Discussion CPU boost frequency and marketing

[BigDaveX]

Long term ? I have 12 year old chips that are still running. Unless you overclock them using extreme voltage, they will be fine.

It's really going to depend on the process, and barring a repeat of Sudden Northwood Death Syndrome or anything similar, it'll probably be at least a few years before we find out exactly how resilient TSMC's 7nm process is when it comes to overvolting.

 

[lopri]

With that said, does it really matter? I mean, what are people doing where they are really only using 1 core with no background processes using cpu time. In today's age, I would worry more about the 2 core boost frequency and beyond. If the 2 core boost level fails in a very lightly threaded load environment, I'd be more worried.

It matters, IMO, because clock speed = performance. People run benches and see numbers changing run by run, and lose confidence in their purchase. It is a natural thing.

Having said that, I do observe a high variation among the BIOS implementation betweeen vendors as well as version numbers. So it may take some time before dust settles.

 

[beginner99]

Long term ? I have 12 year old chips that are still running. Unless you overclock them using extreme voltage, they will be fine.

Can you be more picky ?

So if I buy a brand new Corvette C8 with a new engine and have concerns about it's untested reliability, I can take data from a 2007 model VW Golf as a reference point?

 

[Markfw]

So if I buy a brand new Corvette C8 with a new engine and have concerns about it's untested reliability, I can take data from a 2007 model VW Golf as a reference point?

As far as reliability goes, electronics are in a different reality than cars, that analogy is wrong I even have a e8400 ES CORE 2 DUO chip (the dual core, not the new one) from 2007 and it still works perfectly. I also have an old AthlonXP and it works fine. I think I even have an 8088 chip that works fine.

 

[beginner99]

As far as reliability goes, electronics are in a different reality than cars, that analogy is wrong I even have a e8400 ES CORE 2 DUO chip (the dual core, not the new one) from 2007 and it still works perfectly. I also have an old AthlonXP and it works fine. I think I even have an 8088 chip that works fine.

None of them were made on TSMC 7nm and none of them had the power / voltage regulation features of Zen2. That was my point.

I'm just thinking out loud here. Not claiming anything.Why can't people reach 4.5 Ghz manual OCs (except the one guy) with non-exotic cooling? Yet the chips boost to that level. It's probably due to temporary high voltage. And high voltage can degrade CPUs. Add to this that ryzen 3000 series really seems to be at the clock celling out of the factory and hence if we see reports of these with reduced boost clocks in couple years (after cleaning heatsinks etc), would it really be that surprising? I don't thin so.
But again just thinking out loud. I'm not attacking AMD / Ryzen. I most likely will buy one and that is why I'm putting much thought into it.

 

[IntelUser2000]

Add to this that ryzen 3000 series really seems to be at the clock celling out of the factory and hence if we see reports of these with reduced boost clocks in couple years (after cleaning heatsinks etc), would it really be that surprising?

Degradation will happen though. Intel Fellow Guy Therien talks about it in the interview with Anandtech.

Skylake datasheets say after a few years the clock speed will degrade by a small amount as well(100-200MHz? I forgot).

It's just that when you are at the 100% maximum because the boost algorithm gets it up there, in a few year you get to see the drop.

 

[Vattila]

The following Reddit post argues that the boost algorithm in AGESA 1.0.0.3 had changes compared to version 1.0.0.1, and that the new aggressive behaviour made the CPU fire up more cores, even in single-threaded workloads, causing a slight degradation in the maximum boost frequency seen.

https://www.reddit.com/r/Amd/comments/cibwrv

 

[myocardia]

I'm pretty sure that if the vast majority of 9900Ks were only hitting 4.8-4.9GHz single core despite Intel advertising the maximum turbo speed as 5.0Ghz, people would be flaming Intel to a crisp. I don't think it's entirely unreasonable to hold AMD to the same standard.

You are 100% correct, and AMD's biggest defender on this planet would be the one leading the charge, and that is not an ASSumption. That said, I sure am glad that AMD finally worked out nearly all of the performance problems from their Ryzen line. Now, if I can only convince myself to hold off on buying a 3900X until they release the 3950X, I'll be a step ahead. I'm still not 100% sure which of the top two SKUs is the one for me, and am hoping that the comparisons, as well as the writeups that will obviously go along with them, will allow me to decide. For now, I've been able to tell myself that the UEFIs, AGESAs, et al isn't mature enough. I can only hope that they don't mature before the release of the 3950X, or I may have problems holding out!

As far as the OP is concerned, AMD just needs to lower the claimed boost of their top-of-the-line Ryzens to what they are actually able to boost to, while running actual software that does actual work. You don't get to change definitions, just because you have released a game-changing CPU.

edit: Only performance matters to people like us, not the actual frequencies involved, but AMD needs to remember that the vast majority of humans don't think the way that we think, at least when it involves their CPU frequency. They wouldn't be changing their claimed boost frequencies for people such as ourselves, but for that other 99% that may be buying/considering buying their CPUs.

