News Survey: Only 5.6 Percent of Ryzen 9 3900X Hit Advertised Speeds, Most Other Models Suffer, Too

[FunSurfer]

Actually those two cases have nothing in common. The first case was real, you could verify that the 500mb of ram was different from the rest. The second case is based on one flawed survey then parroted by this hack of a website that is circling the drain.

Actually these cases are very similar: In the first case some part of a VRAM does not reach advertised pre-launch clocks and in the second case some part of a CPU does not reach pre-launch advertised clocks.

 

[OMGPWNTIME]

It's interesting that AMD recommends HWInfo over say Ryzen Master for monitoring clocks.

That said, with my 3800x the highest boost clock it has ever hit is 4415. Also I've noticed whenever trying to run a ST benchmark, it always loads 2 cores which I'm wondering if that's part of the bug that they reportedly identified. It's definitely not temp related as it never breaks 70c under MT loads, and barely breaks 60c under ST. Highest boost with MT isn't bad at ~4250.

 

[jdwii]

Im not really sure how AMD will fix this issue. These chips rarily can hit rated boost on 1 core even when overclocked.

Well with my chip 3 of my cores always hit 4375mhz quite often and with AGESA 1.0.0.2 my chip hit 4.4ghz.

Something Amd changed during that time ruined their turbo speeds. Reviewers also said this in regards to the 3900X so in time i can see this getting fixed as long as Amd keeps their word.

 

[PaulAlcorn]

An FYI guys, AMD has publicly acknowledged the issue. https://www.tomshardware.com/news/amd-ryzen-3000-boost-frequency-bios-fix,40308.html

 

[jdwii]

An FYI guys, AMD has publicly acknowledged the issue. https://www.tomshardware.com/news/amd-ryzen-3000-boost-frequency-bios-fix,40308.html

Thanks to you guys and der8auer for real if it wasn't for the community including you guys who knows what these companies would do.

 

[PaulAlcorn]

Is this really how Tom's Hardware gets their 'News' stories now? Some first class nonsense... Watch a five minute Youtube video and write an article...This site hits a new all time low with every passing day...

Yes, we watched the video and reported on it: that is the essence of reporting. Der8auer did a great job of applying a scientific method to the results analysis and had a large enough sample size to lend the survey some credibility. Perhaps you should watch the video and see why we thought it meets a standard that is worth reporting on.

That said, it wasn't a perfect process, but it does show what we've seen in our own testing and in the thousands, yes, thousands, of complaints that we as a team have seen on reddit, forums, twitter, etc, that customers aren't receiving the rated boost clocks.

We are here to serve the enthusiast community. As an extension of that, we are here to report on issues that are important to that community. If we are in a position to do so, we are here to stand up for the community.

Provided the efforts of others in our sphere are worthy of discussing, we have an obligation to the readers to do so. Without our efforts in our investigation, and the efforts of Der8auer, and others that have also investigated, do you think that AMD would have come out and admitted fault? After two months of silence? This is what reporting is about, and what it should be.

You won't find me hanging my head in shame over the fact that this article is on the site. Quite the opposite.

 

[rhysiam]

Long post from me here, but I personally think this is an important issue and we as (hopefully!) informed enthusiasts should be debating and discussing this. I'd welcome further discussion/debate. Here's my perspective FWIW.

I'm honestly surprised how many people seem willing to waive this away as fake news and give AMD a free pass here. Der8auer discussed his method with AMD and they explicitly recommended he use CB R15 ST for the load and HWInfo for frequency monitoring. He checked every result for the AGESA versions and requested people ensure they use Win10 with the latest updates. The end result?: Under the specific conditions suggested by AMD, only a minority of CPUs were able to achieve (not sustain, just hit for a single polling period!) the frequency that's plastered all over the marketing material and packaging. That's not good for consumers. If that goes unchallenged I believe we'd see boost frequencies become a meaningless marketing label with a few CPU generations.

As some have pointed out, this isn't a perfect survey... but no survey is! The nearly three thousand responses with recruitment focused on Der8auer's enthusiast audience and combined with careful data cleansing make this a solid set of data. Plus, we're not talking about small numbers of CPUs unable to hit boost frequencies. In that case it could be reasonably attributed to erroneous, faked or malicious entries, or hot weather, or any of the other possibly explanations. Here's the breakdown for valid results from the video:
- R5 3600: 272 of 542 (50.2%) <4.2Ghz
- R5 3600X: 163 of 180 (90.6%) <4.4Ghz
- R7 3700X: 886 of 1039) (85.3%) <4.4Ghz
- R7 3800X: 107 of 146 (73.3%) <4.5Ghz
- R9 3900X: 636 of 674 (94.4%) < 4.6Ghz
Anyone dismissing those numbers has to adopt some pretty conspiratorial thinking IMHO.

