Radeon RX 6800 Series Performance Comes Out Even Faster With Newest Linux Code

[bridgman]

BTW if I remember correctly the only generated code in the ROCm stack other than compiled binaries is (a) register headers in the kernel driver (upstream community project) and (b) code generated by Tensile (a compute kernel optimizer) as part of some of the math libraries (maybe just BLAS). I don't think Tensile-generated code is visible except at runtime during a debug session. Is there some other generated code you didn't like ?

One last question for now I hope... the ROCm stack is basically made up of:

- low level drivers (mostly upstream)
- HIP/Clang compiler/runtime
- OpenCL compiler/runtime
- OpenMP compiler/runtime
- a lot of optimized libraries
- porting/tuning of ML frameworks and HPC apps (with the changes going into the upstream versions of those projects)

The compiler/runtime portions are built around upstream LLVM and Clang, along with back-end code that connects to the low level drivers.

Are there specific parts of the stack where you think the code is particularly bad ? My concern is that you may be conflating a historically clunky build/install experience with the actual source code.

Thanks !

[zeb_]

These results are amazing. Having an open source solution that compares in term of performance in the high-end tier is very welcome. Congratulations to the developers.
One thing I would be interested in is image quality, textures etc. I remember the time of the first Radeon ATi (circa 2000?), which had a better rendering than nVidia. Is there a difference between Mesa drivers? Between AMD cards? or versus nVidia?

[lyamc]

The difference is you treat your opinion like it’s a fact

I think it is. But that's just my opinion.

lol, you’re insane.

”In my opinion, my opinion is a fact”

You might want to actually read some of those psychology today articles you’ve been posting. I think it would do you some good.

[ernstp]

These results are amazing. Having an open source solution that compares in term of performance in the high-end tier is very welcome. Congratulations to the developers.
One thing I would be interested in is image quality, textures etc. I remember the time of the first Radeon ATi (circa 2000?), which had a better rendering than nVidia. Is there a difference between Mesa drivers? Between AMD cards? or versus nVidia?

The specifications (Vulkan, OpenGL, etc...) are much better these days and everything should look identical, there is very little room for adjusting things.

Also, some people said that the electrical CRT signals from some cards were better, but now everyone have digital connections so that's also gone...

[duby229]

lol, you’re insane.

”In my opinion, my opinion is a fact”

You might want to actually read some of those psychology today articles you’ve been posting. I think it would do you some good.

Obviously it was a joke.... Duh...

[duby229]

BTW if I remember correctly the only generated code in the ROCm stack other than compiled binaries is (a) register headers in the kernel driver (upstream community project) and (b) code generated by Tensile (a compute kernel optimizer) as part of some of the math libraries (maybe just BLAS). I don't think Tensile-generated code is visible except at runtime during a debug session. Is there some other generated code you didn't like ?

One last question for now I hope... the ROCm stack is basically made up of:

- low level drivers (mostly upstream)
- HIP/Clang compiler/runtime
- OpenCL compiler/runtime
- OpenMP compiler/runtime
- a lot of optimized libraries
- porting/tuning of ML frameworks and HPC apps (with the changes going into the upstream versions of those projects)

The compiler/runtime portions are built around upstream LLVM and Clang, along with back-end code that connects to the low level drivers.

Are there specific parts of the stack where you think the code is particularly bad ? My concern is that you may be conflating a historically clunky build/install experience with the actual source code.

Thanks !

Look, I never meant to offend anybody. I looked through the OpenCL compiler as that is the functionality that would complete my user experience. It definitely feels like cookie cutter or generated code.

[lyamc]

It definitely feels like cookie cutter or generated code.

What are you talking about? Everyone is running generated code.

Look, I never meant to offend anybody.

I seriously doubt that. You started by calling something bad for arbitrary reasons and offering nothing constructive, while attacking anyone asking for something constructive.

And you talked like you knew what you were talking about. That’s going to get people who do know what they’re talking about to go after you for spreading misinformation. It’s your own fault.

[Anarchy]

Really good results.

Hopefully I'll manage to get a 6800xt by the time ubuntu 21.04 comes out. It looks like stock is really tight for these cards and the prices are consequently sky high.

[bridgman]

Look, I never meant to offend anybody. I looked through the OpenCL compiler as that is the functionality that would complete my user experience. It definitely feels like cookie cutter or generated code.

That's fair - the compiler code does look very cookie-cutter and repetitive, however I believe that is a consequence of the way that llvm/clang is designed rather than individual developer decisions. We fill out all the templates (ie yes the code changes are ours) but I don't *think* there is any way to add support to an llvm/clang compiler without the code looking more or less that way.

BTW I don't think anyone is taking offence, just interested in where this is going. You made what was probably intended to be a casual comment that was arguably a bit too broad, and when someone challenged you it seemed that instead of tweaking your comment like everyone expected you kept on digging...

... and everyone wants to be a part of something like that.

[zeb_]

The out-of-the-box experience is not good right now, it looks much better than it was 5700XT, but the experience wont be good until Ubuntu 21.04 or w/e your favorite distro is around that time.

Can you please expand on that? Are concerns about stability or other aspects? I have used nVidia cards for quite a few years on an Archlinux system and they always worked fine with the proprietary driver. Now going open source is a big advantage in my opinion.