Even at -180 Degrees, Ryzen 3000 May Not Hit Boost Clock Speeds

[Admin]

If you want to hit the advertised boost clock on your third-gen Ryzen chip, great cooling could not be enough.

Even at -180 Degrees, Ryzen 3000 May Not Hit Boost Clock Speeds : Read more

 

[Darkbreeze]

The end of the article repeats the same paragraph twice, once before the CPU-Z graphic and once at the end of the article.

 

[NightHawkRMX]

We are seeing clock speeds that are 1% lower than what AMD advertised. I'd consider Ryzen 3000 boost clocks good enough, although I'm not sure if the results would be the same with a decent normal cooler.

This entire article screams "I will complain about AMD being 1% off" since I have nothing useful to report on. Anyone that gripes about getting 99% of what they expect is a real pessimist and wasting their time.

Can I see whether a 9700k hits rated boost clock under the same load and same scenarios?

 

[justin.m.beauvais]

Did you listen to nothing AMD said when they were talking about when they were discussing their boost speeds? They specifically said that Cinebench was NOT the proper way to measure maximum boost clock speed.

I'll give you credit for excellent ABBA puns. It takes quite the Super Trooper to come up with all those. It just seems that by AMD's statements that your testing methodology was fundamentally flawed. Tom's previously tested the ABBA update and found AMD's statement to be accurate as well. They achieved higher momentary boost clocks with other workloads. Soooooooo, yeah. Thank You for the Music but When All is Said and Done I feel like this test is at best inaccurate.

 

[TJ Hooker]

This does seem to be beating a dead horse.

And as said above, AMD has since clarified that a bursty single threaded load is the best way to observe max boost clocks, not consistent sustained loads like Cinebench.

Following the installation of the latest BIOS update, a consumer running a bursty, single threaded application on a PC with the latest software updates and adequate voltage and thermal headroom should see the maximum boost frequency of their processor. PCMark 10 is a good proxy for a user to test the maximum boost frequency of the processor in their system. It is fully expected that if users run a workload like Cinebench, which runs for an extended period of time, the operating frequencies may be lower than maximum throughout the run.

https://community.amd.com/community...ios-updates-for-boost-and-idle-plus-a-new-sdk

 

[NightHawkRMX]

I think the only thing lost here is the writer's time, and not the performance of the CPU tested.

They went through the trouble of testing on liquid nitrogen and still used a workload that AMD has stated will not boast the CPUs performance. This really reminds me of the Verge PC build where they built a high-end PC but yet played league of legends on it.

Really if you test an AMD CPU and have an issue with it, you should show how the Intel equivalent performs to see if that CPU exhibits the same issue.

 

[Darkbreeze]

What those other articles DON'T cover though, is the fact that you can reduce the core voltage (Which we KNOW increases the longevity of the CPU) WHILE increasing the all core manually configured boost frequency, by manually configuring an overclock yourself and simply not allowing any of the automatic features to do their thing.

That is worth noting, however, it isn't anything new and has been universally true across the board for as long as manufacturers have been CREATING boost profiles and automatically configuring voltage, for just about any platform you'd care to look at. So, it's really not a revelation, although it IS the first time I've seen anybody interject that into a major article pertaining to these concerns on Ryzen.

 

[TJ Hooker]

1.49V Vcore under load with stock settings does seem unusually high. I was under the impression that with Ryzen you may see it spike to high 1.4x sporadically under light/bursty load, but it would be less under sustained heavy load.

 

[CountMike]

Cinebench was never meant to or able to push CPU to limits, jus to measure performance for Cinema4D. My 3700x with 1003ABBA gets one or two cores up to 4.391GHz with other cores 4.25GHz while using Geek bench for instance can go over 4.41Ghz. That looks good enough on paper but beside it being close enough to "promised frequency" OS and programs may not hit preferred cores. RM declares that those are 02, 03 and 07 in that order in my case.

