Radeon RX 6600/6700/6800 XT: RADV vs. PRO Vulkan Driver Performance

[linuxgeex]

If you check the Phoronix reviews, I think you'll see that Polaris had a successful launch as well ...
[snip]

Vega was problematic since that was where we switched from legacy display code to DC ... [snip]

You and I had some frank and perhaps uncomfortable discussions around driver quality after the Polaris launch. Polaris was stable at launch, but it took about 5 months before it was truly performant, even on Windows. Maybe some people see value in the performance of the product they bought increasing dramatically months after they bought it. Certainly that can be spun as a "feature" kinda like how GPU-reset was spun as a "feature" to resolve stability problems, lol.

Vega was only really a problem with HDMI. HDMI audio support lagged for a long time, mostly because Linus looked at the 30M SLOC push for DC and basically said no thanks to the large proportion of Windows code that didn't belong in the kernel, along with some "code quality" grumblings.

The community was naively unaware of the disaster that HDMI support was from a standards perspective thanks to DRM fears driving licensing paranoia well beyond the scope of DRM.

Many thanks to AMD for pushing to have many of those licensing issues cleared up. <3

[Daktyl198]

That's how far along the RTRT road we are...

Everything has to start somewhere. RTRT is almost certainly the direction we should be taking gaming. Not only does it make development of games easier and faster, the bigger take away is that lighting (I would argue) is THE most important aspect for modern games. Not even just realistic ones. Valheim is a great example of a game with very good traditional lighting and even with a pixely/cartoonish aesthetic it looks amazing.

Another great example is Minecraft using a RTRT shader like SEUS-PTGI. It both makes Minecraft look a million times better and more modern, and using his HRR (half-resolution rendering) shader I can achieve 70fps in heavily modded Minecraft on my 5700xt (AKA without any special RT cores).

While I don't expect AAA games to be able to fully utilize RTRT at 1440p or higher for a couple more generations, 1080p RTRT is pretty much already here.

[skeevy420]

Wouldn't that be a conflict of interest? If companies were paying Michael, I think that would give the impression of quid pro quo even if it was innocent.

IMHO, since Michael is open and honest about what he does and how he tests things there shouldn't be any conflict of interest. At most he might have to start labeling some articles as Sponsored Articles if AMD/NVIDIA/Intel sends him a GPU, some money, and requests certain tests on a certain setup. As long as the tests are open + fair then hardware donations and paid testing for funding shouldn't matter.

Frankly speaking, if Intel wants to give him their next CPU along with $2000 and a list of tests they'd like ran to see where it stacks up....as long as the testing and reporting stays fair and open, the Phoronix way, what difference does it really make? I suppose a community contract with Michael or for companies that want to pay Michael might be a necessity. A vow of integrity; an agreement that all results will be published even if they're not the expected outcome; that Michael is allowed to have a negative opinions if a product doesn't live up to marketing and hype. Something along those lines.

[wertigon]

Quake2, which is now very nearly *25* years old, with its environments and models that are literally 2 to 3 orders of magnitude simpler than modern assets

You are forgetting that to rasterize a million polygons you need 10x the power of 100 000 polygons. Meanwhile 10x polygons require 10% more power for a pure RT renderer. It is a log scale vs exponential scale.

[agd5f]

Period.
I mean, I don't check EVERY hardware website around here, but from the half a dozen I consult regularly and the ~20 I sometimes check when a particular hardware piece interests me, Phoronix has by far been the most extensive and fair information provider, both in quantity and quality.

A bit of cash without any strings attached of course would be far earned in my opinion and would benefit everyone.

(Edit: just to be clear, I think the same would be true for any hardware provider Michael surveys regularly, and possibly some Linux distribs too *cough Ubuntu / Red Hat cough*. Maybe that's the case and I'd be glad of it for Michael though )

Wouldn't that be a conflict of interest? If companies were paying Michael, I think that would give the impression of quid pro quo even if it was innocent.

[MadeUpName]

Well, that's interesting that they tracked down the issues in Michael's last tests, at least, even if it hasn't actually been fixed yet.

I do wonder what % of consumers actually run a "supported enterprise Linux distribution" though. Makes the driver pretty hard to recommend to anyone other than the corporate situations I assume the driver is targeted at. Nobody is reinstalling their OS just to use this driver.

In a corporation your ether running a way smaller card because people don't need the ooomph. Or you running a way bigger card because compared to the cost of having some one that needs the hardware sitting around waiting for tasks to complete is trivial compared to the cost of the card.

[MrCooper]

I do wonder what % of consumers actually run a "supported enterprise Linux distribution" though. Makes the driver pretty hard to recommend to anyone other than the corporate situations I assume the driver is targeted at. Nobody is reinstalling their OS just to use this driver.

Also, RHEL 8 is supported and defaults to Wayland.

[smitty3268]

All of the benchmarks were running off Ubuntu 20.04 LTS. Following some recent Radeon Software for Linux Vulkan driver tests on Ubuntu 21.04, I was informed by AMD there are currently some known performance issues when using their Vulkan driver in a Wayland environment. With time their Wayland issue should be addressed while for now they recommend users stick to running the packaged driver on the supported enterprise Linux distributions.

Well, that's interesting that they tracked down the issues in Michael's last tests, at least, even if it hasn't actually been fixed yet.

I do wonder what % of consumers actually run a "supported enterprise Linux distribution" though. Makes the driver pretty hard to recommend to anyone other than the corporate situations I assume the driver is targeted at. Nobody is reinstalling their OS just to use this driver.

[Citan]

AMD should support [email protected]

Period.
I mean, I don't check EVERY hardware website around here, but from the half a dozen I consult regularly and the ~20 I sometimes check when a particular hardware piece interests me, Phoronix has by far been the most extensive and fair information provider, both in quantity and quality.

A bit of cash without any strings attached of course would be far earned in my opinion and would benefit everyone.

(Edit: just to be clear, I think the same would be true for any hardware provider Michael surveys regularly, and possibly some Linux distribs too *cough Ubuntu / Red Hat cough*. Maybe that's the case and I'd be glad of it for Michael though )

[arQon]

You know what my biggest takeaway from this article was? That Quake2, which is now very nearly *25* years old, with its environments and models that are literally 2 to 3 orders of magnitude simpler than modern assets, needs a $1500 card to run at 60FPS (which it used to do on a TNT), or a $1000 card to run at the same framerate that the *software* renderer did back then.

That's how far along the RTRT road we are...