Discussion CPU boost frequency and marketing

[SK10H]

AMD is good at sandbagging. AMD makes themselves look bad once they don't sandbag like this thread keeps showing.

They better don't pull this BS again if/when 5GHz is finally attainable.

 

[happy medium]

Only 5.6% of Ryzen 9 3900X Hit Advertised Speeds, Most Other Models Suffer, Too.

https://amp.tomshardware.com/news/a...ing-advertised-boost-speeds-survey,40291.html

You cannot just drop links with no commentary. Read the rules.
Your extra-supersize-colored-fonts, can also be taken as inflammatory, trolling.


esquared
Anandtech Forum Director

 

[VirtualLarry]

You know, if AMD just came out and said, "Those boost speeds are only for LN2 users", maybe that would be a solution.

I mean, we have the extreme high-end mobo category, with 12/16-phase wildly-overspecced VRMs, and BIOS settings for 1.5V-1.8V vcore, on CPUs that normally can only take 1.3V vcore on air. No-one here is claiming that they can't reach / use those vcore settings during nominal operation. (Because they are for LN2 benching.) Same deal.

Just like DRAM clock settings, of like 4800-6000, no-one's going to reach those on air with current DRAMs (I don't think). No-one's suing mobo makers over that.

 

[Markfw]

You know, if AMD just came out and said, "Those boost speeds are only for LN2 users", maybe that would be a solution.

I mean, we have the extreme high-end mobo category, with 12/16-phase wildly-overspecced VRMs, and BIOS settings for 1.5V-1.8V vcore, on CPUs that normally can only take 1.3V vcore on air. No-one here is claiming that they can't reach / use those vcore settings during nominal operation. (Because they are for LN2 benching.)

DrMrLordx got it to spec boost as did others here on this forum. WITHOUT LN2

 

[Vattila]

They better don't pull this BS again if/when 5GHz is finally attainable.

It will be interesting to see how the 3950X fares, considering they have already committed to 4.7 GHz max boost.

[Tom's Hardware reports on DerBauer's findings]

With the main hardware sites picking up the story, I guess AMD is soon forced to respond with a statement on the matter.

 

[cellarnoise]

This should not be understated. My small group that is doing audio and video with only single samples of true driven hard hardware, ... 4700mq - mobile and too loud for our work, 1700x at stock - and never really hit boost specs of 3.8 or 3.9, and a single 1950x that does wonders under air but does not really go above 3.4ish under editing (we love this cpu), and the single 2700x that does 3.15 ish as expected under load.

I am watching the current debate inorder to understand motherboards and such. I'm not so concerned about under 3 thread count but all core should not dip too low based upon whatever TDP or measured benchmarks show. Measured values are more valued than any corporate numbers that I would value.

Heat seems to be a big deal on all these cpus and graphic cards going foward. I'm not so sure that the current cases do so well.

 

[VirtualLarry]

DrMrLordx got it to spec boost as did others here on this forum. WITHOUT LN2

Yeah, my R5 3600 boosts to 4.2Ghz easily too... ONLY AT IDLE!

I mean, it's not like this issue is "made up". Maybe a little bit overstated, because the performance is IMHO there, but boost behavior is DIFFERENT than Intel, and people have gotten entititled to Intel-style boost, especially with the "unofficial overclock" - "MCE", that nearly all Intel boards do, that gives people an all-core overclock, what a single/dual-core boost is supposed to be at on Intel.

Edit: The fact that on AMD Ryzen 3000-series, that the boost behavior is such, that "max" is truely "the max" that it will boost too (formerly, called "XFR limit"), and not a sort of "daily-driver boost speed" that the user sees regularly, just so that they could put a higher number on the box for marketing reasons against Intel's arguably more-capable frequency-wise boost clocks, well, it doesn't sit well with some people.

I understand, they couldn't hit 5Ghz like Intel, and even so, they can still compete performance-wise with them, at a lower clock speed, due to having BETTER IPC, but the optics of the situation could be better. Customers like it better when AMD is sandbagging it, rather than trying to stretch the truth a little bit.

 

[jpiniero]

You know, if AMD just came out and said, "Those boost speeds are only for LN2 users", maybe that would be a solution.

