Ryzen memory clock misreport

[Hicookie]

it's been awhile, there are records out there since ryzen 3000 launch 

I think it's not hard to figure out there are mis readings of ryzen 3000's memory speed up to 5.1GHz

which barely there is no single one software can particularly report it right

people choose not to say or take advantage from it, used for marketing or competition, some may presented for companies

last week I take a chance to check out those weird results, and double check with some softwares

and oscilloscope, and found out those softwares are truly broken and fail to report the right frequencies

as you can see from attached pitcures, all software except ryzen master are reporting the wrong clock

and they still can pass cpuz validation which means software fail to report correctly

though, rzyen master is not always right, becasue it can't detect the out side BCLK, once you jump up 

the BCLK rzyen master only reads the memory clock at 100MHz

here is the validation I just add to show CPUz's buggy memory frequency

CPUz shows 5.916GHz, but actually it's 55*102 not 58*102

https://valid.x86.fr/qx3ezm 

 

[Leeghoofd]

Did you have contact with the CPUZ programmer Cookie? Another AMD debacle is not what we need right now...

[_mat_]

Thank you for the insight, Hicookie! I think we need to take this seriously and have a look at the underlying implementation of the Ryzen memory clock detection in CPU-Z and HWiNFO. This is not the first time that clock frequency detection via software is untrustworthy. This has happened before due to CPU vendors not being able to provide the necessary functionality to read frequencies reliably.

I've contacted Martin from HWiNFO as well this morning to get his take on the issue. Will update as soon as I know more.

Hicookie, have you tried the latest Ryzen Master tool as well (2.0.2.1271 and above)? It should have the new AMD Monitoring SDK built in and might bring some changes: https://community.amd.com/community/gaming/blog/2019/09/10/ryzen-community-update-bios-updates-for-boost-and-idle-plus-a-new-sdk

[Mumak]

Which ABL version was used in the test? At least WABLMT95222040 is required to get proper memory clock setting/readout, earlier versions had a bug there.

[Leeghoofd]

Issue is the validation purpose for the given clocks... If cpuz can't make it happen we have another problem with AMDs latest and greatest

[ObscureParadox]

I think the problem really only lies in allowing AMD in memory frequency results though right? Because ofc you will have skewed data for benchmarks, but the overall score doesn't change and that's the most important thing. At the end of the day we never banned X48 when it reported incorrect ram speeds and it's been doing that since it's inception on certain dividers.

[Leeghoofd]

Yes...  Mem validation will be a big mess

[mllrkllr88]

Kevin Wu showed this on Facebook over a month ago. I have observed the same behavior on ASRock X570 Taichi. The issue is over 5000MHz.  Set 5200 in the bios and all software shows 5300...

[Mumak]

Because encoding of mem clock changes past 5000. This was not initially accounted for in earlier BIOS code (AGESA, ABL) and tools . It was fixed in later tool versions and ABL WABLMT95222040. So if older ABL is used in combination with later tools it might report an incorrect result.

[_mat_]

18 minutes ago, Mumak said:

WABLMT95222040

Do these internal version numbers somehow translate to the external ones that we can see in tools? Like 1.0.0.3 or something like that?

[Mumak]

58 minutes ago, _mat_ said:

Do these internal version numbers somehow translate to the external ones that we can see in tools? Like 1.0.0.3 or something like that?

I don't think so. Each component has its own numbering.