Associate
- Joined
- 19 Jun 2017
- Posts
- 1,029
I was doing some arm chair stuff regarding shader throughput:
Workload distribution
Int 35%
Fp16 13%
Fp32 52%
3080 expected output, full utilisation at 1.71 GHz:
11.1 tera IOPS
4.1 half precision teraflops
16.6 single precision teraflops
Total 31.8 tera ops
There's a hard cap on int units so this is just a very good approximation
I am looking into the rx5700 to extrapolate performance of big navi, haven't been successful yet
Workload distribution
Int 35%
Fp16 13%
Fp32 52%
3080 expected output, full utilisation at 1.71 GHz:
11.1 tera IOPS
4.1 half precision teraflops
16.6 single precision teraflops
Total 31.8 tera ops
There's a hard cap on int units so this is just a very good approximation
I am looking into the rx5700 to extrapolate performance of big navi, haven't been successful yet