AMDGPU vs. Radeon DRM On Linux 4.13 For AMD GCN 1.0/1.1 GPUs

[Michael]

Well clearly the Radeon driver doesn't work right. I'm a lot more interested in seeing AMD solve the issue with Michael's card. I've seen them try on these forums, talked about firmware and such. I feel if they were able to solve it, they already would have. It actually makes me significantly less confident in AMD's driver team that they've not been able to solve it with Michael's card. I'm not willing to accept the idea that Michael's card is messed up until AMD figures out the reason. They seem to be sweeping issues under the carpet rather than honestly trying to fix them. Every time Michael does a new round of benchmarks showing his 290 to continue being problematic, I feel more and more disrespected as a paying customer of AMD's.

Why would they admit that an "experimental" driver needs to be made the default? Gee, I dunno, maybe because it WORKS? Geez.

The R9 290 did work fine when I bought it and then began regressing around Linux ~4.6 (if my memory serves me about versions, but initially it worked fine on both open driver and fglrx).

[Holograph]

The R9 290 did work fine when I bought it and then began regressing around Linux ~4.6 (if my memory serves me about versions, but initially it worked fine on both open driver and fglrx).

What are your current thoughts about why the card has issues? Have you gotten any new clues recently? I don't see a whole lot of detail in the article, so does that mean you've got no new ideas about the reason? Has AMD continued to work with you on trying to fix it recently, or has their contact on the issue fallen off? (too slow on my edit there with additional questions I guess)

[Michael]

What are your current thoughts about why the card has issues? Have you gotten any new clues recently?

Likely some BIOS issue, but then again, it started working correctly on Linux ~4.11 but then regressed again either on 4.12 or now with 4.13. AMDGPU still seems to always work fine.

[schmidtbag]

Well clearly the Radeon driver doesn't work right. I'm a lot more interested in seeing AMD solve the issue with Michael's card. I've seen them try on these forums, talked about firmware and such. I feel if they were able to solve it, they already would have. It actually makes me significantly less confident in AMD's driver team that they've not been able to solve it with Michael's card. I'm not willing to accept the idea that Michael's card is messed up until AMD figures out the reason. They seem to be sweeping issues under the carpet rather than honestly trying to fix them. Every time Michael does a new round of benchmarks showing his 290 to continue being problematic, I feel more and more disrespected as a paying customer of AMD's.

Why would they admit that an "experimental" driver needs to be made the default? Gee, I dunno, maybe because it WORKS? Geez.

Except, the Radeon driver usually does work right. You're acting like the experience you and Michael have overshadows all success stories and therefore, simply because you 2 have a problem, everyone does. But here's the thing: you're the minority. It's difficult to diagnose a problem when so few people have it, and when they can't provide any useful information as to what is wrong with them. I'm not saying Michael himself is useless in this, because this is a problem beyond his knowledge. So unless someone physically gives the driver devs the faulty model, what do you expect them to do about a problem that they can't experiment with? So since they're forced to do this remotely with little to no low-level information, do you really think its a high priority of them to work on a 4-year-old GPU where a relatively small sample group is affected? It wouldn't surprise me if r600 is a higher priority than that.

Meanwhile, you are completely neglecting the huge amount of benefits these drivers have received over the past couple years. More and more disrespected, my ass. You are looking at this way too selfishly, you [should have] had an opportunity to RMA your GPU. I understand your frustration, but when you buy a new car and something is defective the day of you buying it and the mechanics don't know what's wrong, are you going to keep it? In such a scenario, you're already off to a bad and unusual start, so imagine how many more problems you may encounter in the future.

[Holograph]

I've been somewhat negative in this thread, yes, but I'm not always equally negative when I talk about AMD. And I've repeatedly talked poorly of Nvidia for their crappy moves like trying to block virtualization+passthru in their Windows drivers which is actually what caused me to buy an AMD card anyway. At the time I got it to run in Linux but to pass to a Windows VM. Currently I'm not running any Windows VMs, just using WINE on my desktop and my laptop still runs Windows 7. But I didn't really natively game with this card in Linux at first. Fortunately a lot more games have been getting native releases and WINE has been improving, too, so I haven't needed the VM solution (or the dual boot solution) so much anymore. But I'm exposed to any flaws in the performance of the native Linux gaming now.

I will agree with you on certain points here. For starters, while I've seen some AMD guys like bridgman try to help Michael here, I'm sure I have not seen most of the discussion that has gone on between Michael and AMD. Maybe I don't have the whole story on that - in fact, I almost certainly do not. However it's been quite a while now. How long do we need to give them to come up with something, or do you think it's reasonable to permanently give up on cards?

Michael's been having this problem for a long time with that card so the age of the card doesn't really make a valid argument to me. And considering that it's perfectly capable of working properly with AMDGPU, it tells me the card is capable of working properly. Even if something is messed up in his VBIOS... why hasn't AMD sent over some sort of program to dump his VBIOS so they could check that out?

Also, so far, the 290x is still pretty high-end for an AMD card. I mean, no, it's not a Fury-X, but it's still pretty good. It's not really fair to compare the 290x to say an RX580 because Polaris is all low-to-mid range, and the 290x was fairly high-end for the time.

