AMD Ryzen 7 7800X3D Linux Performance

[L_A_G]

Considering this is supposed to be a gaming part these benchmarks are way too heavy on utility, compute and (bizarrely) web hosting benchmarks. Not that it's terrible at these jobs, but the "non-3D" parts are do these jobs either cheaper or better.

Still, it's been pretty hilarious to watch it give the top end of Intel's gaming focused range more than a run for its money while costing less and drawing much less power. AMD has put out a real winner of a CPU with the 7800X. I almost regret not holding out instead of buying a 7700X (with an over-priced AM5 motherboard) at launch last year.

[numacross]

Overall the 7800X3D looks to be a great gaming CPU without having to complicate matters with hybrid designs (from either side). I am shocked by how badly the 13900K behaved in the Total Frame Time tests, this result would probably be perceived as "microstuttering". Could it be caused by the scheduler putting something latency-sensitive on the E-cores?
On the other hand 7800X3D achieved a great result in the same tests.

[middy]

taskset has got you covered.

Truth to be told, neither Intel, nor AMD care too much about Linux. The Linux kernel still doesn't properly support ADL which will soon turn two years of age. I guess the X3D parts will meet the same fate.

Going by the benchmarks Michael has done, including this one, ADL appears to be as fully supported on Linux as it is on Windows 11. Intel did a lot of work getting the scheduler to work with their thread dispatcher. Are there quirks? Of course, but so with windows 11 on alder and raptor lake. The one set Michael did last year showed alder lake doing better than windows 11 on Linux.

Amd on the other hand appears to have zero support for 3d cache ccd priority pinning at all on Linux.

[Michael]

...way too heavy on utility, compute and (bizarrely) web hosting benchmarks.

https://www.phoronix.com/review/amd-ryzen-server .and more AM5 Ryzen servers are on the way...

[avis]

On acpi_cpufreq AMD does not hit its lowest clocks on idle. For example my 5950X idles at 2.2GHz though it can do 800MHz or even 400MHz per core (if I remember Windows numbers correctly). We are just waiting for amd_pstate driver

Frequency reporting for CPUs starting with Zen 2 if not earlier has been totally fake/wrong/incorrect. Start HWiNFO64 in Windows and you'll see the effective CPU cores frequencies. It's not what the CPU frequency subsystem in Linux reports.

[Mark Rose]

Considering this is supposed to be a gaming part these benchmarks are way too heavy on utility, compute and (bizarrely) web hosting benchmarks. Not that it's terrible at these jobs, but the "non-3D" parts are do these jobs either cheaper or better.

Many of us use our computers for things other than gaming, so these benchmarks are very useful to see if this processor is a fit for our tasks. Sadly most review sites will mostly benchmark games, so it's nice to have Phoronix benchmark beyond that.

[Mark Rose]

Typo: Ryzen 7 700X

[V1tol]

Frequency reporting for CPUs starting with Zen 2 if not earlier has been totally fake/wrong/incorrect. Start HWiNFO64 in Windows and you'll see the effective CPU cores frequencies. It's not what the CPU frequency subsystem in Linux reports.

Judging by the power consumption those numbers are correct. With forced amd_pstate driver it reports 800MHz and power consumption is lower. And what I read on the kernel mailing list is that default acpi_cpufreq just don't see all of frequencies thus unable to jump them and amd_pstate corrects that too. Anyways I am quite happy with acpi_cpufreq right now and just waiting proper amd_pstate implementation which won't need kernel cmdline to enable.

[avis]

Judging by the power consumption those numbers are correct. With forced amd_pstate driver it reports 800MHz and power consumption is lower. And what I read on the kernel mailing list is that default acpi_cpufreq just don't see all of frequencies thus unable to jump them and amd_pstate corrects that too. Anyways I am quite happy with acpi_cpufreq right now and just waiting proper amd_pstate implementation which won't need kernel cmdline to enable.

I've noticed zero difference in idle power consumption between acpi-cpufreq and amd-pstate. If anything amd-pstate reported slightly higher idle power consumption - for which I filed a bug report almost a year ago however it looks like AMD no longer cares.

Again, you're paying attention to the wrong numbers. The acpi-cpufreq driver is incompatible with the last 5+ years worth of CPUs and it doesn't report frequencies correctly.

[F.Ultra]

With regards to gaming, if amd wants to continue the multiple ccds with 3d cache, they are going to need to do some work to improve scheduling on linux to pin games to the ccd with the cache like they did on windows with their chipset drivers and using xbox game bar to help feed it info. If you want to game on linux and want amd, the 7800x3d is the only oblivious choice. No scheduling woes.

to be pedantic they do not pin games to the ccd with the cache on windows, they straight out puts the entire second ccd to offline