or for what purpose are you looking for this information. We understand that you are trying to test the performance (speed of execution) of a sparse matrix-vector multiplication algorithm (implemented in sycl) using the Intel Data Center GPU Max 1100 GPU and comparing it with the performance of an equivalent algorithm (implemented in CUDA) using the NVIDIA Tesla V100 GPU. Hence please help in confirming if you are part of the loaner program? Please help confirm on this so that accordingly we can guide you further with your request.

Looking forward to your prompt response.

Regards,

Subhashish_Intel.

[black__edgar]

Hi Subhashish_Intel,

I am a retired researcher looking to learn about SYCL by reading the Data Parallel C++ book and implementing an algorithm, which in my case is the sparse matrix-vector multiplication.

The learning process automatically leads to a comparison between SYCL and CUDA and the performance results of similar GPUs (Nvidia and Intel). I want to be sure that the results I am getting make sense and can be justified.

As far as I know, I am not a part of any loaner program. 

Thank you for your answer,

Edgar Black

[Subhashish]

Hello black_edgar,

Thank you for your response. Thank you for confirming on the loaner program. We are checking this and we will get back to you again at earliest.

Kindly await our next response.

Regards,

Subhashish_Intel.

[Subhashish]

Hello black_edgar,

Greetings !!

We are still working on your request and we will get back to you with an update. Currently this is the available document should you like to refer to the product specifications:

https://ark.intel.com/content/www/us/en/ark/products/232876/intel-data-center-gpu-max-1100.html

Regards,

Subhashish.

[Subhashish]

Hello black_edgar,

Greetings !!

We have checked this internally with our team and we would like to confirm that for Intel Data Center GPU Max 1100, the number of threads are 4096.

Please write back to us for any other query.

Regards,

Subhashish_Intel.

[black__edgar]

Hello Subhashish_Intel,

Thanks to you and the Intel team for the time devoted to my question.

However, I am still a little confused. In my original post, I wondered if the total number of threads of an Intel Data Center GPU Max 1100 was 3584 or 7168 (please, see the links provided in the original post). Now you are telling me that the correct number of threads is 4096.

Could you please double-check this?

Thank you very much again.

[Sachinks]

Hello black_edgar,

Greetings!

Thank you for your reply. We had checked this with our internal team and as per the information provided by them the number of threads in an Intel Data Center GPU Max 1100 is 4096.

However, we will check this again and get back to you with an update at the earliest.

Regards,

Sachin KS

[Subhashish]

Hello black_edgar,

Greeting !!

We would like to inform you that this product is not sold to the end users. The only information we have on this is the thread count which is 4096.

Regards,

Subhashish_Intel.

[black__edgar]

Thanks.

[Subhashish]

Hello black_edgar,

Greetings !!

Thank you for your response. Please let us know if we can go ahead and proceed to archive this ticket.

Regards,

Subhashish_Intel.

[black__edgar]

Yes. You can archive this ticket.

Thanks again,

Edgar Balck

[Subhashish]

Hello black_edgar,

Thank you for your response. We are proceeding to archive this ticket.

Thank you for choosing Intel!

Regards,

Subhashish_Intel.