Folding Forum

kuzko wrote:Hello chaps, If there is a version of the F@H client or benchmark that needs testing for the 5700XT, I have this GPU so just let me know and I'll try to run it.

It will be some time, from the developers the information is support for the new AMD architecture that the 5700 XT uses will be in the next point release of OpenMM. Then a core will need to be created using that release, and FAHBench also updated to use that core.

Thanks for the offer. It looks like the fix will not be back-ported to FAHCore_21 -- though there's a (remote) possiblity that that might change. Work on the upcoming FAHCore_22 is progressing, though rather slowly.

So far, FAHCore_21 is doing a good job on the science side of FAH except for this single limitation. (I have no information about what else Core22 might bring to the science.)

I think we should change the thread title since we have identified that the fix is required on FAH/OpenMM side ... the drivers are fine ...

toTOW wrote:I think we should change the thread title since we have identified that the fix is required on FAH/OpenMM side ... the drivers are fine ...

That's still a good idea, who can change the Thread Title?



Any other news about Core_22 with support for the 5700XT card?

Sven wrote:Any other news about Core_22 with support for the 5700XT card?

No. Basically it stands where it was a couple months ago, in beta. Though the project that was assigned to Core 22 is currently inactive while the researcher goes through the error reports in the returned WU's to see where problems existed.

With the information that they also now have an entire new generation of AMD GPU's that will require using a later revision of OpenMM, don't know where that leaves development of the core for release.

I need specific information about (A) recent changes to the AMD drivers and (B) the status of OpenMM changes required to be compatible with that update. Once that information is understood, I can configure a new GPU Species that can be restricted to FAHCores what incorporate that version of OpenMM. Does anybody have specific information about those two changes?

Next year new GPU generation by Nvidia and AMD are announced. So maybe they delay a new core until tested with these new GPU generations too next year.

bruce wrote:I need specific information about (A) recent changes to the AMD drivers and (B) the status of OpenMM changes required to be compatible with that update. Once that information is understood, I can configure a new GPU Species that can be restricted to FAHCores what incorporate that version of OpenMM. Does anybody have specific information about those two changes?

(A) The combination of GPU, drivers, and FahCore_21 has been broken since launch in July. Navi on the consumer side only includes the 5700/5700 XT. AMD is expected to launch a mainstream 5500 this quarter.

(B) I posted a link to the fix a few pages back. It shipped in the 7.4.0 release of OpenMM. FAH would need to back-port the fix to an older version of OpenMM used in FahCore_22 (if possible) or build a new FahCore_22 using OpenMM 7.4.0 as a base.

According to my information, the only device code0x731f (lspci) that I've found for the navi 10 [Radeon RX 5700 XT] is 0x731f. The Juniper devices 0x68b* apparently work fine, but they are 5700 without the XT and they work fine.

Does anybody have better information?

The Juniper cards are HD 5700 series from years ago and have no relation to the RX 5700 cards that use the Navi architecture. So far two have been released, the RX 5700 and the RX 5700 XT. AMD is supposed to be releasing a RX 5500 before the endow of the year.

However the Juniper based cards do not support DP, so are no longer usable for folding.

In preparation of future testing of a fixed core for Navi, I'm creating a new GPU species (6) for Navi and future RDNA GPUs.

So we should have species 4 for GPU that can run OpenCL core but doesn't support DP (I think there's no work for them, but they're still theoretically compatibles with FAH) and species 5 for GPUs with DP.

The plan is to set core 21 projects to assign to "species = 5" so that Navi won't get them. The new projects that will be created with the fixed core should assign to "species >= 5". This should translate into the following constraints.

Core 21 mixed precision : Core 22 : Note : species definition for NV is different so the above doesn't apply.

I'm ready to test my 5700XT with the new core22, when released.

Is that okay, as I'm not a member of the BETA-Team, but would use the beta flag on my 5700XT PC?

Simply agree to the standard policies of the beta team and ask to join.
viewtopic.php?f=2&t=10364

We'd appreciate another beta tester.

Yes, donors with AMD (and more specifically Navi) GPUs are welcome to join the beta team, we don't have a lot available ...

I also have a 5700XT running on my PC and would like to help in the beta testing. For this reason I registered to this forum (which I have been following as a reader for some time).
Can you please direct me were can I ask to join by agreeing to the policy?
Do I just reply to the link above? (viewtopic.php?f=2&t=10364)