Ticked Off: Clock Bug Leads to False Ryzen Benchmark Scores

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Sep 6, 2019
2
0
10
0
I'm the dude that was interviewed for the article. I'd like to answer your questions and chime in.

The article is neither wrong nor 100%. Its actually both. You have to verify the BIOS supports it and make sure its enabled then you have to force Windows to use it:

That's exactly what the article is saying. It needs to be enabled in BIOS, afterwards enabled in Windows (bcdedit) then rebooted. The system will use HPET as the System Clock Interrupt and for the often used QueryPerformanceCounter (WIN32 API function).

So basically this is a Windows bug meaning all Linux benchmarks such as from phoronix show true results (just without optimization coming in later kernels).
No, Linux is affected as well on the before-mentioned platforms, if the OS is configured to use the LAPIC timer as time source (which normally is the case) and depended timer functions or the RDTSC(P) instruction is used. But Linux doesn't allow to change the bus frequency easily and it's not used for overclocking much, so it's not really a relevant issue.

It's hard to avoid the perception that your article title is clickbait, and anti-AMD clickbait at that. At most it should have said "can lead to false Ryzen benchmark scores".

As it is, you've successfully given the impression that all Ryzen benchmark scores are suspect. Intel marketing is surely happy to see that in a headline.

Does all this somehow not even occur to you? Really?
The article was heavily edited and prematurely released by accident. My interview was pretty much slashed from 14,000 chars to a few lines.

I do very much think that the title is misleading and the whole article is hung up on Zen 2 too much. The use of "Clock Bug" in the title adds to the confusion and is mistaken for the "Boost Clock Bug" a few times in the comments on Facebook.

That said the core of the article was actually about trust in benchmark results. We assume that a platform returns valid results, when we actually shouldn't. The "RTC bug" is one of the reasons why we can't. We see a screen in some forum or a number in a vendor ad and the truth is that we can't rely on that. We might know from our own systems, that we haven't fiddled with the base clock or weren't cheating, but we have no clue what somebody else was doing.

This applies to reviewers a well. Anandtech for example had a wrong timer configuration and all result were skewed due to its performance impact (not skewing!) on the benches. We all saw the numbers there, but had no clue that the author had HPET enabled in Windows. Here is the article about what happened: https://www.anandtech.com/show/12678/a-timely-discovery-examining-amd-2nd-gen-ryzen-results

That's what BenchMate does. It gives you the information to verify for yourself that this score is trustworthy, reliable and repeatable.

Microsoft needs to step it up.

Bug after bug. If your paying over $100 for an operating systrm, you would expect it to work better than a free linux based os.

Every day its something new pertaining to ryzen.
Windows scheduler issues, Windows power management issues, high idle cpu usage issues.

I cant see how some of these are hard to fix.

When Windows detects a ryzen cpu it should change some things. Adapt the sceduler and power settings for good performance but also efficiency
You are right, Microsoft could detect Ryzen and change the time source for the System Clock Interrupt to something else (like RTC in Windows 7). But let's not forget about the requirements of tickless kernels as I've tried to answer in the article. I'd like to think that there was a reason for switching to the flawed LAPIC timer.

On the other hand, AMD can alter their design of their platform to maintain two bus clocks. A single one that's stuck to exactly 100 MHz and a second one that drives memory and CPU. Intel did that with Skylake and I'd like to think intentionally removed that flaw. It might still be there for non-Z mainboards though, as it seems that they only have one bus clock to reduce costs. My guess is that modded H boards should be prone to the same timer skewing as well.

If you have any questions regarding all this, just let me know.
 

splave

Contributing Writer
Editor
Jan 4, 2019
9
6
15
0
Clock Bug Leads to False Ryzen Benchmark Scores

Actual fact: If you've adjusted the base clock of your Ryzen system, benchmark scores will be affected.

It's hard to avoid the perception that your article title is clickbait, and anti-AMD clickbait at that. At most it should have said "can lead to false Ryzen benchmark scores".

As it is, you've successfully given the impression that all Ryzen benchmark scores are suspect. Intel marketing is surely happy to see that in a headline.

Does all this somehow not even occur to you? Really?
I dont make the titles. Im sorry if you feel that way. The fact is it does lead to false ryzen scores. It does not lead to false intel scores. Keeping companies accountable is not a bad thing in my opinion. Turbo boost clocks being too low being exposed caused AMD to release a statement saying they are fixing it. Does that not benefit everyone?

I would agree -- Seems like this article was written by the Intel Marketing Team. Windows 10 bug, affects both Intel & AMD, except maybe the later generation Intel, doesn't say how or using what magic. A better title would be Windows 10 dynamic tic prevents accurate benchmarks. Period. Nah, let's turn it into click bait.
It doesnt effect current Intel though... Would you like me to reference Intel processors from 2014 in this as well? Windows 10 results are flawed when you use AMD Ryzen 3. It is what it is. If im the intel marketing team then why do I have the world record in gpupi using a thread ripper 2990wx and why am I typing this on an x570 taichi with a 3900 and a 3700x sitting next to it?

The article is to show a flaw that will hopefully get fixed. No one is discussing this flaw beyond hwboters that want their points to count in windows 10. Benchmate is the way to do this. I want benchmate to succeed and so should AMD so their results can be verified to be true in the mean while until they can physically fix the flaw or patch it somehow with MS.

For those saying dont touch the bclk then no problem, that is also inaccurate. Stock maybe 100 but it also fluctuates. What then? Its constantly fluctuating.

This is a difficult topic and If I didnt make things clear enough I apologize for that. I'm not a bad guy with any ill intentions. Im an unpaid hobbyist in overclocking. I want everything to be better for everyone honestly.
 
Last edited:

ASK THE COMMUNITY

TRENDING THREADS