Monday, September 2nd 2019

Der8auer: Only Small Percentage of 3rd Gen Ryzen CPUs Hit Their Advertised Speeds

World famous overclocker Der8auer published his survey of boost clocks found on 3rd generation Ryzen CPUs. Collecting data from almost 3,000 entries from people around the world, he has found out that a majority of the 3000 series Ryzen CPUs are not hitting their advertised boost speeds. Perhaps one of the worst results from the entire survey are for the 12-core Ryzen 9 3900X, for which only 5.6% of entries reported have managed to reach the boost speeds AMD advertises. However, the situation is better for lower-end SKUs, with about half of the Ryzen 5 3600 results showing that their CPU is boosting correctly and within advertised numbers.

Der8auer carefully selected the results that went into the survey, where he discarded any numbers that used either specialized cooling like water chillers, Precision Boost Overdrive - PBO or the results which were submitted by "fanboys" who wanted to game the result. Testing was purely scientific using Cinebench R15 and clock speeds were recorded using HWinfo (which got recommendation from AMD), so he could get as precise data as possible.
Der8auer comments that he still recommends Ryzen 3000 series CPUs, as they present a good value and have good performance to back. He just finds it very odd that AMD didn't specify what you need to reach the advertised boost speeds.

If you would like to see the more in depth testing, here is the English version of the video:

Add your own comment

248 Comments on Der8auer: Only Small Percentage of 3rd Gen Ryzen CPUs Hit Their Advertised Speeds

#176
EarthDog
anachron, post: 4109409, member: 189215"
The title of the picture you used is "Average Clockspeeds", and the "Max Boost Clock" line indicate 1898mhz, far above 1733. So unless there are more informations in the article it come from, it doesn't prove that the 1080 didn't at least reached 1733mhz in every game tested at some point. In the case of AMD, it seems to me (i don't have one so it's just based on previous posts) that some peoples can never reach the advertised boost clock.
Here is some context to the cherry picking (and deflection off the real subject).
As a percentage of the maximum boost clock, the average clockspeeds of the the GTX 1080 and GTX 1070 both drop more significantly than with GTX 980, where the latter only drops a few percent from its maximum. This is due to a combination of the temperature compensation effect we discussed earlier and both cards hitting 83C (though so does GTX 980). Either way both cards are still happily running in the 1700MHz range, and the averages for both cards remain north of NVIDIA’s official boost clock. Though this does give us a good idea as to why the official boost clock is so much lower than the cards’ maximum boost clocks.
Banging off the temp limit (as I said is not a good example of)...

The AMD CPUs not reaching the listed boost clock (which is not overclocking - the FACTORY lists that clock) is a SYSTEMIC issue, not a one off like one site showed the GTX 1080 to run in two games.
Posted on Reply
#177
anachron
Vya Domus, post: 4109413, member: 169281"
???

If it's the max and sometimes the max is under that 1733 figure then that means it wouldn't reach it's advertised clock speed. Isn't that as straight forward as it can possibly get ?

No one has "proved" that some AMD CPU's can never reach their max clock speed under any circumstance as far as I know.
As i said, your own chart title his "Average clockspeed", not "Max clockspeed".
Posted on Reply
#178
Midland Dog
anachron, post: 4109409, member: 189215"
The title of the picture you used is "Average Clockspeeds", and the "Max Boost Clock" line indicate 1898mhz, far above 1733. So unless there are more informations in the article it come from, it doesn't prove that the 1080 didn't at least reached 1733mhz in every game tested at some point. In the case of AMD, it seems to me (i don't have one so it's just based on previous posts) that some peoples can never reach the advertised boost clock.
gpus arent cpus, and anyone that knows anything would gladly sacrafice %20 core clock for %20 more vram clock, for a cpu it matters heaps, 4.6ghz got me an average of 103 on bl2 with a min of 66 (battle at overlook is a great bench very consistant test, 4.7ghz gave me 114 and a min of 73, on zen that 100mhz would make far more difference
Posted on Reply
#179
Vya Domus
anachron, post: 4109420, member: 189215"
As i said, your own chart title his "Average clockspeed", not "Max clockspeed".
Maybe I wasn't clear enough. Let's lay it out in the simplest way possible.

