Folding on Ryzen 9 5950X bug

[damric]

Hello

Is anyone out there folding on this? There's some sort of bug that if you fold on both CCDs it will eventually freeze up or crash. This has been documented and tested, but the only solution so far is to use something like Process Lasso to just fold on 8c/16t on one side and it can fold forever. This sucks because obviously we're missing out on half the compute power. It is uncertain whether this is a bug in the f@h client or some sort of problem with the hardware itself. I've personally tested at stock everything, JEDEC speeds, SMT on/off, various motherboards, PSUs, RAM, Windows 10 and 11, ect and eventually it will freeze or crash. Has anyone at Anandtech had this problem or found a workaround? I'd really like to have 14c/28t folding. Official folding forum wasn't very helpful since they aren't really hardware savvy.

Currently I'm rocking:

Ryzen 5950X
B550 PG Velocita (strong VRM)
4x8GB Ballistix
RX 6900XT
LEADEX V 1KW
Custom Water
Windows 11 Pro

 

[TennesseeTony]

Certainly an odd situation, thanks for getting the word out about it. I think most of our heavy hitters are only Folding with GPU though. I guess for now you may want to use the other half of the CPU for BOINC medical projects.

 

[StefanR5R]

FWIW, I had been folding on Zen 2 EPYC 7452, with one task spanning several CCDs and CCXs. But it has been a while; one or perhaps two years ago. And it was on Linux.

With an older FAHCore which they no longer use, I could even spread one task across both sockets of the dual-EPYC = 128 threads. Eventually they moved on to a different FAHCore which no longer scales to that many threads. I don't recall what the limit was, but it was way more than 16. Maybe 64.

I am not going to re-test because I am currently unable to upload results files to Folding@Home.

 

[crashtech]

I know this doesn't happen in Linux for me, so I would tend to think it's a Windows thing, or maybe the F@H Windows client. As Tony says, GPUs are so vastly better than CPUs at Folding that most of us don't bother unless it's a tight competition where every point counts.

 

[StefanR5R]

For the science, it's probably worthwhile to fold on CPUs, because their CPU work batches are for different projects of theirs than their GPU work batches. (Or at least it used to be that way; haven't checked up on that for a long while.) But they give ridiculously low credit to results from CPU work. — Or actually it's the other way around, the credit they give for results from modern GPUs is ridiculously high. — Hence, if one goes by credit (which should be a measure of what the contribution to the project is worth), folding on CPUs is almost worthless. :-(

 

[damric]

Good so it's probably something to do with the Microsoft Windows version of the client.

I honestly don't really care about the points. I've had some family members with cancers so I guess I just like to donate it because it's there.

I am really bad with software, especially Linux but if someone has the patience to teach this cave man I'm all ears. My Linux friends from my other forums are quite snobby and won't hand feed me the knowledge to get it up and running. They are always like google it, read a guide, or whatever, and it might as well be Sumerian to me. I'm great with hardware though.

 

[Skillz]

Join our discord, we will walk you though it. I am assuming you can at least install Linux on your own?

 

[damric]

Join our discord, we will walk you though it. I am assuming you can at least install Linux on your own?

Will do. I have installed Ubuntu before. Is there a preferred distro to run for just folding? I would put it on a separate drive to boot when I'm doing that.

 

[Skillz]

Most of us use Linux Mint 20.3, but all of us use some derivative of Debian. That's Debian, Ubuntu and Mint mostly.

 

[crashtech]

@damric , I'd say that coming from Windows, Mint is probably the friendliest. If you do decide to try Linux and are going to be Folding and/or doing other DC projects for the team, you can expect plenty of assistance from us!

 

[StefanR5R]

@damric, if you are using Windows mainly, but nevertheless want to give Linux a try, then switching back and forth between Windows and Linux on the same computer would be a pain, IMO. Though if you happen to have a spare computer which you'd like to dedicate to Folding@Home, then letting it run Linux most of the time should be feasible and beneficial: While I am not sure what it will do for CPU FAHCore stability, it is well known that GPU FAHCore performance under Linux is better than under Windows.

(From what I learned a few years ago but still might be true, the reason is that the Windows GPU driver stack has got extra features like the ability to update the driver live, and to pull driver crash dumps live. These features come at the cost of extra memory copying during normal operations. The Linux Nvidia driver stack does not have these features, and as a side effect the total bus bandwidth available to GPGPU applications on Linux is higher. FAHCore is one of the few GPGPU applications which desire quite bit of PCIe bandwidth.)

Instead of booting Linux natively, you could of course also run Linux in a VM, e.g. Windows Subsystem for Linux (WSL). That should be good for CPU folding. But GPU pass-through for GPU folding within a VM would require some extra steps, and I am not sure if this would let you enjoy the performance uplift which a native Linux installation would bring. (I never have tried GPU pass-through from Windows to a VM myself. Actually, my entire experience with Windows is rather thin, even more so since after Windows 7…)

 

[Skillz]

From what I have observed getting GPU to pass through the WSL is not easy. I know it's possible as I've heard of people doing it, but no one on our team was able to get it to work and they tried a good bit a while back before giving up.

 

[damric]

@damric, if you are using Windows mainly, but nevertheless want to give Linux a try, then switching back and forth between Windows and Linux on the same computer would be a pain, IMO. Though if you happen to have a spare computer which you'd like to dedicate to Folding@Home, then letting it run Linux most of the time should be feasible and beneficial: While I am not sure what it will do for CPU FAHCore stability, it is well known that GPU FAHCore performance under Linux is better than under Windows.

(From what I learned a few years ago but still might be true, the reason is that the Windows GPU driver stack has got extra features like the ability to update the driver live, and to pull driver crash dumps live. These features come at the cost of extra memory copying during normal operations. The Linux Nvidia driver stack does not have these features, and as a side effect the total bus bandwidth available to GPGPU applications on Linux is higher. FAHCore is one of the few GPGPU applications which desire quite bit of PCIe bandwidth.)

Instead of booting Linux natively, you could of course also run Linux in a VM, e.g. Windows Subsystem for Linux (WSL). That should be good for CPU folding. But GPU pass-through for GPU folding within a VM would require some extra steps, and I am not sure if this would let you enjoy the performance uplift which a native Linux installation would bring. (I never have tried GPU pass-through from Windows to a VM myself. Actually, my entire experience with Windows is rather thin, even more so since after Windows 7…)

I actually do have a separate PC that folds 24/7, and yes I would definitely change that one to Linux. The GPUs that fold on that one are pretty old, and I figured I might as well run them into the ground with folding, but they haven't died after 3 years straight of heavily overclocked folding

That one has:
Ryzen 5 5500 (recently replaced Ryzen 1600AF) folding on 4c/8t
X470 Taichi
Quadro M4000 OC'd to 1507/1750 via firmware main PCIe slot (x8)
Quadro M4000 OC'd to 1507/1750 via firmware 2nd PCIe slot (x8)
Quadro M2000 OC'd to 1507/1750 via firmware 3rd PCIe slot (the runt card gets the x4 slot)
Everything on that rig is under water with $100 worth of the cheapest parts from ebay

I've got more test benches that are constantly getting reconfigured between f@h and HWBOT (I'm very competitive overclocker).

But for my main PC, I do have an unused M.2 slot I could slap a dedicated NVMe drive for when this one folds. I have some spare 256GB drives so I think one of those would be perfect.