vans554

3 ms·

1.
by vans554 3y ago
Pretty interesting. Using nightly + cu121 im getting 8.18 it/s, another 5% improvement vs 7.78 that cu118 gave.
2.
Use pytorch2+cu118 with ADA hardware for 50%+ speedup(gpux.ai)
141 points by vans554 3y ago|35 comments
3.
by vans554 3y ago
I accidentally stumbled upon this and did not expect such a speedup. It seems anything less than cu118 does not properly support the RTX4090 (or H100). Bumping to cuda12.2 with pytorch2.0.1+cu118 made my SDXL go 50% faster and ESRGAN 80% fa
4.
Karlo Diffusion Endpoint(explorer.gpux.ai)
1 points by vans554 4y ago|1 comments
5.
by vans554 4y ago
I put Karlo diffusion by Kakao running as an inference endpoint on GPUX. Karlo Diffusion is more controllable, similar to Dalle-2. Make some images here https://explorer.gpux.ai/inference/gpux/karlo
6.
by vans554 4y ago
Thanks.
7.
by vans554 4y ago
Yea but this is irrelevent unless you implying that FILM is better quality while nvfruc is lower quality but faster so applicable for realtime? if nvfruc is better quality + faster then the comparison is that nvfruc is better.
8.
by vans554 4y ago
very cool how does this compare to NVIDIA nvfruc? https://docs.nvidia.com/video-technologies/optical-flow-sdk/...
9.
by vans554 4y ago
People still go to traditional-school?
10.
by vans554 4y ago
Much needed thank you for doing gods work.
11.
Show HN: GPUX, scalable inference for AI Models(explorer.gpux.ai)
2 points by vans554 4y ago|0 comments
12.
by vans554 4y ago
DDD is just microservices without head of line blocking not sure why everyone hates it. Instead of having 5 microservices you just build an umbrella app exposing the same 5 API. Calls into the umbrella are now at microsecond speed vs 10s of
13.
by vans554 5y ago
Mouse lag needs to be fixed.
14.
by vans554 5y ago
Would be cool to move realworld assets into virtual worlds, and vice versa.