Working precision in Nvidia A6000


My question is about Nvidia A6000 GPU.

What are the available working precisions to perform floating point operations on A6000?

The spec sheet does not mention any double precision processing speed. Does that mean, one ‘cannot’ run DP operations on A6000? In particular does that mean even if I specify my variable as DP it will be taken as SP by default?

Also, according to information on spec sheets, the SP processing speed of A6000 seems to be far higher than the A100 SP processing speed, but still my SP programs run faster on A100. Does it has something to do with memory bandwidth here? Why are SP computations slower on A6000 compared to A100?

context: I do scientific computing using Fortran90 with OpenACC parallelism (nvfortran compiler).
spec sheet links:

Thank you.

No. You can run DP operations. The throughput of those operations will not be as fast as SP operations. For A6000, the throughput of FP64 ordinary arithmetic should be about 1/32 the throughput of FP32 ordinary arithmetic.


It might. People often think that because their program uses arithmetic that the GPU arithmetic performance matters. But for a memory-bound code, even one that does arithmetic, usually memory performance is the strongest indicator of code performance.

1 Like

Thank you, that answers my question.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.