 

[Vattila]

As far as the OP is concerned, AMD just needs to lower the claimed boost of their top-of-the-line Ryzens to what they are actually able to boost to, while running actual software that does actual work. You don't get to change definitions, just because you have released a game-changing CPU.

So you'd prefer the return of XFR — with the advertised boost number being the frequency seen in prolonged workloads (such as Cinebench Single-Threaded), and the real max frequency in the stock operating range being the advertised number + XFR, seen in short bursts — right?

Or do you want them to cap the operating range and leave the extra for overclocking?

 

[PotatoWithEarsOnSide]

Somewhere along the line, a number has to be placed on the box. What people are asking for is a bit more certainty in what they're getting, for what little clockspeeds are actually worth. This is one area that Intel do well in; you know very well what the stock performance and clocks across all cores will be. I appreciate that AMD's whole approach is different, but the number on the box has to be fair for the 95%+ that rely upon it as their metric.

 

[Vattila]

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.

 

[lopri]

Or do you want them to cap the operating range and leave the extra for overclocking?

I do not think that is an option available to AMD due to competitive environment.

 

[VirtualLarry]

Now, in that slide, are they considering "PBO" and "AutoOC" to be the SAME THING? Or are those separate BIOS options? Because on my B450, I have PBO, but not AutoOC, that I've seen. Is AutoOC, just an X570 option?

Edit: Also wanted to report, my R5 3600 CPU, boosts to slightly above 4Ghz, 75C-ish, 1.36V, with a load of 12 WCG threads (mixed tasks) in BOINC. This is with basically stock CPU settings in my B450-F STRIX ATX BIOS, with RAM set to XMP 3600 (CAS22 or something?), GSkill RGB RAM "for AMD". I have NOT tweaked my fabric, so IF clock == 1600, DRAM clock == 1800. Yes, there is a mis-match. But I don't have any instability at this setting. Perhaps I'll get adventurous, and set Fabric clock to 1800 again, and check for stability again.

Right now, though, I'm happy with the BIOS 2501, although a 1.0.0.3ABB wouldn't be a bad thing, I don't think. But this is straight-up usable.

 

[moinmoin]

Tip: Focus on performance, not frequency.

That's always a good idea.(Robert Hallock should have focused on that instead implying PBO offers room for additional frequency increase atop the boost.)

Now, in that slide, are they considering "PBO" and "AutoOC" to be the SAME THING? Or are those separate BIOS options? Because on my B450, I have PBO, but not AutoOC, that I've seen. Is AutoOC, just an X570 option?

As used on AMD systems they are the same thing, yes.

 

[Vattila]

Now, in that slide, are they considering "PBO" and "AutoOC" to be the SAME THING?

My understanding (if anyone is shaking their head reading this, please correct me) is that Precision Boost Overdrive has more levers than just the upper frequency limit setting. These other levers, such as power and current limits, can push up the average boost frequency seen in workloads, without going beyond the maximum frequency ceiling, i.e. still operating in the stock frequency range. I guess this is why they explicitly refer to the maximum frequency lever as "Automatic Overclocking".

In short, Automatic Overclocking is one of many settings within the Precision Boost Overdrive feature.

PS. PBO can even be used, not to overclock, but instead to reduce the power usage of the processor, usually with some sacrifice in frequency.

 

[coercitiv]

PBO can even be used, not to overclock, but instead to reduce the power usage of the processor, usually with some sacrifice in frequency.

This makes little to no sense. It may be an efficient way of overclocking, but it's certainly not more efficient ON than OFF.

 

[lopri]

I think Vattila meant that the CPU can be made more efficient by reducing total power available, which can be done using PBO by lowering PPT. It's akin to underclocking/undervolting to make the CPU run at more efficient voltage curve.

 

[Vattila]

I think Vattila meant that the CPU can be made more efficient by reducing total power available, which can be done using PBO by lowering PPT.

Exactly.

Precision Boost Overdrive lets you manipulate three parameters controlling the Precision Boost algorithm's ability to boost: Power (PPT), Sustained Current (TDC) and Peak Current (EDC).

In the first generation of Overdrive, the processor would still only operate within stock frequency range (which did include some XFR headroom, though, beyond advertised Max Boost).

With the third generation Ryzen, AMD added the Automatic Overclocking setting, which lets you raise the frequency ceiling beyond the stock frequency range (which is now the advertised Max Boost, as XFR has been dropped as a concept).