I don't even care about my specific 290x all that much. What worries me is that AMD doesn't seem interested in solving issues. Michael's 290 is still malfunctioning and AMD are still refusing to talk about the issues people are having with Ryzen and GCC with ASLR and SMT enabled. It just makes me feel like buying AMD is a really bad idea because there's not going to be good support no matter what. And then when you get upset about that and try to discuss it, other consumers get mad at you for speaking of your own experiences. It's kind of crazy.

I actually want AMD to disprove me and fix these issues. I don't want to be right about any of this. But it does matter how a company reacts to problems its customers have and it is absolutely valid for customers to react to that. You can have your opinion on whether any of these problems matter (or are widespread) but it is a fact of life that customers will judge a company based on how they handle issues, and that's not going to change. I won't say you're wrong in the things you've told me but I also don't think I'm wrong to care about this.

[franglais125]

Here is a nice reminder of why AMD/Intel graphics are the way to go, even with the pain. NVIDIA rep: "We currently have no plans to support Xwayland." [1]

It seems no number of middle fingers will do.

[1] https://devtalk.nvidia.com/default/t...88874/#5188874

[schmidtbag]

Currently I'm not running any Windows VMs, just using WINE on my desktop and my laptop still runs Windows 7.

My only suggestion is to try comparing and contrasting your GPU's BIOS to another model known to work, but you're probably going to need Windows to do so. If you like, I can send you mine (bear in mind my hardware is weaker). This article might be a good place to start:
http://www.overclock.net/t/1561372/h...295x2-390-390x
If you're interested in doing this, while you're at it, I'd recommend to try dropping the voltage by 0.05v or if you can, 0.1v. This can help reduce the power consumption by as much as 50W with no performance loss.

How long do we need to give them to come up with something, or do you think it's reasonable to permanently give up on cards?

For this specific issue, I would suggest to permanently give up and move onto AMDGPU. Aside from my point about how old and niche this problem is, AMDGPU works fine and will likely leave experimental status for these GPUs soon.

Michael's been having this problem for a long time with that card so the age of the card doesn't really make a valid argument to me. And considering that it's perfectly capable of working properly with AMDGPU, it tells me the card is capable of working properly. Even if something is messed up in his VBIOS... why hasn't AMD sent over some sort of program to dump his VBIOS so they could check that out?

My point is the older something gets, the more it gets bumped back in the "todo list" as new hardware comes in. The fact this is a device-specific issue (rather than a GPU or architecture issue) would make it an even lower priority.
As to why nobody has requested his VBIOS, I don't know. I bet that's where the issue lies. But like I said before, you probably need Windows tools to work this low-level. I'm not aware of any Hawaii-compatible Linux VBIOS tools.

Also, so far, the 290x is still pretty high-end for an AMD card. I mean, no, it's not a Fury-X, but it's still pretty good. It's not really fair to compare the 290x to say an RX580 because Polaris is all low-to-mid range, and the 290x was fairly high-end for the time.

I agree. I intend to keep my 290 until 4K gaming becomes affordable. Mine has only improved with driver releases.

What worries me is that AMD doesn't seem interested in solving issues. Michael's 290 is still malfunctioning and AMD are still refusing to talk about the issues people are having with Ryzen and GCC with ASLR and SMT enabled. It just makes me feel like buying AMD is a really bad idea because there's not going to be good support no matter what. And then when you get upset about that and try to discuss it, other consumers get mad at you for speaking of your own experiences. It's kind of crazy.

I can't stress this enough: it all comes down to priorities, yours and theirs. At least you have an alternative that will give you a good experience. Consider the DC/DAL issue - that doesn't affect either of us, but there is new hardware that depends on it. Personally, I think buying brand new hardware that is missing a significant feature is more important than an inconvenient anomaly.

To reiterate - I understand your frustration. It is warranted. But the fact of the matter is, you should've known what you were getting into, and therefore, I find your motivation and pessimistic outlook unwarranted. There are sacrifices and sometimes risks involved when buying into AMD on Linux, because the company is relatively small and they're still trying to catch up in both hardware and software. Very often it pays off, but when it doesn't, you kind of just have to accept that. They're cheaper than Intel and Nvidia for a reason, and you have to understand that Linux desktop users are not that profitable to them. So if you consider how low-priority your situation is, all of us are a low-priority in the grand scheme. That being said, yes, it sucks about performance losses and missing features, but a well-informed buyer should know what they're getting into, and therefore cannot rightfully complain.

[haagch]

It actually makes me significantly less confident in AMD's driver team that they've not been able to solve it with Michael's card. I'm not willing to accept the idea that Michael's card is messed up until AMD figures out the reason. They seem to be sweeping issues under the carpet rather than honestly trying to fix them.

I agree that you should have RMA'd it. I kept annoying the developers for some time with my RX 480 too because it had odd GPU hang problems. Turns out after an RMA and getting a new (but exact same model) one it has been performing exactly as expected ever since. It may be possible there is just something wrong with a couple of the GPUs and amdgpu happens to work around it by accident.

[dungeon]

The R9 290 did work fine when I bought it and then began regressing around Linux ~4.6 (if my memory serves me about versions, but initially it worked fine on both open driver and fglrx).

AFAIR Hawaii was not high priority, that didn't worked initially at all with open driver actually... basically AMD opensource devs claimed they don't care by saying it is not high priority

Started working like year later or was something like 10 months after card release, someone from Red Hat found a quirk

[Oxalin]

Two features not tested though, but important for some are VCE and UVD supports. I'm slowly but actively working on VCE 1 support for SI(new child at home). When completed, I'll move to UVD for SI. My two cents here.