Linux 5.11 Drops AMD Zen Voltage/Current Reporting Over Lack Of Documentation

[Mario Junior]

So the "AMD has no drivers" still real. 😂

[birdie]

agd5f bridgman

I'm just curious how the authors of HWiNFO64, HWMonitor, Aida64 get a METRIC TON of information from AMD in order to show a billion of data points in regard to your CPUs.

[Guest]

agd5f bridgman

I'm just curious how the authors of HWiNFO64, HWMonitor, Aida64 get a METRIC TON of information from AMD in order to show a billion of data points in regard to your CPUs.

True. Their explanations just sound like bullshit excuses.

"Atleast it's available on APUs" - because they're going to be used in Chromebooks, no other reason. Why not on desktop CPUs? Why not on all of these other CPUs that consumers pay hundreds of dollars to buy?

Intel provides support well ahead of time, without waiting to see if something has enough market share. If AMD pulled the same stunts on Windows, they'd be trashed and ridiculed.

This is just fucking ridiculous.

[Tomin]

"Atleast it's available on APUs" - because they're going to be used in Chromebooks, no other reason. Why not on desktop CPUs? Why not on all of these other CPUs that consumers pay hundreds of dollars to buy?

Or maybe because they share the interfaces (hardware and/or kernel) with graphics? He did mention that it's through the same interfaces that GPU driver exposes. There may be hardware interfaces that don't exist for CPUs without intergrated graphics.

[bridgman]

agd5f bridgman

I'm just curious how the authors of HWiNFO64, HWMonitor, Aida64 get a METRIC TON of information from AMD in order to show a billion of data points in regard to your CPUs.

At the risk of repeating ourselves, agd5f and I work on the GPU side of the business. We don't know how the closed-source utility developers get their information - all we have to go on right now is answers from the last time we attempted to raise this issue with the CPU folks. We are going to try again but figuring out who to talk to is the first step.

Releasing technical information to closed-source developers is a lot easier than releasing for open source use though.

True. Their explanations just sound like bullshit excuses.

"At least it's available on APUs" - because they're going to be used in Chromebooks, no other reason. Why not on desktop CPUs? Why not on all of these other CPUs that consumers pay hundreds of dollars to buy?

As Tomin suggested, the difference is that on APUs the same HW interface is used for CPUs and GPU. On discrete CPUs there is a different interface.

[torsionbar28]

Who cares about viewing the voltage and current? All that matters is die temperature and clock speed. The only use case I can see for knowing what the voltage and current draw are, is for the kiddies doing their overclocking foolishness on their l33t gam3r peecee. So this has no relevance to servers or commercial workstations. Folks using Linux on the desktop are a miniscule sliver of the market. Folks using Linux on Zen on the desktop are an even smaller sliver. And folks using Linux on Zen on the desktop, and overclocking their hardware are a grain of sand on the beach. It is perfectly understandable why this is not a priority for the folks at AMD.

FWIW I've been using AMD CPU's since late 1980's 386/486 days, and the only time I can ever recall being interested in overclocking, was during the MHz wars of the 1999-2000 era. Nowadays with high base clocks and automatic turbo boost, there is literally no point in manual overclocking.

[BlueCrayon]

agd5f bridgman

I'm just curious how the authors of HWiNFO64, HWMonitor, Aida64 get a METRIC TON of information from AMD in order to show a billion of data points in regard to your CPUs.

Pretty sure they just load up stuff like ryzenmaster (or whatever the OEM provides) in a kernel debugger and work from there. This is how reverse-engineered GPU drivers start off usually.

[birdie]

Pretty sure they just load up stuff like ryzenmaster (or whatever the OEM provides) in a kernel debugger and work from there. This is how reverse-engineered GPU drivers start off usually.

They had the info before Ryzen Master had been released.

[Space Heater]

We are going to try again but figuring out who to talk to is the first step.

Thank you for doing that, hopefully whoever you talk to will listen.

[AdrianBc]

Who cares about viewing the voltage and current? All that matters is die temperature and clock speed. The only use case I can see for knowing what the voltage and current draw are, is for the kiddies doing their overclocking foolishness on their l33t gam3r peecee. So this has no relevance to servers or commercial workstations. Folks using Linux on the desktop are a miniscule sliver of the market. Folks using Linux on Zen on the desktop are an even smaller sliver. And folks using Linux on Zen on the desktop, and overclocking their hardware are a grain of sand on the beach. It is perfectly understandable why this is not a priority for the folks at AMD.

FWIW I've been using AMD CPU's since late 1980's 386/486 days, and the only time I can ever recall being interested in overclocking, was during the MHz wars of the 1999-2000 era. Nowadays with high base clocks and automatic turbo boost, there is literally no point in manual overclocking.

I agree that most users might not care about voltage and current values.

However, serious developers need them.

More precisely, what is needed is the power consumption per core, not necessarily the voltage and current values required to compute it.

For any Intel CPU, it is easy to view the power consumptions while you are running various programs.

For many things that may be computed, there are alternative methods to structure the computations, which will result in different energy consumptions.

There are applications for which one is interested to determine the method that will need the least energy. This is easy to do with Intel CPUs, but too hard to do with AMD CPUs.

There also other problems that make too difficult the optimization of programs for AMD CPUs, e.g. the lack of documentation for some performance counters, or maybe even the lack of some performance counters that are available for Intel CPUs.

Taking these problems into account, there is no wonder that in many cases Intel CPUs may win benchmarks against superior AMD CPUs, because the existing programs are optimized for Intel, but not for AMD.

In my opinion, AMD should better start to provide better documentation for developers, like they provided many years ago, at least at the level at which Intel still does, because it is stupid to spend resources only for designing good hardware, which ends being used below its abilities, while a very small additional effort for better technical documentation could bring as much additional performance as much more costly hardware enhancements.