AMD's Ryzen CPUs (Ryzen/TR/Epyc) & Vega/Polaris/Navi GPUs

It seems like many Ryzen 3000 users are not reaching the advertised boost clocks for their CPU.



There seem to be a lot of factors in how far it boosts (motherboard, BIOS/AGESA version etc), but from what I can see the fault lies with AMD's advertising. Whilst the CPUs have great performance, it seems like AMD's marketing team have not been entirely truthful with their stated boost clocks. Its a shame as AMD's crappy advertising team is what is letting them down at the moment. Their engineers are doing a great job, but they might end up suffering financially because of these poor advertising choices. Intel and Nvidia are also guilty of scummy advertising choices, but that doesn't usually end up influencing important component specs like boost clock etc (although Intel's TDP specs are borderline). AMD really needs to get them dragged back into line to stop them suffering, all they are doing is giving their competitors ammunition to use against them.

 

I just want to grab these whingers and shout in their face: What's more important? Frequency, or work done? If you want a 5ghz CPU instead of one that gets >200cb R15 single core go buy a FX9590!

Roman missed an absolutely critical data point in his analysis. What was the performance score for each result. If there is no correlation between reported max boost and the end single core score, that suggests the hwinfo data is inherently unreliable which renders invalid any conclusion based on it.

Benchmarks in reviews are done in real world usage scenarios. If you buy something based on a specification before there's real world performance data, that's the risk you take.

So you can infer that I think the whole "Bulldozer core" lawsuit was a storm in a teacup where a legal issue was yet again decided on marketing principles as some class action troll lawfirm extracted some shut-up-go-away money from AMD.

Someone who bought a FX because it was "8 core", and not on its performance, based their purchase decision on the wrong type of information and somehow that defective logic entitles them to receive a partial refund where someone who bought eyes open would not.

(I am aware the settlement is for all FX buyers, I'm referring to the technical argument in the class action which was the impact of misleading advertising upon the subjective mind of consumers who claim to have been misled, which would not have been all FX buyers, although it is easy to dishonestly claim as such to meet the conditions of joining the class)

 

But the variances in boosts are likely down to BIOS/AGESA which can be usually tweaked.
AMD probably cannot guarantee those boosts on all CPU's due to a wide variety of OEM motherboards out there. AMD only releases AGESA... how its implemented into the BIOS of the motherboard is entirely up to the OEM (not AMD).
Plus, there's cooling to be taken into account... application of the thermal paste, etc.
This in turn can have unintended consequences for boosts.

So, can it be consistently proved its AMD's fault... or is it more directly traceable to the OEM's and how they implement the Agesa in their BIOS (and possibly the users in how they configure their system)?

Furthermore, the performance differences are negligible due to such small variances in how high a CPU boosts.

EDIT: This is not to say that advertising shouldn't be accurate... it SHOULD. Too many people have been burned by bad or disingenuous adverts... however, I also don't think that given a situation, people take the time to take into account many different perspectives such as independent testing.

Besides, Nvidia and Intel have gotten away with much worse. Latching onto AMD for what may or may not be entirely their fault (and minor one in comparison) seems like people are looking for 'something' to complain. And while technically accurate, the question is, is it actually AMD's fault, or it this a case where the ad should be used as a guideline only and wait for third party reviews?

Plus, its an early 7nm product. Wait for 7nm EUV and you'll likely see some heavy improvements (because comparing this to Intel's well matured 14nm node where hitting 5GhZ boost is relatively easy seems really disingenuous - plus the IPC gains on Zen 2 mostly negate Intel's Ghz advantage).

 

Its harder to test for the performance score since everyone's system is different and has different RAM speeds, different processes running in the background etc which will all impact the score. I don't think anyone is complaining about the performance of Ryzen 3000 though, that is clearly not an issue. AMD should be using performance scores to advertise Ryzen since that is what it does well in, rather than falsely hype up the clock speeds. They don't need to since the IPC improvements are good enough to make up for Intel's lead in that area.

 

As I said, its not only AMD's fault that some users aren't hitting the advertised boost clocks, as this video shows it is also down to the board manufacturers. Saying that though, I think it still lies on AMD for being perhaps a little too generous with their stated boost clocks. You can argue that if you buy a CPU you should expect it to do what it says on the box, though technically it is still legal advertising as long as they have a disclaimer saying that your CPU will not hit that speed in certain situations.

Yes, you are right that the immature silicon process is the main reason why clock speeds are lower than Intel, this will improve over time of course.

I would like to think i can hold AMD to a higher standard, as in the past their advertising has been more honest than the competition. At the end of the day though, they are still a big corporation and they will do what they have to do to make money. My main point is that they need to be careful and avoid bad publicity, since their position is still relatively fragile compared to Intel and Nvidia.

 

AMD is indeed a big corporation and I pretty much always said so.
However, the 'blurp' in AMD advertising by choosing higher boost clocks as a basis for advertising could also be attributed to their GPU's too, but I don't see people making legal (or massive) fusses over those.
Both NV and Intel do the same by quoting best possible results (but people always know to take this with a grain of salt).

So, how is this any different?
AMD has no control over how OEM's make their own motherboards, and if those motherboards have issues in implementing AMD's AGESA adequately which can result in problems with boost clocks, then its the OEM's who are to blame.
I don't think its AMD's job to fix OEM's mistakes.

AMD did say that performance should remain unaffected between B450 and say X570 series mobos... and realistically, there really is no distinction performance-wise.
If a problem did happen, then its quite possible its an OEM issue (not an AMD one).

AMD did remain close to their advertising recently, and I see this scenario as 'an acceptable margin of error which is expected due to differences with mobo OEM's and did NOT impact performance to a perceptible percentage that would be useful/relevant/visible to the user.'
The margin of error is same as with getting different scores in 3d Mark or even Cinebench every time you run them. No meaningful difference

 

AMD thinks they got a fix...

AMD Ryzen‏ Verified account @AMDRyzen
8:00 AM - 3 Sep 2019

https://twitter.com/AMDRyzen/status/1168901636162539536
https://www.reddit.com/r/Amd/comments/cz620f/amd_releases_statement_regarding_the_recent_ryzen/

 

If the 3900x gets to 4.6 GHz consistently and the 3950x to 4.7 then it is another nail in the competitions coffin. Especially if the fix is in before the new TR's are out.

