AMD Ryzen 5 5600G Linux Performance

[avem]

avem I'm going to parrot what Michael always says, he doesn't do any compiler optimizations and runs things as they are, out of the box. The results aren't wrong, that's the results. It means that there are some workloads that perform worse, or have an issue with optimization

You see TPU results, right? Then what makes you believe that 11600 here underperforms so drastically? Should I show you more results of 11600K being comparable to 5600X? Something was wrong with the 11600K system: either it was an inadequate cooler (thus throttling as a result of it) or something, low PL1/PL2 limits, I don't know - I did not run the tests but they are plain wrong and not represent the real life performance of this CPU as measured by hundreds of other reviewers.

Check Phoronix' own previous review for God's sake:

https://www.phoronix.com/scan.php?pa...l-linux&num=22

5600X and 11600K are pretty much equal in it and 11600K is not too mucher slower than 11900K. In this review something is outright broken and 11600K took a heavy beating.

[lyamc]

Uh yeah, 22 pages of like 8 benchmarks on each page will give a better overall picture

But you just goofed, because I see very similar performance with things like AV1-SVT https://www.phoronix.com/scan.php?pa...l-linux&num=14

[avem]

Uh yeah, 22 pages of like 8 benchmarks on each page will give a better overall picture

But you just goofed, because I see very similar performance with things like AV1-SVT https://www.phoronix.com/scan.php?pa...l-linux&num=14

I've not goofed.

The old test results, Blender 2.92, BMW27 : https://www.phoronix.com/scan.php?pa...l-linux&num=16

11900K : 132.80
.5600X : 188.10
11600K : 181.43

The new test results, Blender 2.92, BMW27 : https://www.phoronix.com/scan.php?pa...n5-5600g&num=6

11900K : 132.67 (more or less matches the old results)
.5600X : 181.92 (got faster for no obvious reasons)
11600K : 276.79 (suddenly 52% slower than before)

I must be tripping. Anyways, I'm done and out since no one actually pays any attention to the results while others make up theories why they are still valid despite being outright wrong.

tildearrow

Could you please ask Michael why the 11600K results in this article are so drastically worse than what Michael himself measured just a few months ago?

[lyamc]

Today’s test:

Ubuntu 21.04 was used across the test systems with the Linux 5.11 kernel and Mesa 21.3-devel from 14 August using the Oibaf PPA.

Previous test (with better 11600k perf)

Tests were done on Ubuntu 21.04 with the Linux 5.12 Git kernel and Mesa 21.1-devel

[F.Ultra]

Today’s test:

Ubuntu 21.04 was used across the test systems with the Linux 5.11 kernel and Mesa 21.3-devel from 14 August using the Oibaf PPA.

Previous test (with better 11600k perf)

Tests were done on Ubuntu 21.04 with the Linux 5.12 Git kernel and Mesa 21.1-devel

Also different motherboards. This test uses "ASUS ROG STRIX Z590-I GAMING WIFI" while the old one uses "ASUS ROG MAXIMUS XIII HERO".

And the microcode was 0x40 vs 0x39, perhaps the newer microcode is hitting the 11600k more

[smitty3268]

I must be tripping. Anyways, I'm done and out since no one actually pays any attention to the results while others make up theories why they are still valid despite being outright wrong.

"Wrong" is a matter of perspective. As noted, these tests used a different motherboard. Perhaps it's one that actually implements Intel's recommended TDP limits, rather than what some of the others do which is give infinite time to the PL2 boost.

It's an interesting catch you made, though, and an explanation from Michael by testing a few changes like the MB, kernel, etc. would be appreciated.

[mumar1]

I don't want to scream that the test results are wrong but 11600K must not be so slower than 11900K - we are talking about just a 33% difference in the core count yet in many tests 11900K is twice as fast: https://www.phoronix.com/scan.php?pa...n5-5600g&num=6

Yeah, I'm inclined to believe that the 11600K results are wrong almost everywhere or 11900K was overclocked.

Depends on the motherboard, if 11900K is tested on Z590 and 11600K on B560, depending on the B560 used, expect dramatic differences. This is dur to different power levels implemented (PL1 / PL2) check here for details

https://www.techspot.com/article/2252-intel-b560-disaster/

[ntropy]

Pretty impressive, what AMD did there, the fastest GPU and still only ranked place 3 regarding power consumption.

I bought a 3900X a day after release, as well as a couple of AMD shares and I regret nothing .

[avem]

Also different motherboards. This test uses "ASUS ROG STRIX Z590-I GAMING WIFI" while the old one uses "ASUS ROG MAXIMUS XIII HERO".

And the microcode was 0x40 vs 0x39, perhaps the newer microcode is hitting the 11600k more

If if it's a different motherboard / microcode revision the 11900K should have taken a hit as well. Its results are the same as before. I've trusted Michael's benchmarks previously, not any more, sorry. Not a single reply from him to address in certain cases a 52% performance loss.

[F.Ultra]

If if it's a different motherboard / microcode revision the 11900K should have taken a hit as well. Its results are the same as before. I've trusted Michael's benchmarks previously, not any more, sorry. Not a single reply from him to address in certain cases a 52% performance loss.

Adding to the mystery is that in every single test in the "new article" where the 11900K appears results in a "No Result File Found." when trying to access the raw data from the link to openbenchmarking.org so it's not possible to see if was run using the same MB and microcode as the 11600K. Or there is a significant bug with the 0x40 version of the microcode for the 11600K in particular.

In any way it begs to be investigated. The silence from Michael can be him being busy doing reruns here (which perhaps also explains the missing data from openbenchmarking.org).