Hello. Looked through this forum, but can only find a bit of info here and there about FP32 or FP64 Application requirements.

Planning to use this card for GPU grid once tasks resume.

Would people recommend going the RTX route (better FP32) with a 40 or 50 series, or the Tesla route (better FP64) such as a V100 or P100? Also considering AMD, but doesn't seem very popular on the leaderboard.

Thank you in advance.

GPUGrid currently (and presumably for the foreseeable future) only has CUDA applications available, so if you want to participate in this project, you will want a modern-ish Nvidia GPU.

There has been some recent experimentation with running CUDA apps on AMD GPUs using a compatibility library, but I wouldn't count on this ever being officially supported.

As for FP32 vs FP64, that'll vary based on the application, of which GPUGrid has several. I'll leave it to those more experienced in how the applications works to answer which GPU line would perform better, but on my machine with a GeForce GTX 1060 6GB (which has relatively weak FP64 performance), I never felt like my contributions to this project were going to waste. That's as opposed to, say, Asteroids@Home, where crunching with my GPU was only about 50% faster than a single CPU core.

All of the apps here run a Python environment now and mostly respond to greater parallelism and high CUDA core counts and clocks. So the more modern and high-end the Nvidia card, the better.

it depends on the app.

ATM/ATMML and ACEMD3 all favor the FP32 high core count type cards. RTX 40 and RTX 50 series are likely to do well.

QChem instead favors FP64 performance and large VRAM. so cards like P100, Titan V, V100, etc will perform a lot better.

there are only CUDA apps here. AMD is not an option unless you setup ZLUDA.

---

I noticed that the runtime of all types of ATMML tasks which my hosts processed during the past few days was considerably longer than before, by about 40-60%.
Since this is the case with all 4 hosts on which I crunch these tasks, a technical failure on a given host can be precluded.
Does anyone else make the same experience?

I'm seeing the same thing. Just a bigger input parameter set than prior tasks I assume.

I'm seeing the same thing. Just a bigger input parameter set than prior tasks I assume.

and less credit points compared to some time ago :-(

Credits per task are fixed by the admins. Same credit as before. But there is no longer any bonuses applied since the server move if that is what you meant.

Credits per task are fixed by the admins. Same credit as before. But there is no longer any bonuses applied since the server move if that is what you meant.

oh, okay, your are right. I didn't catch right away the difference in credit points is the bonus which is obviously no longer granted. Any idea what they skipped the bonus ?

We have no idea, we've asked. Forgot to enable it?? Don't know where the code is?? Decided it was not important anymore to get faster turnover of tasks?? Who knows?