Discussion CPU boost frequency and marketing

[lopri]

So where can one find information about the maximum all core clock of the Ryzen 3000's?

Is the current state of play the end of the story, or will a new bios release in a few months or weeks, improve clockspeeds on all core?

From what I gather, Ryzen 3000s do not have any guaranteed boost clock, single or all cores. It's all "opportunistic" boosts when all the parameters are met.

 

[Zstream]

From what I gather, Ryzen 3000s do not have any guaranteed boost clock, single or all cores. It's all "opportunistic" boosts when all the parameters are met.

3600, running on a X570 mobo @ 4200 MHz with and without stock cooler. Not sure why peeps are having issues. It was just a simple boot and it worked.

 

[DrMrLordX]

3600, running on a X570 mobo @ 4200 MHz with and without stock cooler. Not sure why peeps are having issues. It was just a simple boot and it worked.

Is that boost or static OC? If I were you I'd do some testing with static OC and compare MT results (pick something like CBR20 or CBR15) to see if you're really getting the performance afforded by 4.2 GHz clocks.

 

[Zstream]

Is that boost or static OC? If I were you I'd do some testing with static OC and compare MT results (pick something like CBR20 or CBR15) to see if you're really getting the performance afforded by 4.2 GHz clocks.

I’m guessing boost, as RM shows that number on a variety of cores. You mean run CB with a permanent 4.2? What are you expecting the results to be?

 

[DrMrLordX]

I’m guessing boost, as RM shows that number on a variety of cores. You mean run CB with a permanent 4.2? What are you expecting the results to be?

Yeah run CB with 4.2 GHz static if it's stable and in acceptable temp range. I expect that the results based on 4.2 GHz boost may put up lower scores. But that's a maybe, not a definite.

Alternatively, try 4.0 or 4.1 GHz static if it's easier to achieve those speeds and do a comparison of performance #s anyway. The results should be interesting.

 

[Kenmitch]

I’m guessing boost, as RM shows that number on a variety of cores. You mean run CB with a permanent 4.2? What are you expecting the results to be?

That would be the single core boost your seeing. When I tested my 3600 it would always lock onto a single core and maintain it's 4.2 GHz boost during the entire single core run of CB20.

Probably around 3775-3800 or so would be my guess for a 3600 all core 4.2 GHz CB20 run.

 

[Zucker2k]

Seems nobody has posted this. From silicon lottery...……..

 

[moinmoin]

Link for the above table: https://siliconlottery.com/pages/statistics

Edit: I see it even has its own thread already.

 

[lopri]

AMD has updated the Ryzen product pages to be more specific about what "max boost" means:

"Max Boost Clock is the maximum single-core frequency at which the processor is capable of operating under nominal conditions."

Ryzen 7 3700X Specifications

Here I suppose "nominal" has the aerospace jargon meaning: "performing or achieved within expected, acceptable limits; normal and satisfactory." (Dictionary.com)

This makes it clear that the advertised "Max Boost" is the best frequency possible under ideal conditions — the limit of the operating range — unlike earlier generations which had XFR. The slide I posted earlier (see below) further implies that this frequency may not be reached with the stock cooler. The disappointing fact for overclockers though is that it isn't easy to raise the ceiling even with premium cooling and PBO. The top red part of the bar is reportedly vanishingly small for most users. Perhaps we'll see more headroom later with improved firmware, drivers and/or die refinements, but I guess that is wishful thinking.

That said, premium cooling and a quality motherboard with good power delivery should raise the average boost frequency seen within the operating range when running sustained workloads.

Tip: Focus on performance, not frequency.

I've found a clearer explanation from 2 years ago. And as you said, AMD basically turned the XFR range into the new Boost clock.

 

[Vattila]

Excellent investigative work by Hardware Unboxed on the Max Boost issue:

Summary: There is variance between motherboards, and on the best performing motherboards the Ryzen 3000 processors do reach advertised Max Boost.

Interestingly, Steve didn't pick up on the XFR thing, i.e. that XFR was exactly the "50-100 MHz" difference between momentary peaks and sustained single-threaded workloads that he mentions is to be expected. Now that XFR is no longer a thing, users will need to recalibrate expectations.

Important to note: The slight frequency differences seen amount to rounding and margin of error. Focus on performance!