 

Hopefully the fix does the job, i'm glad to see they are addressing this properly rather than sweeping it under the rug.

 

Another bunch of x570 boards tested:

Budget X570 VRM Thermal Performance, A Must Watch For Potential Buyers!
Hardware Unboxed
Published on Sep 4, 2019

 

Rumors;

 

AdoredTV weighs in on Faildozer and AMD boost issues using a number of industry references starting at 06:42

It's been a bad week at AMD
AdoredTV
Published on Sep 6, 2019
Marketing foot in mouth disease.


Pre-release testing results showed higher boost results, similar to the best results from motherboards with the best results now, so a firmware update - some apparently requiring vendor tuning as well to have slower motherboards pick up the slack - should bring back the advertised boosts, but it might take a while for all of the vendors to catch up.

 

Yeah, Ti States (temperature inversion) evidently has a bit to do with AMD tech boosting. I'll comment more potentially later. But it is adjustments to those algos that is happening.

 

https://www.tomshardware.com/news/amd-ryzen-3000-boost-clock-controversy-intel-attack,40231.html

Too Hot to Last? Investigating Intel's Claims About Ryzen Reliability
by Paul Alcorn September 6, 2019 at 1:00 PM

The graphs where they turn off the cooling and observe single thread boost reveal a temperature dependent scaling that commences in the region of 50C, not unlike Nvidia's turboboost 2.0 with Pascal.

 

Tuning like this is based on a small number of factors - you only have so many knobs to turn. I doubt it had anything to do with longevity, and Intel trying to play on people's minds this way shows just what kind of jerks they are - which Intel have such a consistency of personality like this over the years is another reason I have given up on Intel long ago.

"It's also possible that AMD is merely tuning its boost algorithms to provide a more targeted range of effective performance, and these alterations have nothing to do with reliability metrics."

Exactly.

If it's the temperature tuning knob that AMD goes to for their fix then AMD will need to give us the choice to sacrifice boost for temperature - that Asus guys comments about a customizable setting - but again I think that Asus guys personal comment was a guess and may even be a translation issue in that he meant the more likely "stability" reason.

Either way I don't see any of this as cause to purchase Intel CPU's instead of AMD. Intel CPU's are boosting at temps over 105c - and Intel is saying it's just fine - so if I was going to be a worry wort about longevity vs temperature then Intel's 105c+ limit is far more worrying than AMD's 80c limit.

 

Spin, spin, spin... sigh...

If Intel and AMD are using silicone from the same civilization (Earth's), then AMD's 25C 'buffer' seems either like overkill or just plain bad design vs. what Intel guarantees their CPU's for.

Can't have your cake and eat it too.

 

Well yes and no, you have to remember the smaller tracings are affected differently. As a lay term if say through heat 2nm of a trace is hit on 14nm then 12nm could still be viable, at 7nm only 5 would be left viable. do the math and you can see where this could make a huge difference. This goes to other variables as well as far as distance between traces and components etc. etc., etc..

 

You are ignorant, so let me enlighten you. The temps are read from different embedded sensors. The Tctl of the chips is the temp of the hottest sensor. The Tdie is more closely like the avg temp. So, that 25C buffer relates to the average temp while the hot spot is 25C higher than the average in many cases. As such, the silicon is operating the same as Intel, just what is being read and reported is different. This makes sense due to different sensor data being used for boost algorithms and the like.

But because you are unwilling to learn what and why, just keep spouting your ignorance, after all not knowing something has never stopped you from spouting up to date. JFO.

 

Your continuously claiming to know my ignorance only shows yours each time you proclaim it.

Your explanation shows AMD as ignorant as you seem. They really can't add 25C to their estimate?

This constant equating of AMD to Intel when it's not, is tiring, at best.

Let's see if stands on its own without the self professed AMD twins rah rah'ing here each time it's short comings are mentioned.

 

AMD and Intel cores are different, and so it's safe to assume they measure temperature at different locations with different thermal densities. Comparing a degree Celsius number between cores from different manufacturer is not only pointless but it makes no sense. All that matters is whether or not the cores manage temperatures in a way that delivers maximum performance while also meeting reliability targets.

With that in mind, it is safe to say that Zen 2 wasn't ready for release. Much like Radeon GPUs since forever, Zen 2 has shipped with beta software. It is absolutely ridiculous how AMD continue to release products this way, and it is insane that this has now spread to their CPUs as well.

 

I guess you haven't been doing this very long because firmware errata and updates are common from both Intel and AMD and have been for many years. Even with Intel not innovating for so many years there were firmware changes and updates - and new steppings of CPU's.

That's not insane, that's reality. Things rarely come out perfect, and for coding is no different.

I'd rather vendors push ahead with new technology, architecture, processes and I'm willing to put up with updates to optimize performance.

If you want to talk insane, look at Intel's architectural failure with security. That's not only insane, it's reality, and it's getting worse before it gets resolved. Maybe Intel will resolve the security with new architecture in their 7nm era, maybe not.

AMD is outperforming Intel even if some Ryzen 3 CPU's are a few percentage points off their full boost target in some environments, maybe AMD will trim that boost gap further with firmware updates - giving AMD even more headroom against Intel.

Keep complaining to AMD to fix those boost issues guys, that'll make the AMD Ryzen 3 CPU's even faster.

 

https://edge.seas.harvard.edu/files/zu16tistates.pdf
An article discussing, in part, what AMD is telling to accomplish with their boost algo.

 

There is no excusing this release. It has been a mess that is way to reminiscent of their GPU releases in the past. Stellar hardware that is made to look worse by awful software. They need to learn from it.

Questioning how long I’ve been around and bringing up Intel adds nothing to this discussion at all. Why even go there?

Somehow I don’t think this is going to be “accomplished” at current boost voltages of 1.5V+. They definitely are not going for this.

 

"No excusing"? "Mess"? "Awful software"?

Stop catastrophising! The product works and performs as intended. It is more than sufficient to compete, and the market is responding and buying it in droves.