Saw speculation out there that AMD cut the boost via agesa due to long term reliability issues. How long term are we talking about, no idea.

 

[moinmoin]

Saw speculation out there that AMD cut the boost via agesa due to long term reliability issues. How long term are we talking about, no idea.

I think by now it's pretty clear AMD was too optimistic regarding what frequencies are easily attainable with Zen 2 on TSMC's 7nm. For whatever reason they weren't able to backpedal in time so now them not having sandbagged in this instance is consequently hurting them (in the frequency PR fight that Intel will more an more focus on, having nothing else to show).

 

[beginner99]

I mean AMD just got fined for the bulldozer core thing which was stupid really. this is much clearly and I see another double-digit millions fine coming in a couple years. Let's all admit it was a big blunder on AMDs part not putting 100mhz less on the box to avoid the whole debate and probably future class action lawsuit.

Yeah for performance it hardly matters still AMD just always manages to blunder somehow in a very avoidable way.

 

[Thunder 57]

I mean AMD just got fined for the bulldozer core thing which was stupid really. this is much clearly and I see another double-digit millions fine coming in a couple years. Let's all admit it was a big blunder on AMDs part not putting 100mhz less on the box to avoid the whole debate and probably future class action lawsuit.

Yeah for performance it hardly matters still AMD just always manages to blunder somehow in a very avoidable way.

I've never been a fan of AMD's marketing. Gamecache has to be the worst.

 

[Kenmitch]

Yeah, my R5 3600 boosts to 4.2Ghz easily too... ONLY AT IDLE!

My 3600 hit's 4.2 GHz on every core eventually if I monitor it long enough. It'll boost to and maintain 4.2 GHz in Cinebench single core runs. At times 2 or more cores will show 4.2 GHz during the runs. I tried R20, R15, R11.5 and they all do the same thing. Mine looks to tag team single core loads moving the load between core 1 and 2 on the C6H latest uEFI 1.0.0.3abb. Probably Win 10 schedular as I don't think my combos a special config.

My 3700x I think 4.375 GHz is usually the max without uEFI monkey biz. I did see 4.4 GHz playing around with MSI beta bios, but it didn't like my b-die....Go figure.

Different MB's, uEFI versions, default setting, and user input errors can effect boost it looks like.

Going by my 2 examples I see no issue to complain about.

 

[PotatoWithEarsOnSide]

I had no issues at all getting my 3600X to run (and hold) 4.5GHz+ on each core individually. That tells me that it isn't that the cores causing the issue, its the algorithms that are behind boost behaviour.
Admittedly, it wasn't under a CB R20 load, but it wasn't a fleeting clock either.

What I've found interesting is that Ryzen Master doesn't necessarily display as high boosts as I've seen with HWinfo64. Probably just a polling rate thing though.

 

[VirtualLarry]

I had no issues at all getting my 3600X to run (and hold) 4.5GHz+ on each core individually. That tells me that it isn't that the cores causing the issue, its the algorithms that are behind boost behaviour.

Yeah, that could well be true.

 

[Vattila]

OptimumTech reviews DerBauer's findings and adds his voice to the view that AMD has advertised boost frequency poorly, causing confusion and disappointment, by creating unnecessary expectations (+200 MHz PBO) and not meeting obvious expectations (existing Ryzen 1000/2000 and Intel Core boost behaviour). He also mentions the idle boost issue as something that should ideally have been ironed out before launch.

 

[DrMrLordX]

DrMrLordx got it to spec boost as did others here on this forum. WITHOUT LN2

I will add as a caveat that my best boosts happened under the launch UEFI, which crashes with Destiny 2 and has some other flaws.

On my specific setup (x570 Aorus Master), the way to max out boost clocks in any given UEFI is to:

1). Pick DDR4-3200 or lower. Seriously, higher RAM clocks actually makes stability harder to achieve at higher CPU clocks. I've observed this even with my static OC. Now that I'm running DDR4-3733 I can't even get 4.4 GHz stable in CBR20 anymore. At least not without insane voltage.
2). Set CPU and SoC LLC to "Normal". This is the equivalent of LLC OFF. Alternatively you can go LLC Standard, which is LLC 1, and that's very close in performance, but not quite as good. Those settings are specific to the Gigabyte boards. Other boards may exhibit different behavior.
3). Set a CPU voltage offset of -.1000v.