 

[owh88]

Many whiners here complaining about nothing and missing the point. I found this article very informative. The voltage delta between boost and fixed clocks is huge. The performance differential between boost and fixed at near identical clocks is also interesting. I would definitely want to know that, if I owned a Ryzen 3000, or was planning to buy one.

 

[splave]

We are seeing clock speeds that are 1% lower than what AMD advertised. I'd consider Ryzen 3000 boost clocks good enough, although I'm not sure if the results would be the same with a decent normal cooler.

This entire article screams "I will complain about AMD being 1% off" since I have nothing useful to report on. Anyone that gripes about getting 99% of what they expect is a real pessimist and wasting their time.

Can I see whether a 9700k hits rated boost clock under the same load and same scenarios?

Only thing I complained about really was that using a fixed clock speed that is at the same speed as their reported boost of 4.4ghz you can use 0.25v less, have more performance and less heat on ALL cores! The entire article is about that, have another look. I even complimented that the cpu is 3600mhz on the box and runs at 4200mhz all cores without even changing anything in the bios.

Did you listen to nothing AMD said when they were talking about when they were discussing their boost speeds? They specifically said that Cinebench was NOT the proper way to measure maximum boost clock speed.

I don't recall seeing it saying "4.4 boost clock (just not cinebench)" on the box anywhere? This is the point of independent testing. Regardless the article was to show that the maximum boost is irrelevant. Apply that clock to all your cores and enjoy maximum performance all the time while saving power. Please reference the charts.

 

[NightHawkRMX]

There are a lot of factors that contribute to boost clocks.

I wander if you would see full intel boost in cinebench. Like 3700x vs 9700k which would be closer to advertised speeds.

Would one of them meet the rated boost?

 

[PaulAlcorn]

Did you listen to nothing AMD said when they were talking about when they were discussing their boost speeds? They specifically said that Cinebench was NOT the proper way to measure maximum boost clock speed.

I'll give you credit for excellent ABBA puns. It takes quite the Super Trooper to come up with all those. It just seems that by AMD's statements that your testing methodology was fundamentally flawed. Tom's previously tested the ABBA update and found AMD's statement to be accurate as well. They achieved higher momentary boost clocks with other workloads. Soooooooo, yeah. Thank You for the Music but When All is Said and Done I feel like this test is at best inaccurate.

Splave is referring to the goalposts that AMD defined to Der8auer for his survey: The company recommended using Cinebench sT as the test program for respondents to the survey.

 

[PaulAlcorn]

1.49V Vcore under load with stock settings does seem unusually high. I was under the impression that with Ryzen you may see it spike to high 1.4x sporadically under light/bursty load, but it would be less under sustained heavy load.

AMD has stated in a press release that up to 1.5V is acceptable for the 7nm process. These high voltages are likely indicative of something else, though. Voltages are programmed to account for aging. As the chip ages, it requires higher voltages to run within spec. So they program the voltage at the level needed for end-of-life conditions, not out-of-the-box capabilities. Now, considering the fact that these chips can actually run at MUCH lower voltages early in their lifespan, that might say something about the relative aging rate over the three-year warranty period. That's the missing nugget of info that many people aren't taking into consideration in regards to the comparatively-high voltages for 7nm. Stop and think about the voltage deltas between 14/12nm GloFo-produced chips for Ryzen and what you see with 7nm TSMC. There's a reason for this.

 

[PaulAlcorn]

Many whiners here complaining about nothing and missing the point. I found this article very informative. The voltage delta between boost and fixed clocks is huge. The performance differential between boost and fixed at near identical clocks is also interesting. I would definitely want to know that, if I owned a Ryzen 3000, or was planning to buy one.

Good point. AMD's adaptive boosting algorithms can adjust thresholds at a 1ms granularity, but there is still some performance lost in shifting between power states.

 

[drea.drechsler]

1.49V Vcore under load with stock settings does seem unusually high. I was under the impression that with Ryzen you may see it spike to high 1.4x sporadically under light/bursty load, but it would be less under sustained heavy load.