 

[UsandThem]

Excellent investigative work by Hardware Unboxed on the Max Boost issue:

That's an excellent video / mega test.

 

[DrMrLordX]

Important to note: The slight frequency differences seen amount to rounding and margin of error. Focus on performance!

I find that boost clocks are smoke and mirrors. You have to look at bench results. There's a good deal of inconsistency in what settings improve performance in which applications when attempting to use boost.

 

[PotatoWithEarsOnSide]

So I tested each core of my 3600X individually in Ryzen Master in the evening yesterday. Each core was able to hit and hold 4.5GHz at 1.425v during a 3-minute Ryzen Master stress test.
The cores are clearly able to hit (and exceed) their advertised boosts. However, in practice they simply don't do so by themselves.

I also tested all cores at 4.4GHz in Ryzen Master stress test. All cores hit and held the 4.4GHz throughout the 3 minute stress test. Temperatures only hit 70C and the EDC got as high as 90% of 128A.

I then dropped the clocks to 4.35GHz and ran Cinebench R20. Clocks were hit and held throughout. Temperatures hit 82C and I saw 92% EDC. PPT and TDC also were a fair bit higher than under the Ryzen Master stress test.

Back to Ryzen Master and my observations. When testing a single core I simply lowered all other cores to 600MHz instead of disabling them. What I saw was that the cores in the same CCX as the one being tested would also be holding 2.0GHz during the test. For the second CCX testing I saw even more bizarre behaviour. For some reason the scheduler would move the tested core in and out of sleep, thus affecting what clocks could be held. Consequently, I had to disable the 1st CCX for my 2nd CCX to be able to hold the clocks through the stress tests.

Edit: for reference, at stock my 3600X runs Cinebench R20 MT at 4.125GHz, and R20 ST at 4.3GHz. It was scoring 3591 and 481 at stock. At 4.35GHz it scored 3833 and 494.
RAM is still at 3200MHz CL16 at XMP with timings of 16-18-18-36.

 

[Head1985]

ryzen 3000 boost survey.Only 5% 3900x hits 4.6ghz and 14% 3700x hits 4.4ghz

 

[gdansk]

ryzen 3000 boost survey.Only 5% 3900x hits 4.6ghz and 14% 3700x hits 4.4ghz

This looks pretty bad. And it's a real shame because these CPUs perform very well despite missing their advertised boost speeds. If they only reduced the advertised speed by 200-300MHz we would all be happy seeing AMD chips boost to higher than advertised clocks and compete against Intel with lower clock rates.

I understand they were under some 5GHz pressure but this was a mistake on AMD's part. Being proactive and reducing the advertised clock rate will not protect them at all from lawsuits but it would reduce the scope of people who can say AMD lied to them.

 

[Head1985]

This looks pretty bad. And it's a real shame because these CPUs perform very well despite missing their advertised boost speeds. If they only reduced the advertised speed by 200-300MHz we would all be happy seeing AMD chips boost to higher than advertised clocks and compete against Intel with lower clock rates.

I understand they were under some 5GHz pressure but this was a mistake on AMD's part. Being proactive and reducing the advertised clock rate will not protect them at all from lawsuits but it would reduce the scope of people who can say AMD lied to them.

Well it boosted better with older agesa.My 3700x boosted at 4.4ghz, but with 1003ab it hits only 4375mhz(average is only around 4300mhz in cb ST test).So amd nerfed boost with newer agesa.Max voltage is same 1.5V so i am not sure why amd nerfed it.

 

[Kedas]

It's not that bad, it looks bad because he thinks that 4.375 is not the same as 4.4Ghz . sure it isn't the same but do you see AMD putting 4.375 on the box. (it's one resolution step)
This is 0.5% maybe this could be the 100Mhz base clock that is a bit overclocked on some motherboards.

But he is right there are setups that didn't get close enough to see it as a measure/tolerance mistake.
So it would probably be more right if AMD dropped the advertised boost with 0.1Ghz then people can be happy that hey got 0.1Ghz more out of there CPU at stock settings.

 

[TheELF]

It's not that bad, it looks bad because he thinks that 4.375 is not the same as 4.4Ghz .

And he thinks that because it's the peak number, even if it's only reached just for a split second.
Sustained single boost will be another resolution step or two below that.