Nivida says 1733 is the boost clock. I play a game and the average recorded clock speed is under that value, this means that the advertised boost clock wasn't reached most of the time and in fact the clock speed sat under it most of the time.

The comment to which I replied claimed Nvidia's figures were maximums and that the cards always sat at their advertised clocks. This is false, that's not the maximum and 1733mhz is the only advertised figure.
Posted on Reply
#180
Midland Dog
Vya Domus, post: 4109413, member: 169281"
???

If it's the max and sometimes the average is under that 1733 figure then that means it wouldn't reach it's advertised clock speed all the time. Isn't that as straight forward as it can possibly get ?

No one has "proved" that some AMD CPU's can never reach their max clock speed under any circumstance as far as I know.
pascal is highly temp sensitive with clocks, 5c steps it up/down by 13mhz, a 2152 oc on a quiet curve might become 2177 or 2164 on an aggressive curve, heck my aib card boosted to 2025 factory, and does 2190 happily on air. der8auer proved that some (most) zen 2 cpus cant RUN at the ADVERTISED boost only 5.7% were from memory. I personally would be fuming if i bought a chip rated for 4.7ghz and it was a lemon and can only do 4.2ghz id be pissed

Midland Dog, post: 4109433, member: 168254"
pascal is highly temp sensitive with clocks, 5c steps it up/down by 13mhz, a 2152 oc on a quiet curve might become 2177 or 2164 on an aggressive curve, heck my aib card boosted to 2025 factory, and does 2190 happily on air. der8auer proved that some (most) zen 2 cpus cant RUN at the ADVERTISED boost only 5.7% were from memory. I personally would be fuming if i bought a chip rated for 4.7ghz and it was a lemon and can only do 4.2ghz id be pissed
the best solution is that for zen 3 amd lists the max all core boost, because that would be close enough to the max all core oc, in the same way that intels single core boost is close to its max all core oc if not lower
Posted on Reply
#181
Vya Domus
Midland Dog, post: 4109433, member: 168254"
pascal is highly temp sensitive with clocks, 5c steps it up/down by 13mhz, a 2152 oc on a quiet curve might become 2177 or 2164 on an aggressive curve, heck my aib card boosted to 2025 factory, and does 2190 happily on air. der8auer proved that some (most) zen 2 cpus cant RUN at the ADVERTISED boost only 5.7% were from memory. I personally would be fuming if i bought a chip rated for 4.7ghz and it was a lemon and can only do 4.2ghz id be pissed
Irrelevant, Nvidia shipped cards designed by them which they knew would be thermally limited. They set the temperature limit to 83C and with the cooler provided it reached that temperature all the time by default. The matter of the fact is, FE Pascal cards would not stay at their advertised boost clocks all the time. It's as simple as that.
Posted on Reply
#182
EarthDog
Midland Dog, post: 4109433, member: 168254"
pascal is highly temp sensitive with clocks, 5c steps it up/down by 13mhz, a 2152 oc on a quiet curve might become 2177 or 2164 on an aggressive curve, heck my aib card boosted to 2025 factory, and does 2190 happily on air. der8auer proved that some (most) zen 2 cpus cant RUN at the ADVERTISED boost only 5.7% were from memory. I personally would be fuming if i bought a chip rated for 4.7ghz and it was a lemon and can only do 4.2ghz id be pissed
We've also proven that under any circumstances the CPU can't hit the clocks. I'm still waiting as are many others, to see max boost. Again, I've tried over 6 motherboards, multiple BIOS' on each, single threaded loads, have a 3x120mm cooler, lowered polling rates and left the thing up for 24 hours, etc. I CANT HIT IT AND I AM EASILY WITHIN THEIR PARAMETERS. The same with many (most) users are finding. AMD didn't change the verbiage there because there wasn't a problem.......lol