The above quote you provided provides the answer to that as the community update (concerning the ABBA firmware update) goes on to say:

"Please keep in mind, however, that this firmware change is not a cap. The processor must still be free to boost if active workload(s) seriously require it, so you should still expect occasions where the processor will explore its designed and tested voltage range of 0.2V to 1.5V. "

So you should expect to see voltage spikes up to not only 1.49, but 1.5V.

 

[TJ Hooker]

Only thing I complained about really was that using a fixed clock speed that is at the same speed as their reported boost of 4.4ghz you can use 0.25v less, have more performance and less heat on ALL cores! The entire article is about that, have another look. I even complimented that the cpu is 3600mhz on the box and runs at 4200mhz all cores without even changing anything in the bios

I agree, very interesting result. And surprising, as it seems like everything I've seen previously about AMD's PB/XFR for Ryzen 2K/3K had led me to believe you're usually better off leaving settings at default (or using PBO) rather than trying to manually overclock.

For your results comparing static clocks/voltage vs auto behavior, was that testing done under LN2, or the 360mm water cooler?

 

[drea.drechsler]

I agree, very interesting result. And surprising, as it seems like everything I've seen previously about AMD's PB/XFR for Ryzen 2K/3K had led me to believe you're usually better off leaving settings at default (or using PBO) rather than trying to manually overclock.
...

In my experience, and the way I've read those recommendations: you're better off because you gain so very little performance (with a manual overclock) and it takes a lot of effort to achieve it (superior cooling, time and bother testing for stability, etc.).

A 4.2Ghz manual overclock with auto voltage setting won't give me a better CB20 score MT... and a much worse ST score. ST is important because that's the best proxy benchmark I know of to help me predict performance in the bursty environment of playing heavily threaded games. That's where most of the threads are lightly loaded and bursty and so can regularly boost to 4.4Ghz as I've observed on my system. The main thread(s), more heavily loaded, hover around 4.3-4.375 Ghz.

The problem with that behaviour is it's almost impossible to benchmark adequately, or describe. Ryzen 3000 performance is way to complex and hard to evaluate.

 

[TJ Hooker]

@drea.drechsler that was my impression as well, i.e. that with a manual OC you gain only a small amount of multi threaded performance while sacrificing single threaded performance. But the results in this article seem to contradict that.

Also, I'll say that the OC results in this article (i.e. 4.4 GHz at only 1.25V) seem exceptional compared to what I've seen elsewhere. Splave himself said that he wasn't able to break 4.1 GHz without sub-ambient cooling on when the processors first came out. Either the 7nm process has improved significantly in only a few months, or the chip in this article would seem to be a golden sample.

If the latter is the case, I kinda wonder if that has any ramifications on the extent to which the conclusions of this article can be generalized to all Ryzen 3K chips.

Our resident overclocking expert Allen 'Splave' Golibersuch has also spent time with early Ryzen 3000 samples and was unable to break the 4.1 GHz barrier without sub-ambient cooling.

https://www.tomshardware.com/reviews/ryzen-9-3900x-7-3700x-review,6214-4.html

 

[NightHawkRMX]

https://imgur.com/Zb8MMJd

View: https://imgur.com/Zb8MMJd

 

[drea.drechsler]

@TJHooker: I'm not actually sure I'd call that a 'golden sample' if it still can't get boosts to 4.4Ghz after ABBA BIOS updates and latest chipset drivers. Either that or mine is an ultra-golden sample and I don't believe that either. I bought mine from the second lot of 3700X's my Microcenter received so I believe it to be earliest silicon. (I should ferret the date code to know i guess).

One thing though: I merrily encourage explorations as close 1.5V as I can. I've dialed in a small offset (+0.0125) to make sure spikes are peeking above 1.475V in fact (the next step, +0.0250, saw occasional spikes up to 1.51V, too much to make Robert Halleck happy I'd venture to say). I have to think that helps boosting, it's certainly not hurt CB scores, and temps are well controlled.

