search

FindHao / drgpu

A Top-Down Profiler for GPU Applications
BSD 3-Clause "New" or "Revised" License
13 stars 1 forks source link

A100.ini #3

Closed Diksha-Moolchandani closed 6 months ago

Diksha-Moolchandani commented 6 months ago

Hi,

I am trying to create a100.ini May I know what is FB_THROUGHPUT_FIX? What does FB refer to?

Also would you suggest some link to find these specs for a100, specifically TLB information?

FindHao commented 6 months ago

FB_THROUGHPUT_FIX = 1/ memory_bandwith * boost_clock https://github.com/FindHao/drgpu/blob/master/mem_config/gtx1650.ini#L17 It is a computed attribution to show the memory throughput.

I can't remember the exact FB meaning here, but it is the same with GPU_MEMORY.

https://github.com/FindHao/drgpu/blob/6a1ddf4e177ad6023a5b7b2fdd65a73d8edb6362/counters.py#L93

FindHao commented 6 months ago

For A100 memory config, I believe most information can be found in its whitepaper. I don't know the exact TLB related numbers in A100, but I assume you can still use 512KB as L2 slice, 2 uTLB per tpc, and 2 L1 tlbs per gpc.

Diksha-Moolchandani commented 6 months ago

ok thanks

Ruinhuang commented 2 months ago

@Diksha-Moolchandani Hi, i'm also trying to make a a100.ini. could you share you a100.ini file? I want to check that some of my values ​​are consistent with yours. Thanks very much!

Diksha-Moolchandani commented 2 months ago

a100.txt Let me know if these values make sense

Ruinhuang commented 1 month ago

@Diksha-Moolchandani thanks for sharing. it's same as my formula.