If AMD and/or TSMC doesn't have the confidence in the 7nm process yet, it is a good thing that they are being conservative, and not risking product failures. That would be enormously damaging to the entire range of CPU products (especially EPYC) that underpin AMD's financial survival, if this twitstorm is what a temporarily missing couple hundred MHz generates.

 

No, it’s not working as intended or as advertised. That’s the point. Stop trying to handwave and say it’s ok, because it’s not. AMD don’t need you sticking up for them. They know they messed up, and are fixing it. Considering the timing of their statement, it is probably community outrage that has led to the fix.

Stop getting so defensive. Zen 2 is great but it needs fixing, and AMD obviously agree.

 

Because getting all upset about a few mhz shortfall in some builds is abnormal, if you had been doing this as long as many of us have you wouldn't be getting all indignant about it.

Be calm and let AMD work out any issues with vendors firmware especially short on boost making.

It is merely something that needs to be reported back to AMD so AMD can look into vendors with motherboards that aren't delivering full boost.

AMD has been aware of this since the beginning of the release as users report problems with their motherboards, by AMD's report they have been working on this for a couple of months.

It's a surprise to you and many others that don't have the problems - either because they haven't noticed it, don't have the problem with their own hardware (50%+), or don't own the hardware themselves so can't give any useful input to AMD.

If you have the hardware and are affected then gather up the measurements that show your CPU is short and with your motherboard and related details make a report to AMD.

Making bluster in a forum threatening AMD isn't going to solve the problem. But we can help direct you to where to report it:

AMD Support Ticket Process
https://www.amd.com/en/support/contact

Online Service Request
Please use the form below to request technical assistance for an AMD product.
https://www.amd.com/en/support/contact-email-form

AMD Customer Support - 1 (877) 284-1566 (USA)
... click link for worldwide contact info

The best use of displeasure with hardware or software is to report the problem to the vendor. Even if they are already aware as AMD was it can help add additional scope to their response - like if you report a motherboard / CPU combo that hasn't been reported before. Also AMD builds a list of people to ask to test fixes before general release if that fits in your own situation.

Remember that all of these hardware and software products are a moving target, its not like general merchandise support, these products can be patched - fixed - without replacement most of the time.

Intel is going to have to replace all of their CPU's produced now and for the last dozen years to remove their "problem situation" from the world.

Intel's security patches in continuum are all reducing owners performance, where as AMD will release fixes to add performance, and there's a big difference in the problem each are facing right there.

AMD only needs to issue firmware updates be patient and let AMD and the vendors work to address the issue and roll out firmware patches to improve boost performance.

As far as I have seen the difference in boost is a small negligible amount and for the most part single thread boost is rarely used or held long when it comes up, so for most (all?) it's a non problem.

Hopefully the roll out of firmware updates will be done quickly and efficiently. Be sure and contact your vendor to encourage them to work with AMD to quickly issue an update for your motherboard, if you are directly affected.

 

Who am I defending lol. All I'm saying is it's a storm in a teacup. Peak readings are for marketing and trivia. What the CPU is bought for is the work done, area under the curve.

I believe community outrage has resulted only in PR, the "we have listened and engaged with the community" public acknowledgement. They were already aware and "fixing" it behind closed doors weeks ago.

 

I promise you I’m not upset. This has no effect on me. My original post was simply stating that the CPUs weren’t ready for release. And they certainly weren’t.

You can go on about how long you’ve been doing this for as long as you like, it doesn’t change the fact the CPUs weren’t ready, and that this isn’t “normal” for CPU (or GPU) releases. If Nvidia or Intel were having these issues they’d quite rightly be called out on it as well.

Work is a product of IPC * clock. If you don’t care about advertised clocks and what the CPU is actually doing that’s fine. I’d personally be annoyed, but that’s just me.

 

the cpu is on sale for 2 months now so its pretty ready. the problem comes with owning latest and greatest, become lab rat.

looking back i dont think intel's mitigation ever going to get fixed, nor that tiny issue of a prime number that came up just faded cause nobody cared. point is the issue is much smaller than what people claim it to be as those 1 core boost lasts like 0.001% of a 24hr usage it has literally no performance impact.

someone also mentioned in reddit the issue about 1.5v boost. people didnt like it cause they werent use to it but they claim it being latest technology and readings are refreshed much more frequently, as long as its not heavy load 1.5v should be of no problem.

 

It is not that it was not ready for release but that like all other releases Zen2 owners are first adopters. Almost all of those people then have to suffer through the learning stages in getting the systems smoothed out. It is the name of the game, all new designs suffer from.

 

You again show your inexperience with such things by suggesting that the AMD Ryzen 3 CPU's weren't ready. Similarly that this isn't "normal". Normal is a sliding scale of reality that varies over time. "New" things are re-seen repeatedly over and over in different guises. Tweaking / optimizing performance based on feedback from the field is quite normal.

AMD has been releasing CPU's for a very long time, and the Ryzen CPU's for 3 generations, along with desktop CPU's AMD has been releasing APU's, mobile CPU's, HEDT CPU's (ThreadRipper), and Epyc CPU's.

AMD had a good solid release on the hardware they had for testing as well as their vendors and they didn't seen any reason to hold back the CPU's, because there wasn't any reason to hold back the CPU's.

It might be layout issues with the motherboards, or power delivery, or RAM / Chipset issues, or any of a wide range of variations of environment differences that can affect the boost frequency.

Also the algorithm for boost, OC, power, and temperature are all much faster in execution than in the past and the measurement tools - including AMD's - while not at fault in their implementation - are just not fast enough or designed to register the peaks of boost.

Peak reading meters were one adaptation for signal measurements that were adopted for just such situations in the analog world. The meter needle was moving to fast for the refresh rate of vision so the peak reading couldn't be reliably read. The peak reading meter holds the highest reading through some visual hold hysteresis that accommodates human vision and reaction time. Perhaps AMD can incorporate this into their firmware and monitoring tools / api.

Peak Reading Meters
https://www.google.com/search?q=peak+reading+hold+meters
https://sonicscoop.com/2018/03/29/everything-need-know-audio-meteringand/2/

AMD did a good job on the release, has done a good job collecting relevant data, have likely been collecting additional hardware for in house testing - an ongoing exercise that you can't wait to complete before releases of this kind.