I haven't tested that with DDR4-3200 yet, but with DDR4-3733 I can get boosts in the range of 4550 MHz for very brief time windows. I also got one of my highest stock performances in CBR20 ever (~7300). I think if I dropped RAM speeds a bit, I'd get more stability/higher boost clocks. Strangely, this setup is not good for SuperPi, which does not want to go over 4.3 GHz. Might require more testing/research but I think it may be me RAM speeds (DDR4-3733) holding me back on boost clocks.

PBO is a different beast.

Yeah, my R5 3600 boosts to 4.2Ghz easily too... ONLY AT IDLE!

Try fiddling with LLC settings. Set them low. Then maybe try some negative vcore offsets. Also be aware that if you want to run high memory speeds, that lowers boost performance under default behavior. I think a lot of us expected that RAM OC wouldn't affect boost behavior. Sadly, it does. Officially AMD doesn't support anything over DDR4-3200 JEDEC anyway so . . . technically we can't blame them there.

@cellarnoise

Try some of the tweaks I recommended. They seem to be good for multicore/all-core boost. Not so great for single-core boost (1T workloads, like CBR20 ST or Super Pi). I think for your workloads, you're looking more at all-core.

 

[VirtualLarry]

I'm not actually that concerned with only single-core boost, as most of my workloads are all-core, and mostly FPU, although I suppose the mining may be all integer (hashing and whatnot).

 

[Fanatical Meat]

Supposed to do **up to** 4.6 GHz.

Assuming that later BIOS revisions don't rectify something, I agree that they should have labelled it as 4.5 GHz - but "boost up to" does not equate to "will boost to".

Good point, everyone remember “up to” and “less than” are the same statement.

Are there no salespeople posting in this thread, seriously I’ve heard crap like the above statement my entire career.
This will do up to....
Your commission will be up to....

 

[Kenmitch]

https://www.tomshardware.com/news/amd-ryzen-3000-boost-frequency-bios-fix,40308.html

New article about incoming fix possibly.

 

[Markfw]

For those of use who use all cores, this is a non-issue.I think gamers are the ones mostly affected.

 

[UsandThem]

https://www.tomshardware.com/news/amd-ryzen-3000-boost-frequency-bios-fix,40308.html

New article about incoming fix possibly.

For those of use who use all cores, this is a non-issue.I think gamers are the ones mostly affected.

Hopefully this works and gives the majority of users the 25Mhz (according to Der8auer's findings) they were shorted, so everyone can now be happy.

I don't own a Ryzen 3*** series CPU, but I would be one of the people who doesn't really care about this issue. I never look at boost rates of my Ryzen 2700X CPU, as all I care about is the CPU's actual performance. I'm not discounting those who pay attention or who are concerned about their clock rates, it just isn't something I am really concerned about.

 

[Vattila]

Source: AMDRyzen at Twitter

 

[jpiniero]

Hopefully this works and gives the majority of users the 25Mhz (according to Der8auer's findings) they were shorted, so everyone can now be happy.

There's likely a really good reason they nerfed the boosts in the first place, especially given the talk it was done for long term reliability reasons.

 

[Markfw]

Hopefully this works and gives the majority of users the 25Mhz (according to Der8auer's findings) they were shorted, so everyone can now be happy.

I don't own a Ryzen 3*** series CPU, but I would be one of the people who doesn't really care about this issue. I never look at boost rates of my Ryzen 2700X CPU, as all I care about is the CPU's actual performance. I'm not discounting those who pay attention or who are concerned about their clock rates, it's just something I personally not overally concerned about.

I agree. This whole deal about 25 mhz. Even as a computer nerd, I round numbers quite often, and to me its absurd to be upset about 25 mhz.

I used to program statistics. If the claimed boost was within the 95% confidence level based on boneified mass testing (not user testing), then it would be a non-issue to anyone with a brain.

 

[PotatoWithEarsOnSide]

Not sure I agree with the statement that only one of the cores can hit the advertised boost clock. All of mine can exceed it. Problem is that they can't do it on their own.