 

[AnnoyedGrunt]

These chips are too good to be brought down by marketing doubts and user disappointment due to misplaced expectations. IMO the advertised frequencies should be achievable in typical scenarios. I think if AMD advertised 4.5 as the single fore speed, people would be much more satisfied and would feel like they were getting a bit extra if the chip was reaching 4.6. Now you have some who feel like AMD was stretching the truth of what these are capable of.

In the end it is the performance that matters, but better to have a certain level of performance and a feeling of satisfaction and getting a little extra, rather than the same performance but the feeling of being a bit let down.

-AG

 

[CHADBOGA]

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?

 

[VirtualLarry]

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?

I really think that it has to do with the power demands of the load.

I can run 12 WCG (BOINC) tasks, and it hits 4Ghz+ (stock CPU settings on my R5 3600), but if I run PrimeGrid, which is a MUCH more demanding (AVX2) load, I'm lucky to stay at 3.9Ghz. Temps are also higher under PrimeGrid.

Edit: TBH, it would be nice, if these CPUs were a little more like Intel CPUs, inasmuch as they could be overclocked, such that you could have an all-core turbo clock, equal to the highest single-core Turbo clock (which my R5 3600 is listed as 4.2Ghz, which it actually has hit almost 4.3, like 4.27Ghz or so), but I wouldn't hold out hope for a "miracle BIOS" that will improve turbo clocks all that much substantially. I believe that the power-distribution / loading mechanisms at work here, are in fact hard-wired into the CPU, it's myriad network of power / voltage / temp sensors and whatnot, and it would be unlikely for them to behave substantially different, with a new BIOS, IMHO. OTOH, the voltage applied, at "idle" and whatnot, could probably be changed, and somewhat was, with the newest Ryzen (R) Balanced Power Plan, distributed with the newest Chipset Drivers.

Overall, once I got used to the new-ish behavior of the 3rd-Gen Ryzen CPUs, I'm happy with the performance that I'm getting. Even with 12 threads of WCG, with Task Manager graphs full, it still remains snappy for web browsing too, and Skype. I'm very happy with that performance.

 

[CHADBOGA]

I really think that it has to do with the power demands of the load.

I can run 12 WCG (BOINC) tasks, and it hits 4Ghz+ (stock CPU settings on my R5 3600), but if I run PrimeGrid, which is a MUCH more demanding (AVX2) load, I'm lucky to stay at 3.9Ghz. Temps are also higher under PrimeGrid.

Are you talking about all core speeds or just a single core?

I'm trying to work out if there is any point in equipping a case with expensive cooling(i.e. water cooling), if the CPU's won't go above a certain speed, regardless of whether you have outstanding cooling or merely good cooling.

 

[therealmongo]

Are you talking about all core speeds or just a single core?

I'm trying to work out if there is any point in equipping a case with expensive cooling(i.e. water cooling), if the CPU's won't go above a certain speed, regardless of whether you have outstanding cooling or merely good cooling.

I can tell you from first hand experience (custom loop), that in this moment in time, with only a beta bios for my motherboard that I am also waiting to see the same thing, i.e. if the CPUs will go above a certain speed!

What I can tell you from observing other peeps results is that a custom loop does help in sustaining higher all core clock speeds than the stock heatsink, of course this is a given.

For example I tested my CPU (3600x) with stock cooler and Prime95 29.8 build 5, after several minutes the temps (on AVX2 workloads) were around 95C and the all core frequency was around 3750 - 3800 mhz.

The same test with custom loop after running over 12 hours Prime95 29.8 build 5 blend with AVX2 workloads, max temp was around 87 - 89C and all core cpu frequency 4025 - 4075 mhz.

With current BIOS, it seems to me that something is holding the chip back, i.e. not allowing us to push to where we want to go.

Hopefully more mature BIOS will allow is to tweak these restrictions.

But from seeing one of the Stilts posts, it may be that fmax is fixed ......

 

[Markfw]

Are you talking about all core speeds or just a single core?

I'm trying to work out if there is any point in equipping a case with expensive cooling(i.e. water cooling), if the CPU's won't go above a certain speed, regardless of whether you have outstanding cooling or merely good cooling.

Om an inexpensive AIO, both of my 3900x's do over 4 ghz with all 24 threads@100% load at 75c

 

[DrMrLordX]

I'm trying to work out if there is any point in equipping a case with expensive cooling(i.e. water cooling), if the CPU's won't go above a certain speed, regardless of whether you have outstanding cooling or merely good cooling.

Heh! That's what all-core OC is for. I can do 4250 MHz in Prime95 or about 4400 MHz in anything else on custom water with very little effort. Nobody gets all-core turbo that high, and it seems like only a few doggedly single-core applications (like SuperPi or the ST versions of some benchmarks like Cinebench or Geekbench) benefit from default or PBO configurations. Even then I had to tweak the hell out of seemingly-unrelated settings to get single-core boost to hit 4.5-4.6 GHz in those ST applications.