Two more criticisms I'd like to address - again this is my perspectives for the sake of the discussion:

1. But Intel can't hit boost frequencies either: That might be true under some workloads, however most Z series boards support MCE or equivalent. Flip that switch in the BIOS and - provided you have sufficient cooling - you get all cores running at the max boost clock. Yes it can be hard to cool, but the bottom line is, every individual core can run stable at the rated boost frequency at 24/7 voltages. AVX workloads cause issues, and perhaps there's a debate to be had about how all that should be handled, but those are still niche workloads for consumer CPUs. AVX aside, Intel CPUs can demonstrably hit rated boost clocks under typical workloads. That doesn't seem to be the case with 3rd Gen Ryzen
2. But lots of results are within 25mhz of the rated frequency: True, buts lots are substantially lower again. If this were some randomly chosen workload I personally wouldn't quibble about 25Mhz (as long as the vast majority of CPUs actually got that close... and they don't). The problem here is that AMD chose the workload. If the CPU manufacturer tells their consumer to test boost clocks using a specific OS, running specific updates, with specific BIOS & AGESA versions, running a specific workload and measured using a specific tool -> I expect that CPU to hit its rated clocks. They don't.

The frustrating thing is - and Der8auer makes point really well - AMD doesn't need to do this. 3rd Gen Ryzen is fantastic. Why mislead?

My thanks to Gamers Nexus, Hardware Unboxed, Tom's Hardware, Der8auer and anyone else who picked this up. AMD have been forced to respond and you can bet they - and Intel by extension - will be much more careful about boost clocks labels in future releases. IMHO this is a perfect case study for why a free press is so important!

 

[bit_user]

Because the power draw is quite out of proportion compared to the improvement in performance as in much greater.

Actually, it's not. It's only about 36% more power, but hyperthreading typically delivers gains even bigger than that. Depends on the workload, of course.

That said, I wouldn't be at all surprised to see HT use a little more power than the performance gain.

 

[bit_user]

You mention you see your Ryzen 3000s hit advertised boost clocks under low utilization, but that seems like a catch 22 to me: if you're not loading the CPU then you don't really need it to boost in the first place.

It's not, if you think of utilization as what's actually burning power inside the CPU. As you'll know, not all instructions require the same amount of power, with the widest floating-point vector instructions using the most.

Likewise, not all threads have the same amount of potential for instruction-level parallelism. For instance, text file parsing (e.g. Javascript, HTML, C++, etc.) is often very serial, being a mess of branches and data-dependencies. That doesn't mean you don't want it to be fast, however. So, a clock boost when much of the core is idle can still provide benefits in plenty of situations.

 

[bit_user]

This time around i ran each individual core for about 15min in my prime program and it required around 40% of one core.

40% of one core? What do you mean by that? How did you measure it?

 

[bit_user]

Actually these cases are very similar: In the first case some part of a VRAM does not reach advertised pre-launch clocks

No. The GTX 970 issue was more than that. It had to do with the datapath actually being much narrower, for the last 0.5 GB.

https://en.wikipedia.org/wiki/GeForce_900_series#GTX_970_hardware_specifications

 

[bit_user]

1. But Intel can't hit boost frequencies either: That might be true under some workloads, however most Z series boards support MCE or equivalent. Flip that switch in the BIOS and - provided you have sufficient cooling - you get all cores running at the max boost clock. Yes it can be hard to cool, but the bottom line is, every individual core can run stable at the rated boost frequency at 24/7 voltages. AVX workloads cause issues, and perhaps there's a debate to be had about how all that should be handled, but those are still niche workloads for consumer CPUs. AVX aside, Intel CPUs can demonstrably hit rated boost clocks under typical workloads. That doesn't seem to be the case with 3rd Gen Ryzen

You've got to admit that it is sketchy how Intel still handles AVX, with respect to their advertised turbo frequency.

Also, I'm really not sure that it's as much of a corner case as you suggest. Probably most video & graphics tools use AVX, as well as quite a lot of games. Even web browsers are probably using AVX-optimized rendering & compositing engines, when they're not using the GPU.

That said, I wouldn't lose sleep over AVX-heavy workloads dropping a 200 MHz, on an Intel CPU. But, if we're being even-handed, then we should probably also take Intel to task for their advertising of turbo clocks.