You're dead wrong if you think AMD should have waited longer before releasing. Think of the 100's of thousands of happy AMD CPU owners / users out there that would have been disappointed if the Ryzen 3 release didn't happen. Think of Intel forced to sell insecure overpriced CPU's gouging those same potentially happy AMD CPU buyers. The horror.

With any new release this is why I recommend not buying the first production runs of anything, unless you have a mature thick technical skin and must gain knowledge for potential future large roll out of said product. You can count on gotcha's being discovered after release, it happens with all software and hardware, that's just how it works.

You can't possibly test for the full dataset of all possible combinations of hardware and software configurations found in the field, multiplied by the experience level of the person(s) implementing using your product. It's not possible to 100% do that before a product release.

That's why there are patches and updates and user forums and revisions to products as they move through their lifecycle - see I do this kind of work and have for 40+ years - lifecycle management, problem products that can't quite make it out the door, product support for major issues in the field, it's a complex and important task with sometimes 100's of millions of $ on the line for the vendor and the customers.

If you can't stand the heat, stay out of the kitchen. If you keep cutting yourself badly with new products, back off from the cutting edge.

For myself I almost always wait for at least 2 generations of a product before I jump in. For clients of course that doesn't work, and is in fact I have to do totally the opposite - I'm always negotiating with vendors for pre-release hardware to help guide the release of new hardware my client might need to depend on. That's how you make sure to reduce the problems new products have at release if you need to depend on them immediately at first release.

It's a personal journey only you can manage well for yourself, you can't depend on products to be 100% perfect at release, among the myriad of things you will learn over the years.

 

Rumor/leak;
https://www.techradar.com/au/news/amd-ryzen-9-3950x-release-date-may-be-incoming
https://hothardware.com/news/amd-ryzen-9-3950x-16-core-cpu-september-30-release

 

25-75MhZ loss in boost (vs what's advertised) on a single core is hardly something to complain about, especially because this boost wouldn't have been 'sustained' anyway over a long period of time even if the issue wasn't happening.
The performance metrics remain unaffected because the original ones were measured with the same 'issue'.
Also, AMD already released a BIOS patch for this... but the performance 'gain' probably won't be much (if any).
In fact, the differential will likely be the same as scoring slightly higher or lower in synthetic benchmarks per each consecutive run (which is an acceptable margin of error the more you use the CPU).

Point remains that you're complaining about 1 - 2.5% loss in single core clock boost.
Its negligible (you won't notice it in gaming or in productivity)... and its well within an acceptable margin of error which allows for different cooling setups and how OEM's implement BIOS updates.
Furthermore, even with this 'flaw', AMD is still a better option than Intel... its more powerful in IPC, more efficient and surpasses it in multi-core performance while doing all of that at a lower price point.

Seriously... its a non-issue.
If it was causing large performance drops (in the range of over 5%), then sure, it would be seen as a problem... but since this isn't the case, then I doubt there's anything to really worry about.

If this new BIOS update fixes the issue in question, yeah sure... but I doubt you will actually care about that 1 FPS difference in games or a few seconds less in a render... because, that's exactly what's probably going to happen (but I could be wrong).

 

friedmpa 334 points 8 hours ago
"TLDR bios update in 2-3 weeks"

Longer waits likely, depending on the vendor's staffing levels vs product demands.

AMD tweeted info on the boost firmware update:
https://twitter.com/AMDRyzen/status/1171461886870577152

Ryzen Community Update: BIOS Updates for Boost and Idle, Plus a New SDK
Blog Post created by rhallock on Sep 10, 2019
https://community.amd.com/community...ios-updates-for-boost-and-idle-plus-a-new-sdk

" What to Expect Next
AGESA 1003ABBA has now been released to our motherboard partners. Now they will perform additional testing, QA, and implementation work on their specific hardware (versus our reference motherboard). Final BIOSes based on AGESA 1003ABBA will begin to arrive in approximately three weeks, depending on the testing time of your vendor and motherboard.

Going forward, we’ll continue providing updates in this format as the updates are being prepped for release. In the interim, please don’t hesitate to reach out our support page if you have questions or need help."

An Update on 3rd Gen AMD Ryzen Boost Frequencies
Submitted 8 hours ago by AMDOfficial Official AMD Account
https://www.reddit.com/r/Amd/comments/d2a4xs/an_update_on_3rd_gen_amd_ryzen_boost_frequencies/

It's very long and includes links in the text, and check out the comments from users trying the leaked BIOS. Remember the final BIOS from the vendor for your motherboard might show further improvements in boost and other characteristics.

For most reporting the leaked BIOS results it seems to be helping not only with single core boost but also with keeping the boost up and some are actually seeing modest / measureable / consistent CB multi-core increases. Here's one example, check out the reddit thread (sort by new as well):

Raisin_334 1 point 3 hours ago
"I'm hitting 4.575Ghz with ABBA on Aorus Master instead of 4.525-4.550. It's not 4.6 but I'm happy. Most of all this seems to have made a huge difference with the game launchers and iCue I'm running in the background. The boost algorithm seems much better for me now as I was never seeing better than 4.2-4.3 before. I'm now boosting to 4.5Ghz here or there. CPU-Z SC score is largely improved with background processes. I was getting 526 and I'm getting 543 with everything running. Cheers guys!"

Since that BIOS is a non-final beta firmware it's likely when he gets the final version from Gigabyte he might see that last 25mhz boost.

chapstickbomber 3900X stock C6H - 3733 C16tight - RVII 2150/1250 3 points 7 hours ago
"people thinking maximum means typical or sustained is not AMD's fault - they have repeatedly said how the chips work, and yet people are like "nuh uh, I don't like that, you have to change it to what I'm accustomed to" and AMD is just like "lol but we just said they don't work like that, you are mistaken" and then the screeching continues, even now despite a patch to improve the boost behavior, adding 1-2% performance in ST.

I love the new Ryzen 3000 boost behavior. 1ms boost latency utterly shames all other CPUs prior to now. And really, people have it totally backwards. The previous boost numbers were the real farce and marketing BS. If it takes dozens or hundreds of milliseconds to even start boosting, that just means small bursty tasks (most ST loads) basically get nothing out of it. Pointless, but observable in long synthetic tests. Ryzen3000's boost is actually practical, yet instead folks are losing their minds over a few MHz in a synthetic."