If they only reduced the advertised speed by 200-300MHz we would all be happy seeing AMD chips boost to higher than advertised clocks and compete against Intel with lower clock rates.

Yeah that would put their flagship CPUs single core burst boost below or barely at the sustained all core boost of even the lowest i3,the 9100.
That would be a killer for marketing.
4.6 is the highest turbo for intel excluding i7 and i9 so it's much better for marketing.

 

[Vattila]

[In DerBauer's survey] only 5% 3900x hits 4.6ghz and 14% 3700x hits 4.4ghz

In DerBauer's survey, the 3700X has the largest number of submissions, and hence is closest to significant sample size (from submissions that he himself admit is likely to be dominated by unhappy customers). Yet in this sample set, the rounding of mere 25 MHz is decisive for the conclusion drawn.



It is easy to imagine that this 25 MHz difference — only -0.57% from the stated Max Boost — could come down to any number of factors (motherboard, BIOS/AGESA, Windows, testing conditions). Let's look at that difference with a bar chart starting from 0:



Not a big difference, is it?

The premise of his conclusion is also based on a specific interpretation of "Max Boost" as meaning "at least or above", while the meaning of "max" in general use is "up to, but not beyond".

I am disappointed that throughout his investigation of this issue, he has not discussed the fact that XFR has been dropped, and that AMD now really mean "max" when they say "Max Boost". That alone changes the whole perspective on what we should expect.

Again, here is the slide showing how it worked for the first and second generation:

From this slide it is clear that the old definition of "Max Boost" was not the max momentary boost clock achievable, but the max sustained boost clock. It seems likely to me that AMD anticipated the potential for confusion already before launching Ryzen 1000, and mitigated it by introducing the XFR concept. However, for the 3000 series launch I suspect they were under pressure to reach targeted frequencies, and hence decided to drop the XFR headroom and make "max" mean "max momentary boost", allowing them to print larger numbers on the box.

Adding to the confusion is the fact that the boost algorithm is highly complex and dependent on many factors including motherboard implementation, BIOS/AGESA, operating system behaviour and testing conditions. It is a shame that the boost algorithm should be a source of dissatisfaction, considering it ensures that the customer gets more performance than with the old statically set clocks.

In the end though, looking at my bar chart above, DerBauer is making a mountain out of a mole hill on this issue. Again, my advice is to focus on performance, not frequency.

 

[Bouowmx]

Focus on performance, not frequency!

What thread is this? "CPU boost frequency and marketing". Performance is not a prominent demerit of Ryzen 3000.

 

[tamz_msc]

Yet in this sample set, the rounding of mere 25 MHz is decisive for the conclusion drawn.

You cannot round off the clocks like that. He clearly states that these are the maximum achieved clocks, not sustained clocks, so a 4375 MHz max clock could easily mean that the clock settles to 4300-4350 MHz after some time, and this behavior has been observed in practice.

 

[Vattila]

You cannot round off the clocks like that. He clearly states that these are the maximum achieved clocks, not sustained clocks, so a 4375 MHz max clock could easily mean that the clock settles to 4300-4350 MHz after some time, and this behavior has been observed in practice.

Well, I've tried to hammer home that point over and over in this thread. Again, with the 3000 series, the sustained frequency the "clock settles to", is no longer the "Max Boost" printed on the box. It was with the 1000 and 2000 series, and the margin up to the real max boost clock was called "XFR". With the 3000 series, "max" means "max".

My poll is about that change — whether it is acceptable or not. As it stands, about 40% thinks it is acceptable, 12% thinks it is a lie/wholly unacceptable, and the rest would prefer the old meaning, with some XFR headroom so that the stated boost is seen in sustained single-thread workloads.

So, to me it boils down to what the number on the box should mean (momentary max or sustained max), while the fuss about the small rounding error is a distraction.

 

[Vattila]

Performance is not a prominent demerit of Ryzen 3000.

And I do not want to contribute to that unduly becoming so, hence why I remind casual readers that this discussion is almost inconsequential when it comes to the actual performance of the parts.

 

[VirtualLarry]

ence why I remind casual readers that this discussion is almost inconsequential when it comes to the actual performance of the parts.

Indeed.

 

[moinmoin]

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