So funny that example...average when we are talking peaks which the AMD CPU can't even hit. Also calling the boost clock overclocking when its a factory spec, lawllawllawl. Re-FUCKING-diculous.
Posted on Reply
#183
Midland Dog
Vya Domus, post: 4109438, member: 169281"
Irrelevant, Nvidia shipped cards designed by them which they knew would be thermally limited. The matter of the fact is, FE Pascal cards would not stay at their advertised boost clocks all the time. It's as simple as that.
you missed my thesis, pascal is temperature sensitive, if u ramped that blower up it would probably gladly sit in the 1900 range as long as the blower fan doesnt take to much power budget, fun fact the strix 1060 cooler's fans use about 9% of total power budget at max rpm

Midland Dog, post: 4109441, member: 168254"
you missed my thesis, pascal is temperature sensitive, if u ramped that blower up it would probably gladly sit in the 1900 range as long as the blower fan doesnt take to much power budget, fun fact the strix 1060 cooler's fans use about 9% of total power budget at max rpm
more comment sprawl but if the 3900x did 4.7ghz all core id have it in my rig right now and it would be at least %15 faster than my 4690k in the games i play, which is a great gain
Posted on Reply
#184
EarthDog
I'll just post this again...
What we find is that from the start of the run until the end, the GPU clockspeed drops from the maximum boost bin of 1898MHz to a sustained 1822MHz, a drop of 4%, or 6 clockspeed bins. These shifts happen relatively consistently up to 68C, after which they stop.


So when it hits the THERMAL THROTTLING TEMPERATURE, of course its going to drop bins! All one needs to do is a quick fan speed adjustment or raise the temp limit.. done. AMD isn't hitting any throttling point and can't do for most it is the point... no matter what!! Let's not lose the fact that this is a AMD CPU thread and it is silly to go down this hole of a GPU analog.

The point is AMD is struggling mightly for most to hit their advertised (stock) clocks. Period. AMD fans can call it overclocking all they want to, but it isn't 'the same bloody issue' FFS that couldn't be any more obvious we are comparing apples and oranges...

It is something that was 'guaranteed' to hit it IF the parameters they set forth are in play...according to AMD and the jebaited baldy.

I digress on this point.
Posted on Reply
#185
Vya Domus
Midland Dog, post: 4109441, member: 168254"
you missed my thesis, pascal is temperature sensitive, if u ramped that blower up it would probably gladly sit in the 1900 range as long as the blower fan doesnt take to much power budget, fun fact the strix 1060 cooler's fans use about 9% of total power budget at max rpm
We are here to prove AMD is evil and is falsely advertising tubo clocks not that Pascal is temperature sensitive.

If you need a different cooler because the product was temperature sensitive then you either should have shipped a different cooler or advertised different clocks. See ? It's the same bloody issue, everyone is getting liberal with way they come up with these figure.
Posted on Reply
#186
TheLostSwede
Chrispy_, post: 4109357, member: 185623"
Exactly. The only GUARANTEED speed is the base clock. If you can't reach that, something is wrong.
Boost is opportunistic, regardless of whether we're talking about CPUs or GPUs, and regardless of whether it's a red, green, or blue logo on the product.

No, that's specifically about PBO overclocking using top-tier motherboards and comes clearly emphasised with the words "might" and "maybe". The dude even slows down and stresses those words, making it clear to anyone with functioning braincells that is it NOT A GUARANTEE you will get those speeds.

How can people not understand the Silicon Lottery and concept of Overclocking by now?
De8auer's takes the time and effort at the start of the video to very clearly explain that he threw out all of the PBO and PBO+ results and only looked at bone-stock submissions. Don't go bringing PBO+ overclocking into this discussion, it's a strawman argument that isn't remotely helping.
And when people don't even get close to boost speeds, this is AMD spitting in people's face, no?
Judging by der8auer's video, all of two dozen people seems to be hitting speeds above the boost on the 3900X, less on every other chip except the 3800X. And hitting 25-50MHz above boost isn't exactly impressive when AMD talks 100-200MHz above boost.
I have said top-tier board, the right cooling and everything else, yet at one point, I was stuck at 4,400MHz and couldn't go 1MHz above, but a recent UEFI update got me to 4,525MHz during boost on three of my cores.
As such, there are issues, but AMD isn't being forthcoming with what's going on, which I think is what frustrates people the most.
Posted on Reply
#187
Midland Dog
Vya Domus, post: 4109448, member: 169281"
We are here to prove AMD is evil and is falsely advertising tubo clocks not that Pascal is temperature sensitive.