AMD BIOS Boost Fix Leaked and Tested! Ray Tracing Cores in NAVI?
Boot Sequence
Published on Sep 10, 2019
Let’s get started with AMD! today is September 10th and we got an update from AMD about the boost problem fix that many ryzen 3000 owners have been waiting for. The new agesa firmware called 1003 ABBA will fix a bug that reduced boost speeds of 3rd gen CPU's. AMD claims that the reduction was between 25 to 50Mhz but others saw different results. Anyways, Right now, this new firmware has been sent to motherboard vendors and should be available within 2 to 3 weeks directly through your motherboard's website.

Spoiler: More...

While the Firmware fix hasn’t been released officially by AMD to consumers yet, it seems like a leaked version of the fix has been making the rounds since yesterday. It was leaked on the Chiphell forum and while you can go download and install it for yourself , I would highly advise you to wait for an official version. You dont want to mess with your CPU’s microcode with a beta version.

In any case, toms hardware took it upon themselves to test it out and the results are... Kind of a mixed bag. They tested the new Agesa 1.0.0.3 ABBA on the 8core 3700x and the 12core 3900x. In their previous testing to see if those chips truly had an issue with boosting, toms hardware concluded that both those chips boosted really close to advertised speeds although it never actually reached it. In their case, it was only 25Mhz below advertised clocks but as we saw with Derbauer’s survey, some were as low as 250Mhz below.

Let’s get into the new results, this time with the leaked firmware. With this one, their Ryzen 3700X now reaches its advertised clock of 4.4ghz without any problems.its worth noting that their setup kept the cpu under 55 degree’s Celcius throughout their tests.

Now for the 3900x. In this one, the results are very different. Before, their chip would reach 4.575Ghz across the whole test on the old firmware. With this new one though, the cpu starts off great within the first few seconds. Even peaking above advertised clocks. But a few seconds in and it drops lower than it used to at 4.55Ghz. Moving even further into the test and the clocks drop again to 4.45Ghz and stay there for the remainder of the test. That’s 150Mhz below advertised clocks meaning the sustained load must cause the CPU to want to switch cores or something. You can even see the cores trading places across the test with the 3900x as indicated by the color change.

Now there is a chance that the leaked version is not the final one despite the fact that it is also Agesa 1003ABBA. The SMU version might be different but we won't know until motherboard vendors start to distribute the new bios. In any case, toms hardware’s testing does show there is still some weird behavior happening here but I think AMD did fix one of the problems. The CPU's will now likely reach the single-core boost clock but assigning the fastest core to the most demanding workload is still something that needs to be worked on for system with 2 chiplets. That’s why the 3700x boosted properly. With only one 8 core chiplet, its easier to keep the workloads managed. When you jump to 2 chiplets, than the ryzen scheduler needs to figure out where to assign the fastest core. At least that’s one problem down for AMD, with time optimization for systems with multiple chiplets should fix this issue.

 

wccftech published a well thought out and balanced response to the firmware update release from AMD, which brought out many aspects beyond the boost fix, which AMD has been working on over the last couple of months. I'll include the link and the closing paragraph, and please check out the article in full (it is short) if you haven't already read it:

AMD Announces BIOS Updates For Ryzen 3000 Series Boosting Performance And More
By Keith May, Sep 10
https://wccftech.com/amd-announces-bios-updates-for-ryzen-3000-series-boosting-performance-and-more/

"To wrap this one up, AMD has announced more than I expected today with the news of the new AGESA 1.0.0.3ABBA, new Idle Enhancements that will be coming with it, and the new Monitoring SDK. But, I think two of the most important notes came in the admission that previous AGESA updates were not to extend the lifespan of the parts and the timeline which to expect new BIOSes to become available from board partners. The timeline is important because we’ve already seen supposed leaked BIOSes, and subsequent tests completed off of those leaked BIOSes, so I would likely disregard those as not accurate results since AMD is just now releasing this AGESA today and that pretty much makes having gotten it several days ago an impossible task."

Whether the leaked BIOS is part of the development stream - an earlier version than the final released by AMD to vendors - or it's something else, we should wait for the "real thing" before complaining that the leaked BIOS isn't enough.

It is nice for those that can find benefit from it in the interim, but it's not the final build - and there will likely be ongoing tweaks and optimizations ongoing between now and Zen 3 / Ryzen 4.

AMD is intent on continually moving forward with optimization based on owner feedback besides their own internal testing, so please keep filing feedback with AMD.

Another Reddit thread tracking beta firmware results:

100MHZ increase with ABBA Beta Bios w/ performance to match! 3900X + Aorus Elite x570
Submitted 17 hours ago * by BaitForWenches
https://www.reddit.com/r/Amd/comments/d2hd82/100mhz_increase_with_abba_beta_bios_w_performance/

"I'm coming from the F4 Bios, Aorus Elite X570 to the new Beta Bios F5A ABBA.

Default Bios Settings with only XMP memory setting enabled.

https://i.imgur.com/4E9cHve.png - Old Bios - 4.516 MAX

https://i.imgur.com/lD8Ld3V.png - New Bios - 4.616 MAX

My performance is now right under a 9900k, when it was under an 8700k previously! (single core)

https://i.imgur.com/YXe33Se.png

Update: Disabled Speed Spectrum and Now I'm getting 4.625 MAX and CPU Z single core score is the exact same as 9900k at 543."

Polkfan 25 points 16 hours ago
"It's rare to actually get stuff like this fixed Amd did a good job on this update"

Koyomi_Arararagi3900X//Aorus Master//48 GB 3533C14//1080 Ti 8 points 13 hours ago*
https://imgur.com/a/w1AzpTI
"Also, disable spread spectrum so your bclk is 100 (99.98) instead of 99.80. Get that free 9 mhz"

That's a good point, for many simply syncing the BCLK to 100 instead of a smaller fraction will gain mhz and get you closer to the advertised boost. Some of this is down to motherboard makers BIOS and initial settings, that's why you see mention of other parameters that can affect the boost reading besides this firmware update - find them all for your motherboard BIOS and reap the benefits even before the patch comes through from your motherboard vendor.

