When GPU makers quote performance in FLOPS...
> GeForce GTX 680: 3 TFLOPS
> Radeon RX480: 5 TFLOPS
> Tesla P100: 10.6 TFLOPS
Are they all using the FP32 FMADD instruction? What code are they executing to get these numbers, just one instruction in a loop across all cores?
Its the number their sales team come up with
It's how many floppy dicks can fit in their CEO's mouth and the time the card is released.
>>55145839
>Shillary 30 FLOPS
>>55146037
as much as i'd love to agree that their marketing droids pull these numbers from their ass, i suspect there's some actual benchmarking going on. is there some simple CUDA program (PTX assembly?) you can run to verify these FLOPS?
>>55145839
>>55146037
It's a theoretical peak. Assume you can have a program executing a floating point operation on each core each cycle. Off of this you can get an idea of how much FLOPS you can get on a real world application.
>>55146738
i see. thanks