AMD's Initial Graphics Updates For Linux 5.3 Include PowerPlay Improvements, HMM Usage

[phoronix]

Phoronix: AMD's Initial Graphics Updates For Linux 5.3 Include PowerPlay Improvements, HMM Usage

While the Linux 5.2 kernel won't see its debut until July followed by the opening of the Linux 5.3 kernel cycle, the AMD developers sent in today their initial set of staged changes to DRM-Next for queuing their preliminary AMDGPU/AMDKFD driver changes they want to get into this next kernel cycle. There are some notable additions but what we are expecting/hoping for and haven't seen yet is the Navi support...

http://www.phoronix.com/scan.php?page=news_item&px=AMDGPU-Linux-5.3-First

[humbug]

Though with the RX 5700 series shipping in July, it will mean launch-day customers either need to be building the drivers from source (seeing as the code didn't make it into Linux 5.2 or Mesa 19.1 that will be stable versions then) or be riding the Radeon Software for Linux packaged binary driver stack on a supported enterprise Linux distribution.

Bad look.

I really thought AMD's Linux driver team will achieve a smooth launch this time. Navi has been cooking for a long time.

[Termy]

Let's hope the best - if navi makes it into 5.3 and mesa 19.2 it is at least close enough to launch to not be too bad - if it misses 19.2, it will be an annoying delay :/

[ThoreauHD]

I think Navi is GCN, and that AMD is great at compute, but can't figure out high perf low powwr gaming rasterization to save their ass.

June 11 is E3. Supposedly the full announcement with launch on 7/7.

That either means the driver/card isn't ready, or they suddenly don't care about the Linux market. They aren't playing coy because they are competing against the R7, not Nvidia. The only thing Nvidia affects with this launch is price, and that can be changed on the fly. Driver support, not so much.

These name changes and shady moves smell like a delay tactic in a parade of BS. I hope I'm wrong, but I get the feeling AMD will simply be focusing on datacenter compute gpu's if Navi gaming goes sideways.

[Peter Fodrek]

Bad look.

I really thought AMD's Linux driver team will achieve a smooth launch this time. Navi has been cooking for a long time.

Maybe no. Maybe it will work as Vega or Polaris in compatibility mode. It ssems tahat NGG fast path is not used by Vega, yet. There is possibility to enable Navi in same "obsolete compatibilty mode" at launch same situation as Vega was or is, yet

[Veto]

It is also a possibility the Navi changes largely already *are* in the code. But until the PCI IDs are posted we just cant see the relevant code paths. It should be possible to merge additional PCI IDs relatively late and possibly as a point release of the kernel/Mesa.

Just hoping :-) Navi looks like a possible buy for me.

[jrch2k8]

I think Navi is GCN, and that AMD is great at compute, but can't figure out high perf low powwr gaming rasterization to save their ass.

June 11 is E3. Supposedly the full announcement with launch on 7/7.

That either means the driver/card isn't ready, or they suddenly don't care about the Linux market. They aren't playing coy because they are competing against the R7, not Nvidia. The only thing Nvidia affects with this launch is price, and that can be changed on the fly. Driver support, not so much.

These name changes and shady moves smell like a delay tactic in a parade of BS. I hope I'm wrong, but I get the feeling AMD will simply be focusing on datacenter compute gpu's if Navi gaming goes sideways.

I do suspect as you do that RDNA is very very close to GCN from a driver perspective, i suspect also RDNA is just an electronic level optimization of GCN as more ROPS, positioning changes of the shader units, etc. to optimize bandwidth and performance of each compute block instead of a full redesign.

If thing go as i suspect NAVI support won't require much work on Linux outside some golden registers and some new firmware and maybe few changes for new features.

Of course after E3 we should have a more concise idea of how different is from GCN

[skeevy420]

Bad look.

I really thought AMD's Linux driver team will achieve a smooth launch this time. Navi has been cooking for a long time.

That's why I suggested that AMD needs to host their own repositories containing the all the open AMDGPU source code for the same distributions they support with AMDGPU-Pro.

This is the exact opposite of the Catalyst days where we needed custom AMD repos that contained old software; now we need them with bleeding-edge software.

Buy an AMD GPU, add the AMD Repository, update+install a couple of packages, reboot, and done...if it were only that simple...but nope, they basically have us do the debianxfce method of adding unofficial PPAs, compiling custom kernels most people probably don't even know exist, and other things that the average end-user shouldn't have to do to get their $400 GPU to work correctly.

[schmidtbag]

As long as Navi uses radeonsi, I'm sure it's safe to say that it is in fact very similar to GCN. I still don't fully understand what it is that makes Navi so different, but, I don't think being similar to GCN is a bad thing. GCN is actually a pretty fast and efficient architecture, but, only under very specific workloads (which as we all know, gaming is not one of them). To me, the main thing that gives Nvidia such a lead is how their performance-per-watt is consistent regardless of what you're running. To me, all AMD needs to do is to get better performance-per-watt results than Nvidia in gaming. To me, they don't have to have something faster than a 2070, they just need it to be more efficient.

[Awesomeness]

they basically have us do the debianxfce method of adding unofficial PPAs, compiling custom kernels most people probably don't even know exist, and other things that the average end-user shouldn't have to do to get their $400 GPU to work correctly.

No, end users shouldn't do that. Distributors should. They have the QA pipeline in place to test if the whole thing causes issues with other tweaks and patches they have.