Also, memory OC'ing has been tied to reduced boost. When going too far in the memory OC you can appear stable but get less performance in various metrics including maximum boost. You can try going back to default / XMP settings to see if you gain boost to get closer to advertised value. Other tuning parameters setup before hand to fine tune for your CPU can also help.

It's nice AMD is giving more performance, but check your own BIOS / tuning optimizations to make sure they are giving you the best CPU performance. Hopefully this hulabalo will bring out these tuning details specific to various motherboards BIOS and generic tuning benefits too to more people so they can tune their setups even better than before.

In that respect this can wake people up to their best settings / tuning outside of AMD's own tuning improvements if they put in the time to research the giant wave of knowledge coming out now.

zaggynl 3900X | RX 56 1 point 1 day ago
"Sounds good, I'll wait a bit until they are officially released.
Link for those looking for the Beta BIOSes:
https://forums.tweaktown.com/gigabyte/28441-gigabyte-beta-bios.html
Note: GIGABYTE is sharing these BIOSes for testing purposes only and are not meant for general release"

 

amd better get that memory latency reduced to under 55ns with the best ram, and up that infinity fabric frequency. also reduce latency between ccd/ccx and i want all of these on zen 3 with at least a 200mhz boost. pls thanks amd.

edit: oh and 10% more power efficient over 7nm and 10% IPC.. kinda too much to ask but more improvement the better

edit2: add optane dimm support while you're at it

 

You tell'em @ole!!!:

AMD Support Ticket Process
https://www.amd.com/en/support/contact

Online Service Request
Please use the form below to request technical assistance for an AMD product.
https://www.amd.com/en/support/contact-email-form

AMD Customer Support - 1 (877) 284-1566 (USA)
... click link for worldwide contact info

 

Here's someone that listened to the Ryzen CPU boost tuning tips and implemented them to optimize CPU boost and his 3900x is now getting almost 4.7ghz boost:

Y0yOy0 4 points 18 hours ago
"Got my bios updated to the new ABBA agesa which got me a small increase in boost frequency. I also downclocked my memory to just below 3600 mhz (got my cpu to boost even higher with this). Additionally I changed BCLK to 101 instead of 100 and now my 3900x boosts almost up to 4.7 ghz. Pretty cool. The mobo I'm using is the Gigabyte x570 aorus elite."

This is also a good idea:

Largemonitor1 1 point 2 days ago
"Is it not better to just manually set these things with PBO off? I have had a much faster system since doing the same using less voltage and MUCH lower temps with my 3800x set to 4.4ghz constantly."

Some GPU reviews, Thunderbolt on AMD4, plus more...

IT WAS AN EPIC BATTLE - RX 5700 XT & 5700 Roundup
HardwareCanucks
Published on Sep 12, 2019
Today we test some beefy custom AMD Radeon RX 5700 XT & RX 5700 cards. If you're thinking of picking one up, hopefully this video will help you narrow down a particular model. Enjoy!

Spoiler: Links for boards tested:

Enermax AIO White: https://geni.us/LIQII360
XFX RX 5700 XT - https://geni.us/XFX5700-XT
MSI RX 5700 XT - https://geni.us/MSI5700-XT
MSI RX 5700 XT OC - https://geni.us/MSI5700-XTOC
Sapphire RX 5700 XT OC - https://geni.us/Sapphire5700-XT
ROG STRIX RX 5700 XT OC - https://geni.us/STRIX5700-XT
XFX RX 5700 - https://geni.us/XFX5700
MSI RX 5700 - https://geni.us/MSI5700
MSI RX 5700 OC - https://geni.us/MSI5700-OC
PowerColor RX 5700 - https://geni.us/PowerColor5700

PowerColor RX 5700 Red Dragon Review: Non-XT Overclocking, Thermals, & Noise
Gamers Nexus
Published on Sep 10, 2019
This is the first of our non-XT RX 5700 series reviews, featuring the PowerColor RX 5700 Red Dragon primarily versus the AMD reference card. In this review, we're looking at the PowerColor RX 5700 non-XT Red Dragon video card, diving into GPU thermals, acoustics/noise levels, gaming performance, and overclocking. The RX 5700 Red Dragon is the first of the RX 5700-series cards from board partners that we'll be looking at, with others to follow in short order. Until now, we've only had the AMD reference 5700 with blower, but now we can look at reference vs. partner cards. The Red Dragon is among the cheaper board partner models, making it viable and competitive with the 2060 Super.
Article: https://www.gamersnexus.net/hwrevie...dragon-review-benchmark-overclocking-thermals


Ryzen & Thunderbolt - Together At LAST!
Level1Techs
Published on Sep 12, 2019
Thunderbolt AM4 Build


More takes on the AMD beta / soon to release firmware features:

AMD improves real-world boost and idle on Ryzen 3000
The extra 25-50MHz boost speed probably means less than the lower idle, really.
JIM SALTER - 9/11/2019, 12:24 PM
https://arstechnica.com/gadgets/2019/09/amd-improves-real-world-boost-and-idle-on-ryzen-3000/
Reader Comments

TomWestrick, SEP 12, 2019 4:49 AM
" Legatum_of_Kain wrote:
"Yep, as a matter of fact, I have a machine that's a 6700K that does not get to the advertised speeds at all, even overclocking it, it's off by about 40 mhz.

AMD is just getting hammered because there's a lot of people at Intel and using Intel that are having a meltdown on the overall value of the new AMD chips."
Pun intended?"

AMD ANNOUNCES RYZEN BIOS UPDATES FOR BOOST AND IDLE, NEW SDK
Posted by Sebastian Peak | Sep 10, 2019
https://pcper.com/2019/09/amd-ryzen-bios-updates-new-sdk/
Comments

AMD issues updated BIOS to fix Ryzen boost bug, promising up to 50MHz more boost
A new monitoring SDK will also make an appearance, providing fine-grained insights into what's going on under the hood of your Ryzen 3000 processor.
https://www.pcworld.com/article/343...ost-bug-promising-up-to-50mhz-more-boost.html
Comments aren't enabled...

