Learning Forum

[skikuchi]

According to the document, GPU Scaling for Discovery Live is applied "Single GPU" and "Single Node".

Even though Tesla M60 has 2046 x 2 core, I can use only 2046 core and 8GB with Tesla M60 on Discovery Live.

Is this right?

https://www.nvidia.co.jp/content/gpu-applications/PDF/gpu-applications-catalog.pdf

[Naresh Patre]

Hello Sho

I am checking with the team and will let you know.

[Naresh Patre]

Hello Sho

Yes, Since Discovery Live is "Single GPU Single Node", it will only use per GPU cores and memory i.e. 2048 cuda cores and 8 GB memory

[skikuchi]

Hello Naresh Patre

Thank you for your answer. I understand it.

One more question.

Tesla M60 has 2048 cuda cores and 8GB memory. Quadro P4000 has 1792 cuda cores and 8GB memory.

Are there anything differences with same model on Discovery Live such as results, speed and so on because of amount cuda cores even though same memory ?      

[Naresh Patre]

Hello Sho 

The fidelity range (and therefore accuracy of results) is solely determined by GPU memory. In that regard, the cards should solve the same set of problems. As for speed to solve, this is determined by a number of factors (# cores, core clock, memory bandwidth) and depend on the specific geometry being simulated as to which is the limiting factor. 

The difference in speed between these two cards is not going to be all that significant. The M60 is likely faster.

[skikuchi]

Hello Naresh Patre 

Thank you so much!