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
peevee - Friday, September 20, 2019 - link
Sunny Cove is about just as new as Zen2 compared to Zen+, if not more.Only not "architecture" but "microarchitecture", architecture is the same Von Neumann prevalent since 1940s, or its Harvard variant prevalent since 1980s.
Jovec - Tuesday, September 17, 2019 - link
Might go to credibility and past behavior. Bulldozer couldn't maintain stock speeds under load (although the lawsuit was for something else IIRC).evilspoons - Tuesday, September 17, 2019 - link
Thanks for the informative article. The modal frequency being so close to the rating on all those 3000-series chips, BEFORE the +25-50 MHz fix, means it's not even worth worrying about. As consumers it's good to be informed and keep an eye on the companies we buy from, but there's no conspiracy here. Just confusion.Karmena - Wednesday, September 18, 2019 - link
Companies have to be held accountable on what they write on the boxes. Even if that is off by 25 out of 4400. Just be honest and write 4375.Atari2600 - Wednesday, September 18, 2019 - link
I'm sure your bright enough to be able to round up to 1 decimal place.What does 4.375 become when rounded to 1 decimal place?
ianisiam - Wednesday, September 18, 2019 - link
Except for the fact that 4400 isn't guaranteed. Like evilspoons said, it's just confusion.eva02langley - Wednesday, September 18, 2019 - link
You didn`t read the article obviously... these are not guaranteed. If you apply thermal paste and HSF pressure like an amateur, your results will not be in line with AMD numbers, but of course you will scream rip-off...Karmena - Monday, September 23, 2019 - link
Will not scream, just it would be nice to have every little bit. With ABBA, that is the case, in the end I need good mobo, good cooling, some memory OC prooves and I do not have to worry about CPU OC as that is brought to maximum performance by itself already.limitedaccess - Tuesday, September 17, 2019 - link
I feel there is one more aspect to this that wasn't fully addressed nor have seen it very addressed in discussions.According to your article (which matches previous understanding) is that Intel (even if they won't guarantee it the boost speeds) will bin CPUs so that every core is capable of running at the listed turbo speeds. In your example every core can reach 4.6ghz for the 9600k.
While AMD with Zen is only binning one core as capable of reaching the boost speeds. Although you list a more likely typical example in your article does this not in theory mean that a 3600x which can only reach 4.4ghz on a single core, while all other 7 can only hit 3.8ghz will pass (even if statistically such a CPU is extremely unlikely)? Or is their binning actually tighter then this but not disclosed anywhere?
This has been one of my concerns with respect the this type of situation going forward. It seems like there needs to be more disclosure/data points on binning requirements going forward.
ajlueke - Tuesday, September 17, 2019 - link
Thanks Ian!Could you highlight what the nebulous "limits of the silicon" are? I have noticed, that in low current (core) work loads with Zen 2, the performance seems to be fixed. I'm not reaching any temperature, EDC, TDC or PPT boundary, but changing the scalar, increasing fmax, or increasing the aforementioned limits. It seems something else is limiting the processor's ability to boost, but there is no clear indication what that is. I have observed the Stilt mention the Fitness Monitoring Tool (FIT) as a voltage limit baked into the silicon. Zen 2, tend to be at this limit, and it simply will not go higher regardless of what you set PPT, TDC and EDC at. Do you have any additional information on this limiter?
It does seem fundamentally different for overclockers. The voltage limit on boosting seems to be reached before thermal limits, meaning that Zen2 CPUs with the stock cooler boost the same as those with far better heat dissipation, at least in lightly threaded workloads. I guess this is a good thing for users overall, as everyone gets the same performance out of the box, but jarring for enthusiasts who are used to seeing much high numbers on their systems. I noticed the delta between the bottom 5th percentile and upper 95th percentile of the 3900X on userbenchmark was 12.4%. Incredibly tight, and demonstrates that there likely isn't much to ring out of these systems.