RBD10100 AMD Ryzen 3600X | Radeon VII | ASUS B350-F STRIX 1 point 21 hours ago
"Just watched Gamers Nexus and their 3700X is maintaining 4.4 GHz in Cinebench 1 thread, and the 3900X actually did the 4.6GHz during some of the test, but it falls to 4.4 GHz at times they said. Still, it looks like things improved on their samples. It's holding 4.6GHz in the Total Warhammer game they tested. Here's hoping yours reaches higher too."
AMD Ryzen Boost Frequency Fix Benchmarked: AGESA ABBA Tested - Snowflake Edition
Gamers Nexus
Published on Sep 12, 2019
AMD has released AGESA "ABBA" with Ryzen 3000 boosting frequency fixes for the R9 3900X, 3700X, and other recent CPUs. We're benchmarking it. AMD's new AGESA update is now out [No, he's using a leaked BETA not a released version, that will be 2-3+ weeks before the release version comes out from motherboard makers] to motherboard manufacturers, and each of them will soon be adding it to BIOS revisions for boards. The new AGESA version (ABBA) includes frequency fixes, so our focus for testing will be: (1) if frequency is actually higher, (2) if there's any voltage or temperature impact as a result, (3) how much that change impacts some gaming or production performance.

 

AMD “will consider adding support” for RIS on Vega, but has no plans to right now

https://www.pcgamesn.com/amd/vega-radeon-image-sharpening

Seriously?
They skip this on the grounds of 'popularity'?
Meh... I'm starting to think AMD decided to shun Vega completely (which might explain latest driver problems Helios 500 users experience with static noise on screen when bringing the laptop out of sleep mode and some other issue desktop users experienced).

 

AMD's got so many actions to fill with staff that they need to limit things somewhere draw the line and deciding whether to continue a new feature across the product lines is one thing that makes sense - they can delay for resources for VEGA while watching for feature acceptance and use with high use it will be an easy justification for allocating those resources.

To you and me it initially makes sense to push out a new feature across all the GPU architectures at once to ensure acceptance with developers to encorporate in new games - but with RIS it's a tool that doesn't need developer buy-in it needs customer buy-in, and hopefully it is useful and does get high acceptance and use, and AMD can justify quickly pushing out RIS for Vega.

I'm more interested in the next generation Navi and beyond and I want AMD to not delay that more than not delaying RIS for Vega - if one really affects the other - only time will tell.

And, yes Vega in laptops would seem to be much larger than initially comes to mind, with the Vega in APC's + Consoles + future use, but it may be Vega is a minor part moving forward so it's getting pushed to the end of the resource allocation list, and that's what is happening.

AMD confirms it has no plans to add Radeon Image Sharpening to Vega but "will consider adding support" should there be demand
Submitted 18 hours ago by ninjap0wz Nitro+ RX 580 | i5 4590
https://www.reddit.com/r/Amd/comments/d3rt3x/amd_confirms_it_has_no_plans_to_add_radeon_image/

Interesting video digging in to VEGA and Navi architecture:

NAVI vs VEGA: Architectures
Naoki Watanabe
Published on Aug 29, 2019
Vega slow? Navi fast? We look at the differences between GCN & RDNA and why they matter to gaming.

 

This is apparently where AMD has Feedback support to see how much interest there is in expanding certain features on their GPU's (but nothing for specific GPU's).
https://www.feedback.amd.com/se/5A1E27D203B57D32

Radeon Image Sharpening and Anti-Lag would be my top ones... but we can apparently include vote for ALL of those features (except I don't particularly find much interest in the Gallery one).

 

Answered here:
http://forum.notebookreview.com/thr...zen-2-helios-500.817796/page-74#post-10950995

 

Anyone know for Ryzen 2 PBO if you use negative offset same as Ryzen + ?

Sent from my ONEPLUS A6013 using Tapatalk

 

See:
https://www.anandtech.com/show/1487...gning-perception-with-amds-frequency-metrics-

See:
https://www.tomshardware.com/news/amd-ryzen-3000-boost-fix-cores,40398.html

The saga continues (AMD achieving parity with Intel platforms in all aspects).

What is interesting is how we have to learn new definitions for what 'Turbo' is.

More interesting is that AMD doesn't even have a handle on their terminology or execution either (second link).

I'm sure it will sort out in the end (but this gives Intel an open door to implement even weirder 'definitions' too).

As long as my workloads/workflows (as anyone else's, of course) continue to improve, definitions be damned!

As usual, don't buy into the marketing fluff (from either side). Doing your own testing on your own workload(s) and workflow(s) and seeing actual improvements over what you are currently running is all that matters in the end. That is what you buy/spend your $$$$ on. Not the 'feelies' that marketing people try to sell you on.

 

You mean like Intel did with TDP and would do whenever convenient for them and their marketing team? Yep.

 

He referenced similar shenanigans, both actual and potential, from Intel:

 

Seems pretty basic, we've always suggested checking power settings in Windows as well as boost and other tuning suggestions as most PC's come set up with Balanced or worse performing settings (no Turbo) and other settings are a mixed bag across motherboards

With the variance in results from all of the people reporting (3000?) Ryzen boost results the first thing that came to mind was their variation in system settings like power, tuning, cooling - and the vast gap in understanding of how to set it all up correctly.

Coupled with the new tools and rules for Ryzen 3 I thought that would account for most of the poor results new owners were seeing - we see this all the time here when people come looking for performance tuning problems.

I didn't look at @der8auer's participation text for reporting results in his survey, but I hope he had basic set up requirements / suggestions included to help participants verify their basic settings to get the most from their set up before reporting results.

Dr. Ian Cutress @IanCutress
"For people still 300 MHz down on Ryzen, check your Event Viewer and latest system boot details. I just had an email from someone who had 100/100/100 for Kernel-Processor-Power, rather than something like 145/55/14. There might be some Windows Power Plan issues at play here."
https://twitter.com/IanCutress/status/1174259710587002882

Matsumura @Matsumuray 2h Replying to @IanCutress
"Mine is at 100/100/100 unless I specifically select "Enable" Cool'n'Quiet in BIOS (the default setting Auto does not solve it). MSI MEG X570 ACE w/3900X, using Ryzen Balanced Power Plan."

Dr. Ian Cutress @IanCutress Replying to @Matsumuray
"That'll do it. Cool n Quiet enables ACPI, and by extension CPPC."

