AMD Linux Gaming Performance Largely Unchanged With Linux 6.6 Git

[phoronix]

Phoronix: AMD Linux Gaming Performance Largely Unchanged With Linux 6.6 Git

With the AMD performance uplift on the Linux 6.6 kernel due to the EEVDF scheduler code, the workqueue enhancements for chiplet-based processor designs, and other improvements, many Phoronix readers have speculated over AMD Linux gaming performance improvements with this in-development kernel...

https://www.phoronix.com/news/AMD-Gaming-Linux-6.6-6.5

[Mitch]

I'm glad to see testing here. I wasn't sure what to expect exactly.

I imagine EEVDF's desktop improvements would come in the form of better UI responsiveness under heavy load, and, for gaming, possibly better low 1% FPS and low 0.1% FPS scores, especially when background work is going on.

[Errinwright]

I'm glad to see testing here. I wasn't sure what to expect exactly.

I imagine EEVDF's desktop improvements would come in the form of better UI responsiveness under heavy load, and, for gaming, possibly better low 1% FPS and low 0.1% FPS scores, especially when background work is going on.

I suspect 1% lows would be more relevant to the general player base and I concur with regards to the statement. It would be interesting to see such tests, in addition to testing the BORE-EEVDF 'burstiness' optimizations for gaming.

[creative]

From 5.15.X to 6.+ going to be honest 5.15 is a great gaming kernel. What people are more than likely to benefit the most from is certain libraries being updated, stuff like that. Running a development distribution however is snappier with certain things like Plasma logout, that's a Plasma base version difference and things getting cleaned up.

Even on Manjaro 5.15 was excellent. 5.15 at least for me has been the most memorable kernel version and the one I have payed the closest attention to personally. I never really have issues with kernels but that version across the board seemed in terms of performance to be just the right amount of snappy and balanced.

I pretty much stay in stable yet current development on my selected OS certain things are quicker but is no game changer enough to daily drive it like stable, I just maintain it and keep its sets of software mirrored to my daily driver. Both share the same gaming library drives.

My hardware is a generation or two old though.

I put my old i7 7700 65 watt +GTX 1050 TI as a backup system back together this morning and booted up Liveslak 15 for testing it, great system. I need to get a drive for it, all of its drives got migrated over to this one. I have been wanting to turn it into a giant outbound multi effect DSP system for a while now.

[JEBjames]

Michael

Typo "but basedo n the initial" should be "but based on the initial"

[timofonic]

I hoped improvements would be better and less regressions, but well...

[Kjell]

I hoped improvements would be better and less regressions, but well...

EEVDF is designed to provide latency reductions while maintaining good throughput over CFS. Snappier desktop and lower input lag is a major win in my book

[yump]

This is a good test for the CCX-bound workqueues, but if the EEVDF makes any difference in gaming, I'd expect it to show up on on much lower core count CPUs. The 7950X is a 32-thread processor, and games almost never have that much concurrency. It makes sense that the scheduler would make effectively no difference if newly-runnable threads can always be punted to an idle CPU.

[geearf]

This is a good test for the CCX-bound workqueues, but if the EEVDF makes any difference in gaming, I'd expect it to show up on on much lower core count CPUs. The 7950X is a 32-thread processor, and games almost never have that much concurrency. It makes sense that the scheduler would make effectively no difference if newly-runnable threads can always be punted to an idle CPU.

Maybe redoing this test with heavy compilation in the background? That should occupy the CPU some more.

[skeevy420]

Maybe redoing this test with heavy compilation in the background? That should occupy the CPU some more.

Or one of the multi-CCD X3D chips since those have the wonky cache layouts that EEVDF should help with or an Intel hybrid big.LITTLE x86 CPU. All things considered, I'm not really that surprised that the best AM5 CPU without 3D cache had next to no change under gaming loads since it's a balanced CPU without any of the unbalanced oddities that EEVDF is supposed to assist with.

It makes perfect sense that a regular CPU stays the same. The oddballs are what should have the most improvements.

EEVDF does feel a hell of a lot more snappy. To put that in perspective, I upgraded hardware last Tuesday and I noticed a significant feels improvement going from a 4650G Zen 2 to a 7800X3D Zen 4 in regards to Firefox scrolling, KDE menus, and other UI tasks (same OS, same SSD, new CPU/RAM/MB). With the 7800X3D, going from CFS to EEVDF was even more noticeable in the feels department in regards to just using the KDE desktop and doing non-intensive tasks like opening up makepkg.conf in a text editor. Using the feels standard, it's the best any computer I've ever owned has felt....granted, I have one of the best processors in the world so it goddamn better pass the feels test.

I never ran EEVDF with the 4650G....and I'm not going to disassemble my PC to do a CPU/RAM/MB swap to find out.

The only noticeable downside with EEVDF was that I had to lower my PBO Curve Optimizer voltage from -40 back down to -30 due to random crashing. CFS ran -40 stable, not one crash ever, while EEVDF crashes above -30 within in a minute of seeing the desktop which makes me wonder if more efficient threading decisions can affect a slightly too undervolted CPU.