Ryzen 7 CPUFreq Governor Comparison For Linux Gaming On 4.12

[phoronix]

Phoronix: Ryzen 7 CPUFreq Governor Comparison For Linux Gaming On 4.12

A few days back I posted some fresh P-State and CPUFreq governor tests on Intel hardware while now is a similar comparison on the AMD side with a Ryzen 7 1800X processor and Radeon R9 Fury graphics card.

http://www.phoronix.com/vr.php?view=24858

[tildearrow]

Typo:

Metro Last Light Redux didn't see mufh of a change in performance until hitting the powersave governor.

[NotMine999]

Maybe I am just not seeing it in the page 1 details table, but it would be nice if the PTS readout/table clearly indicates the version of the BIOS on the motherboard being used. I can interpret that "v1.0" value at the end of the MSI string in the "motherboard" line in different ways.

Some vendors, like ASUS (just look through their Support web site for examples), have different hardware versions of their motherboards along with different BIOS versions, sometimes hardware-specific, for those motherboards. And... those ASUS hardware and software versions do not "line up", i.e. there is rarely if ever any correlation between the two. Hence my statement that I can interpret "v1.0" in different ways.

Why do I think that this is so important that it should be reported to the reader?

Other online sites have reported Ryzen performance differences in their testing with the only variable being the motherboard BIOS in use. Some motherboard vendors are making the effort to improve their BIOS based on bug fixing, performance adjustments, and AMD AGESA updates.

Said another way, when comparing test results for the same hardware and software from different online sites it is nice to make as much of an "apples to apples" comparison as possible.

IMHO side note:
MSI is one of those vendors that rarely updates it's motherboard BIOS; Supermicro is comparable with it's infrequent BIOS updates. On the other hand, ASUS seems to make BIOS updates a semi-annual or even more frequent event.

[Michael]

Maybe I am just not seeing it in the page 1 details table, but it would be nice if the PTS readout/table clearly indicates the version of the BIOS on the motherboard being used. I can interpret that "v1.0" value at the end of the MSI string in the "motherboard" line in different ways.

Some vendors, like ASUS (just look through their Support web site for examples), have different hardware versions of their motherboards along with different BIOS versions, sometimes hardware-specific, for those motherboards. And... those ASUS hardware and software versions do not "line up", i.e. there is rarely if ever any correlation between the two. Hence my statement that I can interpret "v1.0" in different ways.

Why do I think that this is so important that it should be reported to the reader?

Other online sites have reported Ryzen performance differences in their testing with the only variable being the motherboard BIOS in use. Some motherboard vendors are making the effort to improve their BIOS based on bug fixing, performance adjustments, and AMD AGESA updates.

Said another way, when comparing test results for the same hardware and software from different online sites it is nice to make as much of an "apples to apples" comparison as possible.

IMHO side note:
MSI is one of those vendors that rarely updates it's motherboard BIOS; Supermicro is comparable with it's infrequent BIOS updates. On the other hand, ASUS seems to make BIOS updates a semi-annual or even more frequent event.

It's more of a Linux problem.... PTS does report it when available. The v1.0 is what is being reported as the BIOS version, even though this board was like 15/1.0.5 (the latest at time of testing, ended in 5 if my memory serves me), but the problem is some motherboards expose their version in different way (not always the same field), some BIOS parsing only works as root, etc. If there is some other new interface available for reliably reading the BIOS version as a user, happy to use it but AFAIK no new options since last checking.

[fallenbit]

I have 2 questions:

1) power draw on idle (desktop) for each governor
On my Kaveri it is only +3W or so with performance - would think, that all the hold states are reached, and that is the more important thing?

2) figures for having no governor at all
During the introduction days of Rizen I've read (here?) that the CPU can controll it's states allone and with best results. Am I mixing something up? Do I remember this wrong?

A further hint:
Have you ever tried to eliminate the decimal fractions from all these figures (avg, min, max)? It would look much more professional. Zero decimal fractions on anything from 10 and above, 1 decimal fraction for figures below 10. Why? Readability. Management papers would not allow any of this pseudo details. These figures change every second with new releases of any software used, PC load changes, hell - even case temperature. This is no exact science. But for the human brain, it is much easier to only read full numbers without fractions.

[Herem]

I have 2 questions:

1) power draw on idle (desktop) for each governor
On my Kaveri it is only +3W or so with performance - would think, that all the hold states are reached, and that is the more important thing?

2) figures for having no governor at all
During the introduction days of Rizen I've read (here?) that the CPU can controll it's states allone and with best results. Am I mixing something up? Do I remember this wrong?

A further hint:
Have you ever tried to eliminate the decimal fractions from all these figures (avg, min, max)? It would look much more professional. Zero decimal fractions on anything from 10 and above, 1 decimal fraction for figures below 10. Why? Readability. Management papers would not allow any of this pseudo details. These figures change every second with new releases of any software used, PC load changes, hell - even case temperature. This is no exact science. But for the human brain, it is much easier to only read full numbers without fractions.

Dropping the fraction part of numbers over 10 could lead to changes of up to 5% not being observable.

While this might be fine for managers (who often struggle to understand anything that's not pie shaped), it's not much help for describing subtle variations in technical data.

[xpander]

just like with FX cpus and now with my Ryzen 7 1700X. Been always using performance mode when going into gaming and ondemand for destkop use. Just because of the performance drop in ondemand mode in some games
keyboard binds on my macro keys to make it simple and fast to switch.

[microcode]

Note that on an APU system with a heatsink that can't handle the max dissipation of the chip, using governors other than performance can yield a performance improvement.

[starshipeleven]

2) figures for having no governor at all
During the introduction days of Rizen I've read (here?) that the CPU can controll it's states allone and with best results. Am I mixing something up? Do I remember this wrong?

Carrizo or later can do that fine, I've seen some laptops whose ACPI tables were horribly borked.
They were unable to expose CPU frequency control to the OS so you could simply not set governors period, and the CPU was still changing frequency on its own. Sure it might have had a crappy logic to do so, but from comparisons with Windows (where everything was ok as the acpi tables were using windows-only functions) it seemed to act the same.

So it is imho entirely possible that desktop processors do that on their own too.

[starshipeleven]

Note that on an APU system with a heatsink that can't handle the max dissipation of the chip, using governors other than performance can yield a performance improvement.

Translating for the plebs: "APU system with a heatsink that can't handle the max dissipation of the chip" = most laptops