If you need a different cooler because the product was temperature sensitive then you either should have shipped a different cooler or advertised different clocks. See ? It's the same bloody issue, everyone is getting liberal with way they come up with these figure.
1: 1080 fe doesnt drop below boost, 2: no one wants to prove amd is evil, im actually dissapointed that zen 2 didnt compell me to upgrade some of my old d3d9 games really need more st perf, 3: amd SAID it should boost to 4.7ghz and yet only a small portion get close, the majority dont even in a controlled environment. The core wars are more retarded than i am, it wont get us anywhere, the best way to increase multi threaded perf is to make faster individual cores, not ad more cores and complexity, extra cores need to be fed, fewer faster cores will still need to be fed but it will be far easier to design them and older programs and newer ones alike would scale
Posted on Reply
#188
anachron
Vya Domus, post: 4109431, member: 169281"
Maybe I wasn't clear enough. Let's lay it out in the simplest way possible.

Nivida says 1733 is the boost clock. I play a game and the average recorded clock speed is under that value, this means that the advertised boost clock wasn't reached most of the time and in fact the clock speed sat under it most of the time.

The comment to which I replied claimed Nvidia's figures were maximums and that the cards always sat at their advertised clocks. This is false, that's not the maximum and 1733mhz is the only advertised figure.
While i agree with you that nvidia is not the perfect example on advertising done right, the card reach the advertised boost clock in at least seven of the nine games of your charts, it's still better than never reach it. So i understand why people are reacting more angrily to the zen 2 issue.
Posted on Reply
#189
Vya Domus
Midland Dog, post: 4109452, member: 168254"
1: 1080 fe doesnt drop below boost
Sure it does, maybe you missed it have a look at this picture again.




You know how averages work, right ? If some measurement is supposed to stay above value X but the average is below X then that means it dropped below that number.
Posted on Reply
#190
Midland Dog
i actually enjoy analysing the cpu market at the moment all though i feel that amd has lost site of what a cpu should be. a cpu is designed to do serial tasks in a timely manner, gpus do parallel, ryzen feels more like a hybrid uArch, trying to parallelise a cpu if you will, its a great idea but they already have a gpu department that does that. intel on the other hand builds a cpu to do what a cpu should do

Vya Domus, post: 4109456, member: 169281"
Sure it does, maybe you missed it have a look at this picture again.




You know how averages work, right ? If some measurement is supposed to stay above value X but the average is below X then that means it dropped below that number.
average clocks that are 1 step off of the boost, or average clocks that are 400 mhz off of the boost which one sounds better, keeping in mind that 13mhz gpu core equates to a hole 71 gflops id say it makes less of a difference there
Posted on Reply
#191
Vya Domus
anachron, post: 4109454, member: 189215"
While i agree with you that nvidia is not the perfect example on advertising done right
So that's my point. Getting mad on a per basis case is useless, these guy don't advertise these things based on what some guys on a form say. They advertise them based on what everyone else does, so everyone needs to stop.

Midland Dog, post: 4109459, member: 168254"
less of a difference there
Same goes for a 3900X then, 4550 is close enough. This is exactly what I am talking about, people are fine with one manufacturer doing it but not with the other.
Posted on Reply
#192
TheLostSwede
Midland Dog, post: 4109459, member: 168254"
intel on the other hand builds a cpu to do what a cpu should do
Which is what exactly?
Are you saying that Intel CPU's have been the same since the early days of the 4004 in how they work?
I guess you've never heard of AVX? Afaik, AVX is massively parallel...
Posted on Reply
#193
Midland Dog
TheLostSwede, post: 4109465, member: 3382"
Which is what exactly?
Are you saying that Intel CPU's have been the same since the early days of the 4004 in how they work?
I guess you've never heard of AVX? Afaik, AVX is massively parallel...
yes avx is a parallel task however intels focus is still to get the most throughput out of what cores it has, amds current design goal is to add cores to hit the required throughput. i dont forsee a physical ipc wall but i do forsee a core scaling wall

