AMDGPU Appears Ready To Flip On ASPM For More GPUs To Help With Power Savings

[phoronix]

Phoronix: AMDGPU Appears Ready To Flip On ASPM For More GPUs To Help With Power Savings

With Linux 5.13 AMD began enabling ASPM by default in the AMDGPU DRM driver for Navi 1x, Vega, and Polaris GPUs. Looking ahead to potentially 5.14, AMD appears to be ready to flip on this power-savings feature for the Radeon RX 6000 series (Navi 2x) along with older pre-Polaris GPUs too...

https://www.phoronix.com/scan.php?page=news_item&px=AMDGPU-ASPM-Ready

[aufkrawall]

It's quirky on Windows as well, especially with PCIe 4.0 QA seems to have gone down the drain (who knows who is to blame). At least I'm currently suspecting ASPM to be responsible for causing me hard reboots under Windows with Rocket Lake + RDNA2, thus having it completely turned off in bios (Linux didn't crash, but total system idle consumption also was ~15W higher, despite of CPU and GPU fully clocking down)...

[stalkerg]

ASPM is horrible... I worked on a kernel driver for a device with DMA and it's ruined a lot of things.

[Adarion]

Interesting. Very interesting. I'm eager to see comparisons by Michael and I'll do my own as far as I can. Got a RX 560 and a 590 here, and some older thingies.
I'm already happy my RX 590 doesn't mess around like it did before (idle way too much power consumptions, severe problems on monitor power saving and esp. returning (lost link, wrong resolution, EDID detection failed on some EIZO monitor that I have in use with older AMD models for a long time). However, it was messy on Windows, too.
I'm curious if I'll see some improvements.

(running with pcie_aspm=force as kernel boot parameter for years, though; not sure if there are blacklists/quirks in the kernel)

[intelfx]

ASPM is horrible... I worked on a kernel driver for a device with DMA and it's ruined a lot of things.

How so?