AMD's R9 Fury On Open-Source: Prepare for Disappointment, For Now

An anonymous reader writes: With Linux 4.3 AMD is adding the initial open-source driver for the R9 Fury graphics cards. Unfortunate for Linux gamers, the R9 Fury isn't yet in good shape on the open-source driver and it's not good with the Catalyst Linux driver either as previously discussed. With the initial code going into Linux 4.3, the $550 R9 Fury runs slower than graphics cards like the lower-cost and older R7 370 and HD 7950 GPUs, since AMD's open-source developers haven't yet found the time to implement power management / re-clocking support. The R9 Fury also only advertises OpenGL 3.0 support while the hardware is GL4.5-capable and the other open-source AMD GCN driver ships OpenGL 4.1. It doesn't look like AMD has any near-term R9 Fury Linux fix for either driver, but at least their older hardware is performing well with the open-source code.

Comments

What kind of post is this?

[Ecuador]

As far as I can understand, AMD has released the specs for the new GPUs, which is what many Linux / Open Source advocates care about, right? Sure they haven't yet added the support for the new cards on their own, but other people could do it if they are in a hurry, right? Then you have Nvidia not releasing any specs for open drivers, but adding support (more quickly perhaps?) on their proprietary binary drivers. This upsets (hopefully a different set of) Linux users.
And in the end, this is just about games (because I never had a problem with the Linux desktop in general, even over multiple displays with mostly AMD cards), right? Well, guess what, Linux is not a good gaming platform. It is great in many things, why should it also be good for games? Why would Nvidia and AMD spend significant resources so that very few people (compared to the total market) can play games in a specific platform that, let's face it, is not gamer-oriented? Well, they don't, so don't complain. Sorry for the rant-ish post, perhaps I would post different 15 years ago when I was still into computer games ;)

Re:What kind of post is this?

[Kjella]

As far as I can understand, AMD has released the specs for the new GPUs, which is what many Linux / Open Source advocates care about, right? Sure they haven't yet added the support for the new cards on their own, but other people could do it if they are in a hurry, right?

Not for Fury, not yet. The shader instruction set architecture (ISA) is up to date, since the shaders are still GCN 1.2 (also known as 3rd gen GCN, IPv8) so you can write shader code for it. But the code to initialize and manage the card changes from generation to generation, if I recall correctly at launch it didn't work at all. Those parts are typically written by AMD, with the code typically preceding and more or less being the initial documentation. From what I've understood it's because AMD's legal department finds it easier to approve low level code than the high level documentation you'd need to write your own from scratch, since the former is less likely to reveal any trade secrets or plans for the future and the implementation is de facto proof the information is sufficient. So until AMD drops some code to dynamically manage clocks on Fury, there's not documentation either.

Once upon a time AMD had a plan to reach launch parity with Windows, since the work doesn't get any significant less by waiting - in fact it would be easier if they could ask the closed source people towards the end of their implementation while it's still fresh before they start moving on to the next generation of cards, but with AMD's financial troubles that's not going to happen. If the open source driver wasn't important for the embedded/semi-custom market, one of the niches AMD is betting on I don't think any of them would have a job today. In fact, I'm not sure how long anybody at AMD has a job as they now have a stockholder's deficit meaning they owe more than their assets. Technically that's not a problem until they can't pay their bills, but I'm not sure how much longer they can get financing since they're now losing the bank's money.

The whole posting is disengenuous.

[tlambert]

The whole posting is disengenuous.

"the R9 Fury isn't yet in good shape on the open-source driver"

The card won't be changing to fix this; the driver will have to change to accommodate the card; therefore it is more correct to say "The Open Source driver is not yet in good shape on the R9 Fury". In other words, it's not the hardware's fault that the driver doesn't support it yet.

"AMD's open-source developers haven't yet found the time to implement power management / re-clocking support"

The power management model in Linux is Linux's responsibility, not AMD's. The authors of the Open Source driver are accountable *only* for writing callbacks for the device power management component, and populating the structure. It's my understanding that Linux lacks a uniform model for use by all graphics drivers, in this regard. his is a Linux issue, not an AMD issue.

Also:

In general, in a hardware world, you either NDA people, or the Open Source is going to lag the closed source, period. This is because openly manipulating code related to an unreleased hardware product in a publicly accessible source repository, instead of a privately held repository, is tantamount to preannouncing your hardware to competitors. You might as well have the CEO call a press conference, and then shoot themselves in the head in public.

Open Source projects have a secondary problem in that, even if the driver source was developed entirely by engineers within AMD, and released the same day as the hardware was made available, the Open Source projects aren't going to be happy just integrating the code as is. They will insist on peeing on it to make it smell like themselves, just as cats do with new furniture, and this will take time. You can either have closed source, or you can have it integrated later than the release date, but you can not have both.

Re: Talk about unfair.

[WarJolt]

I hate to break it to you, but most open source drivers are written by corporations.

Re: roflmao

[WarJolt]

Ever since 802.11ac came out I feel like we've lost about a decade of progress when it comes to hardware support.

To be fair though graphics drivers have always sucked.

Re:All 5 Linux gamers outraged!

[epyT-R]

Buying a gpu and expecting quality drivers for supported OSs (open or closed) is hardly 'freetard entitlement'.

so?

[electrosoccertux]

I just use glxgears to benchmark. glxgears is a great benchmark. I recommend glxgears if you want to benchmark your liinicks. thanks.

Exactly the same kind.

[tlambert]

Re the last paragraph. That is not entirely true, as Intel appear to be able to integrate new chipsets during the time they are released (but only to next tier manufacture) before the public can buy anything using it.

What kind of "peeing on it" has been done to the Intel drivers to get them integrated ?

Exactly the same kind. It's possible to do for anything, it just takes time.

The reason Intel is able to do this ahead of general release, when other vendors aren't, is that it does not lose them a competitive advantage.

First, there is no issue of another manufacturer producing "pin and register compatible devices", and undercutting Intel, because Intel's graphics are integrated into the CPU; you'd have to build an entire Intel compatible CPU as well, and you'd have to do it competitively in terms of price point.

Second, no one really wants to emulate Intel Integrated Graphics in silicon, since there's really no advantage to doing so, since the chips have inferior performance relative to the competition.

So there's really nothing lost by Intel pre-announcing all of the information needed to make a driver, or even publishing source code for the driver, since doing so will sell more Intel chips, not less. For other GPU vendors, this is simply not the case, and there's no economic value in such pre-disclosure.

Re:Not That Bad

[apharmdq]

Whoops, this was me, somehow wasn't logged in.