Reaching for Turbo: Aligning Perception with AMD’s Frequency Metrics
by Dr. Ian Cutress on September 17, 2019 10:00 AM ESTA Short Detour on Mobile CPUs
For our readers that focus purely on the desktop space, I want to dive a bit into what happens with mobile SoCs and how turbo comes into effect there.
Most Arm based SoCs use a mechanism called EAS (Energy-Aware Scheduling) to manage how it implements both turbo but also which cores are active within a mobile CPU. A mobile CPU has one other aspect to deal with: not all cores are the same. A mobile CPU has both low power/low performance cores, and high power/high performance cores. Ideally the cores should have a crossover point where it makes sense to move the workload onto the big cores and spend more power to get them done faster. A workload in this instance will often start on the smaller low performance cores until it hits a utilization threshold and then be moved onto a large core, should one be available.
For example, here's Samsung's Exynos 9820, which has three types of cores: A55, A75, and M4. Each core is configured to a different performance/power window, with some overlap.
Peak Turbo on these CPUs is defined in the same way as Intel does on its desktop processors, but without the Turbo tables. Both the small CPUs and the big CPUs will have defined idle and maximum frequencies, but they will conform to a chip-to-chip defined voltage/frequency curve with points along that curve. When the utilization of a big core is high, the system will react and offer it the highest voltage/frequency up that curve as is possible. This means that the strongest workloads get the strongest frequency.
However, in Energy Aware Scheduling, because the devices that these chips go into are small and often have thermal limitations, the power can be limited by battery or thermals. There is no point for the chip to stay at maximum frequency only to burn in the hand. So the system will apply an Energy Aware algorithm, combined with the thermal probes inside the device, to ensure that the turbo and workload tend towards a peak skin temperature of the device (assuming a consistent, heavy workload). This power is balanced across the CPU, the GPU, and any additional accelerators within the system, and the proportion of that balance can be configured by the device manufacturer to respond to what proportion of CPU/GPU/NPU instructions are being fed to the chip.
As a result, when we see a mobile processor that advertises ‘2.96 GHz’, it will likely hit that frequency but the design of the device (and the binning of the chip) will determine how long before thermal limits kick in.
Facebook
Twitter
RSS
144 Comments
View All Comments
0siris - Tuesday, September 17, 2019 - link
In the end, this doesn't change anything for those of us who read reviews and decided based on those if the performance in our workloads was acceptable for the price AMD was asking. But if your marketing is going to dumb the performance down to a (set of) number(s), then you will also have to deal with the complaints of people whose understanding doesn't go beyond those simple numbers. Looks like the "bigger number!" marketing we've seen with is finally starting to hurt companies.It seems completely logical to me that boosting algorithms will become ever more refined to ensure optimal performance at all times. That's a good thing for all of us.
WaltC - Tuesday, September 17, 2019 - link
The so-called Bulldozer lawsuit was settled by the lawyers who brought the suit (not just by AMD, as is erroneously often implied), who agreed to take the money and run rather than chance the loss of their suit, ultimately, because AMD surely would have won, anyway. AMD agreed to the payout as it amounted to less money than it would have cost AMD to actually go all the way and win the suit--this is why companies settle spurious lawsuits--and is the only reason they do. Additionally, in writing, the suing attorney's agreed with AMD that AMD had done nothing wrong--no false advertising, etc. Whenever this "issue" *cough* is raised, it is never explained fully--with a sort of mindless "AMD did something wrong" emphasis. Of course, nothing about this *old* lawsuit has anything at all to do with the burst frequency issue which AMD recently corrected in the ABBA AGESA is distributed a week ago--so what is the purpose of mentioning a lawsuit that the suing attorney's have dropped and agreed with AMD that AMD had done nothing wrong?....;) Clearly, the suing attorneys believed they could not win, else they would not have agreed to such a tiny settlement, eh? Or agreed to signing a statement clearing AMD of any wrongdoing whatsoever. (There is something wrong with the civil system when this kind of garbage is allowed to clog up the courts, imo, but that's another issue entirely.)I really don't understand why some people don't understand that "teething" issues with a new architecture release are common--and it doesn't matter whether AMD makes the CPU or Intel makes it--new architectures all demand a period of software adjustment--six months to a year is common for both companies when new architectures are released! I marvel that this needs explanation...;) It's been so long since Intel has shipped a ground-up new architecture, though, that likely few remember Intel-architecture teething problems. I hate to see all the red faces appearing when and if Intel ever makes it to 7nm with a new architecture....)
AMD has also made it clear that the AGESA bugs caused the loss of boost MHz, in all cases. They've surely fixed it--my 3600X boosts now like it did with AGESAs 1002/3 in bios releases for my x570Arus Master--and a tad faster--to 4.424GHz, instead of 4.401GHz. (Wowee...;)) Later AGESAs up through ABB cut max boost by ~125Mhz. Happy to report that has been remedied with the newer AMD AGESA ABBA bios for my x570 Master.
hansmuff - Tuesday, September 17, 2019 - link
Outrage culture and the lawyers are just cashing in on it. Can't blame them either.WaltC - Tuesday, September 17, 2019 - link
Yes! I really don't blame the starving lawyers quite as much as I blame the judges who greenlight spurious stuff like this. I mean, it creates the impression that either the judges don't understand the issues involved, or else that it's a racket with "everyone" hoping to share the ill-gotten gains...;) Because in the end--it's we the customers who pay for all of the spurious junk that isn't thrown out on day 1--just as much as if it was a direct tax. This reminds me so much of the infamous class-action suit they tried and failed to throw at the HDD makers a few years back. The lawyers alleged that everyone who bought a hard drive was "mislead" by false advertising because the HDD makers decided to call a "megabyte" 1,000,000 bytes instead of 1,024,000 bytes--*even though* all of the HDD makers posted that information not only in all their product advertising, but also on the HDD product boxes! The judge in that case was savvy enough to throw it out on Day 1, IIRC. In the Bulldozer case, of course, the lawyers will earn $6M-$8M (or more) of the $12M and the Bulldozer owners might be lucky to get $10 each....;) What a racket!...;)Targon - Tuesday, September 17, 2019 - link
Intel hasn't had a new architecture for so long at this point, many people have forgotten about the problems.WaltC - Tuesday, September 17, 2019 - link
Shades of Prescott..;)Korguz - Tuesday, September 17, 2019 - link
Targon, not according to one person that comments on here, sunny cove is a new architecture, and is completely different then all the cpu's intel has made over the last 5 or so years.but yea.. its time intel stopped milking the same architecture, and its customers, and came out with something that is actually new
Phynaz - Wednesday, September 18, 2019 - link
Why do you care, you would never buy intel?FunBunny2 - Wednesday, September 18, 2019 - link
"its time intel stopped milking the same architecture, and its customers, and came out with something that is actually new"'architecture' is constrained by the Law's of God, aka maths, so once the 'best' has been found, that's it. IMHO, all these 'architectures' amount to:
1) pulling previously off-chip function on-chip
2) rearranging them to implement Cray's shorter wires
3) adding more transistors to make certain functions bigger
not really much 'architecture' progress to any of that. in any case, how much more 'architecture' does it take to do word processing, spreadsheets, and e-mail? you know, the Three Killer Apps of the PC?
Korguz - Wednesday, September 18, 2019 - link
once the best has been found ??? intel hasnt released a new architecture because they have had no reason to. but i bet now, with Zen, intel is doing just that, working on something that is actually new and not based on the same cpu we have had from them for the last 5 or so years.