Matsumura @Matsumuray 1h Replying to @IanCutress
"Shame MSI don't have it enabled by default/auto. I bet many people might be having the same issue due to the way it's set up out of the box."

Go to the thread on twitter, there's a lot more back and forth to read, in progress.

I wouldn't be surprised if this whole "tempest" isn't down to user "error" or inability to know what's going on with their setup and failed to configure adequately for best performance:

Razvan @dotMCL 5h Replying to @IanCutress
"This is ridiculous, really. I shouldn't need a Windows update (tweak) for my CPU to work correctly. I have no issues with how it performs, I do have an issue with how it is handled."

Charlie Demerjian @CDemerjian 3h Replying to @IanCutress
"Pebcak errors. "

"problem exists between chair and keyboard."

 

Definitely pebcak/pebkac errors for all getting hundreds of MHz lower. Der8auer even noted he excluded those, generally, from the sampling on distribution as low outliers, statistically, and because they had to be caused by errors in setup, errors in cooling, or other similar behavior. So even though shown on the graph, it was not used to pull down the average.

I know. I was being cheeky there, pointing out the TDP argument that is as old as time, it seems now.

His use of "even" before "weirder" signified Intel already had done the fudging before.

And, yes, AMD is playing it fast and loose on the definition of boost this gen. I have a feeling both sides will be doing that when Intel finally has 10nm on something other than mobile (which there is a chance the single core boost will rarely be seen on even the mobile parts although it will be at the rated speed! Lol).

Boost algorithms, similar to GPUs, are going to be implemented more and more as time moves on. And Lord knows we've all seen AIBs' boost claims challenged over the past half a decade. It is now moving to the CPU, just with more variables.

But I get your point and hope @tilleroftheearth saw that last comment as poking fun, not as "partisanship."

 

I think you missed my point if that's your reaction.

I think der8auer's exclusions weren't broad enough. I wouldn't be surprised to find that the % of actual 25mhz-50mhz were the bulk of the properly configured owners properly reading and reporting.

Given my (our) many years of experience helping people with new hardware understand the operation and configuration for optimal performance - helping them properly understand and configure their hardware to get all they are supposed to get out of new systems, I believe the bulk of those reporting higher than 25mhz-50mhz were outliers with improper configuration or monitoring: most likely Windows power settings, BIOS CPU settings, Wattman tuning, memory tuning, and other settings mis-configuration from misunderstanding or inexperience - or simply mis-reading or mis-interpreting the data.

Contributing to the outliers problems there might also be similar vendor mis-understanding and mis-configuration of vendor firmware. As reported by reviewers testing the same Ryzen CPU with the same settings on a range of different make / model motherboards resulting in about 50% getting the expected results and about 50% getting less performance than expected on a descending scale - which is of course outside the control of the end user, or AMD. The lower performing 50% of reviewed motherboards may be down to firmware issues, or design and production related problems. It might even be component tolerance issues from their suppliers.

Whittling down the outliers to the set of users seeing the 25mhz-50mhz shortfall AMD and solving those shortfall's with the AGESA firmware update should bring everyone with proper configuration fully in line with expectations.

The rest need to dig in and consider all of their configuration and tuning settings to see what's wrong with their setup, and if their motherboard is one of the "bad ones" to contact their vendor to get them to fix the shortfall they are responsible for with firmware or hardware ECO's.

I think we'll find the uplift in the education level of Ryzen owners brought about by owners digging in to proper settings will bring about better configured systems, and that combined with the new AMD AGESA + Vendor tweaks will bring their motherboards in line with expected performance.

Even though the 25mhz-50mhz firmware fix from AMD may not give much of a performance uplift, all those owners self-correcting their configurations by learning about correct settings while investigating their own systems performance will reap even greater benefits and happier AMD CPU owners.

I hope the 3950x gets out the door with better vendor awareness of proper firmware / motherboard tuning and get's it right the first time - AMD can announce the 3950x + new motherboards (if any), we're waiting.

 

What?! *Another* AMD 2nd gen EPYC CPU??! Already?

AMD drops surprise announcement of new 64-core 2nd generation EPYC processor
James Dawson 12 hours ago
https://www.kitguru.net/components/...of-new-64-core-2nd-generation-epyc-processor/

"...This afternoon, AMD announced a number of things, including a new addition to the 2nd generation family of EPYC ROME CPUs. The AMD EPYC 7H12 will be another 64 core/128 thread monster, this time boasting a 2.6Ghz base frequency and a maximum boost frequency of 3.3Ghz.

Just like the current crop of 2nd generation EPYC CPUs, the 7H12 will be specifically built for HPC customers and workloads.

The EPYC 7H12 turns out to be AMDs most power hungry CPU ever, with a 280W TDP!

Therefore, it is best suited to water cooled applications to ensure boost speeds are at a peak. ATOS has tested the CPU on its BullSequana XH2000 water cooled hybrid supercomputer. During these tests, the new EPYC 7H12 processor achieved a LINPACK score of 4.2 TeraFLOPS, which is an 11% improvement over the AMD EPYC 7742 processor. This comes as no surprise as the 7H12 basically seems to be an overclocked, water cooled EPYC 7742."

AMD's picked up on Intel's assisted cooling *stretch* the tech move.

"At the same event in Rome, AMD also showcased five new DELL EMC PowerEdge platforms powered by 2nd generation EPYC processors. IBM detailed how 2nd generation EPYC processors will help improve cloud security, provide better memory bandwidth for data and analytics workloads, and help core scaling and breakthrough performance for container workloads.

TMSC also announced its adoption of 2nd generation AMD EPYC to help power its next generation research and leading process technology among other things.

Since this is brand new information we are yet to hear details of pricing, however, we will be keen to share this once we are informed officially."

EPYC is so good TSMC is gonna use them to develop the next gen processes EPYC 3 / 4 will use.

I'd bet Intel wishes they could use 10nm Desktop CPU's and 10nm Server CPU's in their 7nm process development... too soon?

AMD’s New 280W 64-Core Rome CPU: The EPYC 7H12
by Dr. Ian Cutress on September 18, 2019 9:15 AM EST
https://www.anandtech.com/show/14882/amds-new-280w-64core-rome-cpu-the-epyc-7h12