Folding Forum

Any way to reproduce the Intel folding issue on vanilla open source OpenMM? I could play around with it a little.

Damn no HIP still.

Still curious to compare the RX 7900XTX with HIP to the RTX 5080, which is one of the reasons I haven't looked too hard for 5080s yet! I have been reading the OpenMM GitHub but haven't seen many updates. Last I read was something you wrote in 2023.

7900xtx on hip is equal to 4080
To reproduce intel stuff you need intel b500 series gpu

Is there much room for improvement? Its TDP is about the same as a 5080, so if it can eventually get closer to its performance then it will be worth it.

And I don't have that Intel GPU series, oh well.

5080 is one generation newer.
We need hip FAHcore first, then I can talk to AMD to improve things

muziqaz wrote: ↑Wed Apr 23, 2025 5:44 am

arisu wrote: ↑Wed Apr 23, 2025 4:42 am What issues was core26 having? I've never experienced any issues with it on Nvidia with OpenCL, Nvidia with CUDA, or AMD with OpenCL.

Not seeing CUDA on certain OS hardware combinations.
Core folds on CPU if Intel GPU is present.
Core fails on certain AMD hardware scenarios.
No HIP (not an issue, but a ticket to mental institution for certain individual )

Core27 still has intel issue and still has no HIP

OpenCL libraries detection issues on Linux too ...

What version of OpenMM is Core27 based on?

8.2.0 based on logs?

OpenMM on core26 and core27 is the same, and should be the same on next core release

muziqaz wrote: ↑Wed Apr 23, 2025 9:14 am 5080 is one generation newer.
We need hip FAHcore first, then I can talk to AMD to improve things

It looks like despite being a generation newer, the 5080 has worse FP32 performance per watt than the 7900XTX (6.39 vs 5.78 watts per TFLOP).

And yet with HIP it is only as fast as the 4080 for folding. I don't know if that's because there's still more performance to be squeezed out of HIP, but I'm seriously considering discontinuing my search for near-MSRP 5080s and instead focus on an all-AMD system (I have everything but the GPUs).

Hip does not care about performance per watt. 5080 has more performance

Oh I see. Then AMD must be overstating their FP32 figures, or perhaps Nvidia is understating theirs! 5080 is 56.28 TFLOPS (360W), 7900XTX is 61.39 TFLOPS (355W). At least officially. That led me to believe that, even with HIP, there was some overhead that caused it to perform more poorly in folding.

I guess there is a lot more to folding performance than raw FP32 TFLOPS.

Tflops is marketing nonsense. It died long time ago.
AMD has this dual issue design, which kinda doubles the tflops. While they have no intention to optimise for that neat hardware feature, marketeers make sure to still use bigger number better

arisu wrote: ↑Wed Apr 30, 2025 4:40 am

muziqaz wrote: ↑Wed Apr 23, 2025 9:14 am 5080 is one generation newer.
We need hip FAHcore first, then I can talk to AMD to improve things

It looks like despite being a generation newer, the 5080 has worse FP32 performance per watt than the 7900XTX (6.39 vs 5.78 watts per TFLOP).

And yet with HIP it is only as fast as the 4080 for folding. I don't know if that's because there's still more performance to be squeezed out of HIP, but I'm seriously considering discontinuing my search for near-MSRP 5080s and instead focus on an all-AMD system (I have everything but the GPUs).

If nothing else, it would be a second person (other than Muziqaz) to bug AMD devs into finally getting HIP out in the public. Props to you guys that can mess with things for hours to figure them out, I would never have the patience to do it. Maybe I burned myself out back in the early days of computer geekery and now would rather just not deal with it.

AMD has HIP out. It is now down to us to implement it. Initially core25 was supposed to be hip. Due to issues with other stuff, core26 was born, which fixed some issues, but had other issues, again HIP integration to backseat. Core27 then was thought to have HIP. But it was pushed without hip to make sure that all the issues with core26 were fixed on 27. It seems we have majority of things fixed now, so I hope core28 or revision of core27 will have hip, as excuses and reasoning is running out now

AMD will listen if we, as in FAH, have something to show and tell them that things can be improved, etc. I cannot tell them anything right now because we have nothing to work on yet.

Well I hope it performs as well as expected, for the sake of all the AMD users.

Core 30 HIP. By Christmas. You heard it here first. *












































+ Standard disclaimer. WHICH Christmas is unknown. Core 30 is only a guess.