Gotta go *fast* - Rewritten OpenGL drivers make AMD’s GPUs “up to 72%” faster in some pro apps

[Lightwreather]

 

Summary

The majority of graphics driver development effort these days, whether from Nvidia, Intel, or AMD, is concentrated on new APIs like DirectX 12 or Vulkan, increasingly powerful upscaling technologies, and particular upgrades for new game releases. However, AMD has been focused on an old trouble area for their graphics drivers this year: OpenGL performance.

Over the summer, AMD released a rewritten OpenGL driver that it said would boost the performance of Minecraft by up to 79 percent (independent testing also found gains in other OpenGL games and benchmarks, though not always to the same degree). Now those same optimizations are coming to AMD's officially validated GPU drivers for its Radeon Pro-series workstation cards, providing big boosts to professional apps like Solidworks and Autodesk Maya.

AMD claims that its new drivers can enhance Solidworks rendering performance by up to 52 percent at 4K and 28 percent at 1080p resolutions when used with a Radeon Pro W6800 workstation GPU. Autodesk Maya performance improves by 34% at 4K and 72% at the default resolution. The size of the improvements varies depending on the app and the GPU, but AMD's testing shows significant, consistent improvements across the board on the Radeon Pro W6800, W6600, and W6400 GPUs, which AMD claims will help those GPUs outperform analogous Nvidia workstation GPUs like the RTX A5000 and A2000, as well as the Nvidia T600.

Quotes

Quote

"The AMD Software: PRO Edition 22.Q3 driver has been tested and approved by Dell, HP, and Lenovo for stability and is available through their driver downloads," the company wrote in its blog post. "AMD continues to work with software developers to certify the latest drivers."

Quote

The release of AMD Software: PRO Edition 22.Q3 on September 29, 2022, is no exception as it brings our most significant performance advancements to date in all OpenGL applications and many of your other favorite creating, designing, modeling, and CAD software applications. The latest improvements are edging us towards and beyond the competition, such as Autodesk Maya, where we see improvements up to 41% greater on the AMD Radeon™ PRO W6800 GPU versus an NVIDIA RTX A5000 GPU using the Quadro Optimal Driver for Enterprise (ODE) 516.94 (2).

What do these numbers mean? These numbers provide a way of comparing hardware and driver performance on an even playing field. The better the score, the more streamlined experience the user will have with creating models and assemblies, working with detailed scenes, and a multitude of other interactive activities.

We continue working with software developers to advance future drivers to the next level of performance, reliability, and efficiency.

 

My thoughts

Well, that's interesting. Those Performance improvements seem quite nice, especially since these performance improvements are coming to older cards. Good on you AMD. Though it should be noted that these improvements apply solely to OpenGL and so doesn't carry over to macOS as well (since it was deprecated, unless you somehow compiled it on macOS, which I think someone has done). Would've been nice to see them on OpenCL and Vulkan but c'est la vie.

 

Sources

ArsTechnica

AMD - Blog, Adrenaline Release Notes

[SeriousDad69]

Good job AMD, it only took you 12+ years to make OpenGL usable in Windows with Radeon GPUs lol. Is this the AMD fine wine I'm always hearing about?

[LAwLz]

OpenGL performance has always been an issue with AMD so it's nice to see them finally fix it.

But I wonder if it's a bit too late. OpenGL performance was a pretty big deal back when Vulkan wasn't a thing, at least on non-Windows OSes. OpenGL performance has gotten less and less important so while it is nice that they finally fixed it, I feel like in most cases the issue has fixed itself.

Except of course some niche applications like some of the ones highlighted in the article. For the few people who need those applications this is great (although they probably use Nvidia).

 

Good that they fixed it, but this would have been way bigger news if it had been fixed 5 years ago.

 

 

30 minutes ago, SeriousDad69 said:

Good job AMD, it only took you 12+ years to make OpenGL usable in Windows with Radeon GPUs lol. Is this the AMD fine wine I'm always hearing about?

Well it did age like wine.

It's just that it was a type of wine that tasted terribly for the first 10+ years and then it suddenly started tasting decently.

[SeriousDad69]

On 10/4/2022 at 7:24 AM, LAwLz said:

OpenGL performance has always been an issue with AMD so it's nice to see them finally fix it.

But I wonder if it's a bit too late. OpenGL performance was a pretty big deal back when Vulkan wasn't a thing, at least on non-Windows OSes. OpenGL performance has gotten less and less important so while it is nice that they finally fixed it, I feel like in most cases the issue has fixed itself.

Except of course some niche applications like some of the ones highlighted in the article. For the few people who need those applications this is great (although they probably use Nvidia).

 

Good that they fixed it, but this would have been way bigger news if it had been fixed 5 years ago.

 

 

Well it did age like wine.

It's just that it was a type of wine that tasted terribly for the first 10+ years and then it suddenly started tasting decently.

The real knee slapper imo is AMD prematurely dumped support for all GPUs older than Polaris, so the vast majority of the GPUs that released with these horrible performance problems in OpenGL, DX9, DX10, and DX11 wont see any benefit lol. And people call me a fanboy because I'm willing to spend 20% extra for an Nvidia card lol

[PocketNerd]

On 10/7/2022 at 7:16 PM, SeriousDad69 said:

The real knee slapper imo is AMD prematurely dumped support for all GPUs older than Polaris, so the vast majority of the GPUs that released with these horrible performance problems in OpenGL, DX9, DX10, and DX11 wont see any benefit lol. And people call me a fanboy because I'm willing to spend 20% extra for an Nvidia card lol

That would be older than the RX 400, which honestly are probably not good enough even if they got optimized.

[SeriousDad69]

2 hours ago, PocketNerd said:

That would be older than the RX 400, which honestly are probably not good enough even if they got optimized.

R9 390 is somewhere between a 1650 or 1060, but with 8GB of VRAM. A perfectly usable card for most gamers. And it's not even that old, it came out in 2015. There's other cards that are a lot faster(and newer), like the Fury, Fury X, Fury Nano, etc.

Up until the last ~2 years, AMD's drivers had been so bad for so long that you cant accurately gauge how fast the hardware in these older cards actually is, but I'm guessing these cards would probably be 10-15% faster with proper drivers, which would easily elevate them past their Nvidia equivalents. But yeah AMD dumping support for perfectly good cards while Nvidia is still supporting Maxwell just goes to show you get what you pay for. AMD is like, you save $50 but get ~4-6 less years of driver support. I'm going to need to see the RX 400 series get driver updates at least until 2026 before I trust them again.

[DuckDodgers]

Proper OpenGL ICD support has been a curse for most of the graphics vendors in the last 25 years. Nvidia really did their homework from the first shot and it is the reason their hardware has been a default choice for many users and developers dealing with this API.