AMD Renoir Running Smooth On Linux 5.10

[phoronix]

Phoronix: AMD Renoir Running Smooth On Linux 5.10

After last week sharing some Intel Tiger Lake benchmarks on Linux 5.10, the tables have turned and here are some similar tests when running Linux 5.10 on an AMD Ryzen 4000 series "Renoir" notebook...

http://www.phoronix.com/scan.php?page=news_item&px=AMD-Renoir-Linux-5.10

[euduvda]

Nice testing.

If I understand correctly, some tests regressed and some have improved.
I know scheduler changes sometimes mess with workloads, but isn't this too much?

[plinkyplonky]

Would the previous generation of AMD Ryzen processors benefit from the newer Linux kernel? Those using the Zen+ architecture like the 3500U? Thanks.

[drakonas777]

Waiting for offended birdie to express his garbage emotion-driven and lies filled opinion on how Renoir laptops are overrated and Tiger Lake laptops are better for average user due stronger single thread performance (which is unsustainable in long runs).

[Grinness]

Waiting for offended birdie to express his garbage emotion-driven and lies filled opinion on how Renoir laptops are overrated and Tiger Lake laptops are better for average user due stronger single thread performance (which is unsustainable in long runs).

Yes, I second that!

Please Mr Birdie, enlighten us all with a beautiful chart mixing premium, non-premium, budget and 10 years old hardware and 'clearly' demonstrate how your PoV is the sole and only truth

[leipero]

Would the previous generation of AMD Ryzen processors benefit from the newer Linux kernel? Those using the Zen+ architecture like the 3500U? Thanks.

I assume it would, because those two archs are not so different. I'm using "Schedutil" (defaulted to it for some reason) for few months now on Zen+ APU, switched to "ondemand", for my workflow, there's 0 difference so I just got it back to the "Schedutil".

[Guest]

Waiting for offended birdie to express his garbage emotion-driven and lies filled opinion on how Renoir laptops are overrated and Tiger Lake laptops are better for average user due stronger single thread performance (which is unsustainable in long runs).

Well to be fair, Intel CPUs will consistently hit their Turbo 2.0 boost clocks if given enough power and cooled sufficiently, while Zen 2 doesn't ever seem to reach it's specified max boost clock, even when cooled really well and with power limits unlocked.

On the other hand, higher core/thread count is definitely helping Renoir in multi-threaded workloads.

On another hand, Tigerlake's single core performance is pretty strong, and if they put it in a desktop chip with higher power limits, it might actually beat Zen 3.

On the final hand, Intel's Linux support is still significantly superior to AMD's, so if you want a less troublesome experience on Linux, Intel is the better choice.

[3diStan]

It's all nice and swell to read that Renoir "runs" smoothly on Linux, but as far as I see, there's still a crucial part missing here: ROCm support. I'm not sure that AMD realizes how much of a handicap that is - not even just or primarily for the APU business but for their computing products as well: Developers absolutely need to be able to develop and test stuff ON THEIR NOTEBOOKS too, even for computing software that is then mostly run on big computing hardware. And that is currently possible with NVidia/CUDA but unfortunately still impossible with AMD/ROCm. As much as devs and deciders would like to prefert to buy AMD (because of the open source policy, where NVidia is lacking), that lack of ROCm support is making it impossible.

So as long as ROCm doesn't work on APU notebooks, developers who do computing are de-facto forced to buy Notebooks with NVidia GPUs… and then in consequence they tend to use CUDA and in consequence buy NVidia hardware for the big computing hardware too. And the same goes for embedded hardware (typically Camera+AI/DNN based image processing and classification): NVidia's offerings there (e.g. the Jetson family) runs CUDA… whereas the embedded APU boards cannot run ROCm. It just hurts.

The fact that AMD doesn't have ROCm support yet for APUs is all the harder to understand as there is actually a project by some swedish quasi-one-man-consulting shop that is exactly that: a ROCm port for APUs. The problem being: the guy explicitly refuses to make the sources of that available (despite it being a port of the open source ROCm he got from github). So his closed source binary ubuntu packages of ROCm for APU aren't a viable option for anyone.

So please AMD, make ROCm available for APUs!

[bridgman]

I'm not 100% sure where we are re: ROCm support for Renoir but will find out and give it a kick if I don't like what I see.

If I remember correctly we shifted APU support starting with Picasso from "HSA paths" (ATC+IOMMUv2 with OS-allocated memory) to "dGPU paths" (GPUVM with API-allocated memory) so that APUs could run exactly the same code as dGPUs.

I think we now have at least an uneasy consensus that we need to make ROCm easily accessible to developers if we want to get real adoption, so hopefully things should start falling into place re: APU support going forward.

[phoronix_is_awesome]

It's all nice and swell to read that Renoir "runs" smoothly on Linux, but as far as I see, there's still a crucial part missing here: ROCm support. I'm not sure that AMD realizes how much of a handicap that is - not even just or primarily for the APU business but for their computing products as well: Developers absolutely need to be able to develop and test stuff ON THEIR NOTEBOOKS too, even for computing software that is then mostly run on big computing hardware. And that is currently possible with NVidia/CUDA but unfortunately still impossible with AMD/ROCm. As much as devs and deciders would like to prefert to buy AMD (because of the open source policy, where NVidia is lacking), that lack of ROCm support is making it impossible.

So as long as ROCm doesn't work on APU notebooks, developers who do computing are de-facto forced to buy Notebooks with NVidia GPUs… and then in consequence they tend to use CUDA and in consequence buy NVidia hardware for the big computing hardware too. And the same goes for embedded hardware (typically Camera+AI/DNN based image processing and classification): NVidia's offerings there (e.g. the Jetson family) runs CUDA… whereas the embedded APU boards cannot run ROCm. It just hurts.

The fact that AMD doesn't have ROCm support yet for APUs is all the harder to understand as there is actually a project by some swedish quasi-one-man-consulting shop that is exactly that: a ROCm port for APUs. The problem being: the guy explicitly refuses to make the sources of that available (despite it being a port of the open source ROCm he got from github). So his closed source binary ubuntu packages of ROCm for APU aren't a viable option for anyone.

So please AMD, make ROCm available for APUs!

This is exactly why CUDA dominates the market. I don't know if AMD realize this or not. Making ROCM only run on Polaris/Radeon Vii and not NAVI 1x NAVI 2x, Renoir and APUs is a strategic mistake