Midland Dog, post: 4109468, member: 168254"
yes avx is a parallel task however intels focus is still to get the most throughput out of what cores it has, amds current design goal is to add cores to hit the required throughput. i dont forsee a physical ipc wall but i do forsee a core scaling wall
heres a good read, amdahl's law https://en.wikipedia.org/wiki/Amdahl%27s_law
Posted on Reply
#194
Vya Domus
Another funny thing is that we are talking about something that was never meant to be guaranteed, never has been and never will. Every manufacturer has been clear on this from day one ever since "turbo clocks" became a thing.

Turbo clocks being a requirement or a guarantee is something people came up with on their own.
Posted on Reply
#195
Midland Dog
Vya Domus, post: 4109474, member: 169281"
Another funny thing is that we are talking about something that was never meant to be guaranteed. Every manufacturer has been clear on this from day one ever since "turbo clocks" became a thing.

Turbo clocks being a requirement or guartee is something people came up with on their own.
except that for the last however many years boost clock was basically guaranteed, tell me an intel or amd gen prior to ryzen that couldnt hit its boost on a large scale. we arent just talking about half of the ryzens not hitting boost we are talking about ONLY 5.7% of a sku hits its boost, please tell me that you dont believe that only 5.7% of 9900k chips have hit 5ghz boost, because that is soooooo wrong

Midland Dog, post: 4109479, member: 168254"
except that for the last however many years boost clock was basically guaranteed, tell me an intel or amd gen prior to ryzen that couldnt hit its boost on a large scale. we arent just talking about half of the ryzens not hitting boost we are talking about ONLY 5.7% of a sku hits its boost, please tell me that you dont believe that only 5.7% of 9900k chips have hit 5ghz boost, because that is soooooo wrong
9590 certainly wouldnt have sold if the advertised 5ghz only meant 4.6ghz in the real world
Posted on Reply
#196
Vya Domus
Midland Dog, post: 4109479, member: 168254"
except that for the last however many years boost clock was basically guaranteed
Except that's not the case and it wouldn't even make sense. If the turbo clock is a guarantee then they would just write one number, however they don't, they say "this is the base clock" (guaranteed) and this is the "turbo clock" (mileage may vary).

The distinction is there, they are not the same thing and they are not both guaranteed. Then it would just be one number.
Posted on Reply
#197
Midland Dog
Vya Domus, post: 4109486, member: 169281"
Except that's not the case and it wouldn't even make sense. If the turbo clock is a guarantee then they would just write one number, however they don't, they say "this is the base clock" (guaranteed) and this is the "turbo clock" (mileage may vary).

The distinction is there, they are not the same thing and they are not both guaranteed. Then it would just be one number.
i feel that i havent expressed my point clearly enough, AMD CANT LABEL A CHIP AS 4.7GHZ IF THE SILLICON IS SO BORKED THAT IT WONT DO 4.3GHZ, INTEL CAN LABEL A BOOST CLOCK AT 5GHZ BECAUSE ALL 9900K CAN DO IT ON AT LEAST 1 THREAD. dont sell borked sillicon at impossible clocks
Posted on Reply
#198
Vya Domus
Midland Dog, post: 4109491, member: 168254"
i feel that i havent expressed my point clearly enough
You did but unfortunately, you're wrong. Turbo clocks have never been guaranteed by anyone, you might get them you may not and to varying degrees.
Posted on Reply
#199
Midland Dog
Vya Domus, post: 4109492, member: 169281"
You did but unfortunately, you're wrong. Turbo clocks have never been guaranteed by anyone, you might get them you may not and to varying degrees.
well EVERY INTEL chip ive owned hit its turbo, if the sillicon cant hit those clocks then tell us so we dont expect intel crushing perf in older titles
Posted on Reply
Add your own comment