 

[TJ Hooker]

I'm not actually sure I'd call that a 'golden sample' if it still can't get boosts to 4.4Ghz after ABBA BIOS updates and latest chipset drivers.

Given that boost behavior seems to be a bit of a crapshoot that can seemingly depend on all kinds of things other than silicon quality (including load, and we've already established that cinebench as used in this article is not what AMD recommends to see max boost), I'm not sure if we can draw a direct correlation between max stock boost clocks and silicon quality/overclock-ability.

All I know is that when looking at Ryzen 3K reviews, most of them couldn't get 4.4 GHz on all cores even going up to 1.4-1.45V, let alone 1.25V.

 

[drea.drechsler]

Given that boost behavior seems to be a bit of a crapshoot that can seemingly depend on all kinds of things other than silicon quality (including load, and we've already established that cinebench as used in this article is not what AMD recommends to see max boost), I'm not sure if we can draw a direct correlation between max boost clocks and silicon quality/overclock-ability.

That's exactly right... CB isn't good for it. I think it's because even in a ST mode it's putting a heavy load on the one thread and the processor won't boost high for that but a tiny bit, at very first of the bench mark.

What I've found is a good way is to run a heavily multithreaded game with HWInfo running in background. A game doesn't really load all it's threads heavily, and the lightly loaded threads seem to be quite bursty in nature as the scene changes, so there's ample opportunity for some of them to boost up to 4.4Ghz. I've observed this in Doom, BF1 and Shadow of the Tomb Raider.

I adjust HWInfo as Derbaur suggests: turn off monitoring of everything but the essential things to minimize impact and adjust the strobe to 500mS and no faster. I monitor SVI2 Vcore, regular VCore and the core multiplier for each of the 8 cores and plot them on screen. Play the game a bit then alt-tab back to it and look at all the boosts to 4.4Ghz on three cores, way more in the 4.3-4.35 range across every core. My gold star core is holding 4.25-4.3 Ghz almost steadily; that would be the heaviest loaded thread. And yet there is plenty of idle time (3.6Ghz which is probably actually deep sleep state) in between the boosts of the boosting cores.

This seems to me text book example of the what I've read from AMD of how it should work. The Windows scheduler is even cooperationg in that it puts the heaviest thread on the gold star core, the one that can hold the heaviest load with least thermal impact. Which suggest to me being gold star core doesn't mean it can boost highest: it means it can carry the heaviest load without contributing to CPU thermals and dragging down achievable clocks.

That's not a benchmark though, hard to say it's actually more beneficial than some all-core overclock. But it does seem to illustrate how the boosting behavior should work.

 

[JayGau]

Did anyone actually try it?

I did yesterday. I locked the CPU at 4.4GHz and put 1.25V manually and the computer never wanted to boot. I even got a POST loop until I manually turn off everything and reset the BIO, hopping that my machine was still alive (it was).

I did some more research and there is no way you can make this chip boot and be stable with all cores at 4.4GHz and 1.25V it's just not possible:

https://www.pcgamesn.com/amd/ryzen-7-3700x-overclocking-benchmarks-performance

 

[drea.drechsler]

Did anyone actually try it?

I did yesterday. I locked the CPU at 4.4GHz and put 1.25V manually and the computer never wanted to boot. I even got a POST loop until I manually turn off everything and reset the BIO, hopping that my machine was still alive (it was).

I did some more research and there is no way you can make this chip boot and be stable with all cores at 4.4GHz and 1.25V it's just not possible:

https://www.pcgamesn.com/amd/ryzen-7-3700x-overclocking-benchmarks-performance

I'm curious how VCore is being measured, among other things.

My results were much closer to PCGamesn. With my results at 4.35Ghz (bootloop like yours) I felt it hopeless to try 4.4Ghz. I chickened out at a fixed 1.45V for a manual overclock to be stable at Prime95, AVX off.

Either there is some hellatious variance in silicon quality or something else is going on.