 

[NightHawkRMX]

I'm honestly surprised how many people seem willing to waive this away as fake news and give AMD a free pass here.

I don't feel thats what happened.

I have heard several threads on ryzen 3000 not hitting rated boost speeds.

I have also seen other postings on other websites about this issue.

If nobody really was talking about the news, i dont feel amd would have adressed this situation specifically and would make an effort to fix it.

 

[jimmysmitty]

You've got to admit that it is sketchy how Intel still handles AVX, with respect to their advertised turbo frequency.

Also, I'm really not sure that it's as much of a corner case as you suggest. Probably most video & graphics tools use AVX, as well as quite a lot of games. Even web browsers are probably using AVX-optimized rendering & compositing engines, when they're not using the GPU.

That said, I wouldn't lose sleep over AVX-heavy workloads dropping a 200 MHz, on an Intel CPU. But, if we're being even-handed, then we should probably also take Intel to task for their advertising of turbo clocks.

I would agree so long as Intel has not specifically stated their max turbos workloads. So long as they acknowledge that certain workloads may not be able to hit max turbo I see no problem with it.

 

[jdwii]

No. The GTX 970 issue was more than that. It had to do with the datapath actually being much narrower, for the last 0.5 GB.

https://en.wikipedia.org/wiki/GeForce_900_series#GTX_970_hardware_specifications

To this day i have no idea why others are bringing up this its all over the web this is 100% irrelevant to this discussion.

Anyways i measured CPU usage per core using hwinfo

 

[jimmysmitty]

To this day i have no idea why others are bringing up this its all over the web this is 100% irrelevant to this discussion.

Anyways i measured CPU usage per core using hwinfo

Its a false equivalence . People like to use it as some way to justify what issues AMD might have or try to make them not seem as bad. In my eyes they are both issues and need to be fixed by their respective manufactures. So long as they fix the issue who cares.

They forget that nVidia settled the lawsuit for the 970 and paid people who owned one, much like the AMD FX lawsuit was nothing really.

 

[geogan]

I find HWInfo doesn't appear to be working correctly, unless I am doing something wrong - it's not showing the correct core clock values at all.

The "Minimum" voltages never change on mine and show around 3400 MHz for all 12 cores all the time, even though I can see in Ryzen Master that ten cores are asleep and the two doing something are only at less than 500 MHz.

So if it can't show minimum properly, then what confidence do you have in it showing maximum correctly either.... not much to me.

Screenshot of HWInfo & Ryzen Master running

 

[rhysiam]

You've got to admit that it is sketchy how Intel still handles AVX, with respect to their advertised turbo frequency.

Also, I'm really not sure that it's as much of a corner case as you suggest. Probably most video & graphics tools use AVX, as well as quite a lot of games. Even web browsers are probably using AVX-optimized rendering & compositing engines, when they're not using the GPU.

That said, I wouldn't lose sleep over AVX-heavy workloads dropping a 200 MHz, on an Intel CPU. But, if we're being even-handed, then we should probably also take Intel to task for their advertising of turbo clocks.

Sure, I absolutely agree Intel should be much more transparent about boost frequencies under AVX workloads. That said, it's not the same thing. On Intel at present, there are specific workloads (i.e. AVX) where boost frequencies won't reach advertised rates. For AMD, what this survey suggested, was that despite using the very specific workload recommended by AMD, most CPUs were unable to hit their boost frequencies at all - or at least for long enough for the reporting software to measure it.

Absolutely let's keep pressure on Intel to be more transparent about boost frequencies, but I don't think that argument should undercut the significance of the issues Ryzen 3 were having. Thankfully, it looks like AMD has a solid plan to address the issue which should be available through BIOS updates within a few weeks. That's the right outcome here.

As an aside, it will be interesting to see how this plays out as boost CPU boost and power management features become increasingly complex. I expect future Intel CPUs to have similarly complex behaviours where boost frequencies are far more workload dependent. ST performance still matters a lot, but gains are increasingly hard to come by, so it's logical to expect manufacturers to work hard to extract every bit of ST performance they can from their chips.

I don't feel thats what happened.

I agree it's been widely reported - which is good. I was more surprised with the number of responses in this thread which just disregarded the survey entirely or waved it away for whatever reason.

 

[bit_user]

For AMD, what this survey suggested, was that despite using the very specific workload recommended by AMD ... I was more surprised with the number of responses in this thread which just disregarded the survey entirely or waved it away for whatever reason.

Just to reiterate, I wasn't initially aware that that AMD had made boost claims about this specific benchmark. So, I agree that's pretty bad.

I'm sure some others in this thread were also unaware.