EDIT - pro drivers released by AMD on 2020-09-29 work under Ubuntu.

I'm writing this post with the dual goals of warning other potential buyers and perhaps receiving some helpful feedback.

In short, the Radeon RX 5700 XT is absolutely useless for Go analysis as of November 2019.

In Windows 10:

• Leela Zero will - if it even starts at all - crash within seconds due to invalid mathematical results
• KataGo will - if it even starts at all - crash within seconds due to invalid mathematical results
• Lizzie app - numerous graphical glitches in menus whether using CPU or GPU version

In Ubuntu 18.x, 19.x:

• clinfo doesn't detect the card
• cannot build a functioning GPU Leela Zero from source
• numerous system crashes due to GPU driver instability when playing games

I built this PC for Adobe-based photo and video editing, gaming, and Go. It performs the first two tasks admirably well.

Sadly, it's a miserable failure for the third.

On a related note, these cards also send bad data to SETI@Home when crunching work units. Again, this is an OpenCL support issue.

I'm hopeful that AMD gets this straightened out, but with these cards on the market for 3 months, racking up complaints and crash reports for 3 months, and no improvement yet... I'm not optimistic.

Caveat emptor.

Thanks a lot for the valuable info.

As far as I know, the RX 5700 XT are good cards but there has been a lot of drivers issues. I've heard less complain about it lately. Are your drivers up to date?

@explo: Please explain how a RX 5700 XT is a "good card" for go related software. I would be really interested in positive experiences with a recent AMD card in this use case. I have not found any yet.

My drivers are up to date on both platforms.

If you know someone who can run Leela Zero on the 5700 XT, please collect as many implementation details as possible.

SETI@Home is basically banning the 5700 XT right now because the OpenCL implementation is so poor (bug-ridden).

I've never said they are good cards for go related softwares. Drew's experience is the only one I know about. Saying the RX 5700 XT are not good cards because they can't handle go softwares is absurd in my opinion, because go softwares are such a niche use. Judging those cards for their video games and video editing performances is a lot fairer.

explo, you must understand that we're in the ComputerGo section of 19x19. Any GPU discussion by default is regarding Go software performance.

If I was only answering your post about go softwares I would have said it's functionnal. Based on your post, the card is not bad, it's straight up not working.

Some very minor good news: AMD's most recent Radeon driver release notes (link) includes the following under Known Issues:

SETI@Home may be provided incorrect results from Radeon RX 5700 series graphics products.

SETI@Home suffers from the same problem all the Go software does - bogus results/crashes from OpenCL.

Acknowledging you have a problem is the first step in solving it... right?

It's a general bug with OpenCL in RDNA.

Can someone with a new AMD card submit a bug report about OpenCL to AMD directly? I would have done this but I don't have one to provide relevant details.

Update for those of you out there like me with this card: today I heard from an AMD development rep that the fix has been ported to the mainline/release build and it is expected to be released soon.

Almost one year after release, there's light at the end of the tunnel...

@Drew: what was your reason for buying an AMD Radeon card? I really admire the ideas of open frameworks like OpenCL (and OpenGL and the like), but isn't it really hard to beat the performance of NVIDIA GPUs with tensor cores in combination with NVIDIA's proprietary CUDA platform? At least, that's my experience for training neural networks. I recently bought an NVIDIA RTX 2060 for a friend to run Go analyses and it was plug 'n play.

I'm not trying to rub in your disappointing experience. It's just that I bought the NVIDIA card practically without thinking, so I'm curious about your reasoning.

It's a fair question. The computer in question was built to maximize gaming performance value, and in that respect the 5700XT was an excellent choice. A performance-equivalent nVidia card would have been almost 2x the cost. It just shipped with non-functional OpenCL support which no one in the gaming world cares about because it's used by a vanishingly small number of projects these days.

Unfortunately, one of those projects is Go AI software.

TL;DR: money.

Thanks for your reply. That makes total sense.

Good to hear that there's a patch coming up to solve your issue!

Well im having the same problem as you. The release didnt happen yet, right?

Fellow Go fan and chagrined new 5700 XT owner here. I knew when I bought it that AMD wasn't exactly the leader in deep-learning support, but I didn't realize that this specific model was so broken. I am joining the forum and thread to keep posted about any developments around this so I can hopefully play KataGo and Leela Zero one of these days. Thanks for all updates.

I just found this thread and I’m now nervous. I bought a new iMac with Radeon Pro 5700 xt. It’ll be delivered in a couple of weeks.

One of the reasons was I figured in the likely event that Apple fully dropped support for Open CL in its next OS (already deprecated), I could still use boot camp to run katago in Windows.

Is there any news on this front lately? Have recent updates worked? Is it just a windows problem?

Well, at least Apple has a generous return policy if it fails.

Thanks.

_________________
- Brady
Want to see videos of low-dan mistakes and what to learn from them? Brady's Blunders

Hi, let me just say that I actually managed to get Katago running on Win 10 with 5700 XT seemingly without a hitch or any special steps. Runs fast, plays strong. (Note that I am not doing any neural-net training so far, just using the existing weights.)

AMD has patched the bug on Windows with latest drivers. Allegedly also for Ubuntu 20.X, but I haven't been able to install the amdgpu-pro driver package successfully as of yet.