Most likely, many have heard about Intel Turbo Boost technology (but not about AMD PBO, but it will be discussed below): they say this is a magic feature that allows processors to accelerate and work faster on their own. In part, this is really true – the technology is magical, but the magic here works more towards the manufacturers of CPUs than towards ordinary users. And taking into account the fact that there is very little information about the operation of auto-acceleration functions on the Internet, especially on official websites (for obvious reasons), you have to collect it bit by bit, depending on how different processors work in different conditions.
But overclocking is not real!
Modern processors from Intel (we’ll talk about AMD separately because it’s still more confusing) have many limitations – which, surprise, may not be fulfilled if the motherboard manufacturer disables them by default in the BIOS. The first and most important limitation is the maximum temperature, about 100-105 degrees for various destopny processors. When approaching it, the CPU will begin to throttle, in other words, seriously reduce the frequency in order to keep the temperature within acceptable limits. If even at the minimum operating frequency of 800 MHz, the processor cannot cope with overheating, it either crashes (at this moment the picture on the monitor freezes), or the board reboots.
It would seem that this is a great limitation that works perfectly and does not allow the processor to get into silicon Valhalla ahead of time. In fact, everything is somewhat more complicated. Firstly, temperature sensors inside the CPU chip are not everywhere, and if the maximum recorded heat, for example, is 80 degrees, then the processor may well have a place that heats up to 85. Secondly, the crystal itself is heated unevenly: the hottest places, of course, are the nuclei. But the integrated graphics, various controllers and the cache can warm less than a dozen or two degrees – especially if the cores are heated at a hundred degrees. Of course, such temperature differences within a single crystal are far from useful for silicon.
So there is nothing surprising in the fact that Intel decided to introduce new limits. The most famous of them is TDP or Thermal Design Power. This is a very tricky figure: they say that it is precisely so much heat that the cooling system should take away from the processor. In practice, it is still more interesting: it is to this figure that the heat dissipation of the processor tends to under continuous load. And then the first “oops” happens: take, for example, the popular mobile Core i5-8250U. It has a native frequency of just 1.6 GHz, but Turbo Boost allows it to accelerate to 3.4 GHz. It has a TDP of 15 watts, which allows you to put it in ultrabooks – well, let’s conduct a stress test and check what the real frequency will be with a long load:
2.4 GHz Formally, all is well – the frequency is higher than the native 1.6 GHz, and one and a half times. But, on the other hand, this is not 3.4 GHz: 20-25% percent is lost, which is also quite significant. Okay, let’s start the game now – it loads the processor less, it “fits” at 15 W and operates at a maximum frequency of 3.4 GHz.
So we see the first cunning, which is hidden in the phrase “up to 3.4 GHz”: after all, 2 GHz – “before.” And 2.5 is also “before”. But this is only the beginning – most manufacturers of desktop motherboards pretend that they do not know about TDP and trivially disable this limit! What does this lead to? Moreover, the 6-core Core i5-8400T, which formally has a heat dissipation of 35 W, begins to consume 60 and 70 in some tasks, without reducing the frequencies. It would seem – here it is happiness, productivity does not fall? So yes, but not really: if a 35 W aluminum box cooler easily takes away, then it may not be able to cope with 70 W. Of course, as I wrote above, the processor is unlikely to burn out from overheating, but it is unlikely that you will be satisfied with constant slowdowns in work. There are obviously two ways out – either enable the TDP limit in the BIOS, or buy a more powerful cooler.
Of course, this is a slightly far-fetched problem: for the most part, on the contrary, they turn off various limits and energy-saving functions in the BIOS so that the processor can operate at the maximum possible frequency. But this shows very well that the manufacturers of boards wanted to sneeze on Intel specifications (and AMD too).
The second cunning is even more interesting: for example, in some processors the indicated Turbo Boost frequency is achieved only … with a single core. So, the same i7-8550U with a load on one core can operate at a frequency of up to 4.0 GHz, two – only 3.8 GHz, well, and all four can not “boost” above 3.7 GHz. So even if you create ideal conditions for this processor – 4 GHz with a serious load, you will never see. Generally speaking, you will never see this figure, because in the modern world a task that will load only one core still needs to be looked for, and in realistic tasks with a high load and in the absence of constraints, the real frequency will be at the level of 3.7-3.8 GHz. Moreover, formally, Intel again does not find fault: is this frequency higher than the native 1.8 GHz? Above. Well, the fact that you can’t do a single-core load is your problem.
But back to laptops and ultrabooks. Intel is well aware that most of the serious workloads are short: how long does it take to open a program? Seconds 5-10, no more. It takes even less to load a page in a browser. At the same time, the cooling system has a great thermal inertness: in order to warm it up to hundreds of degrees, it will take no less than several tens of seconds, or even whole minutes. Conclusion – for a short period of time, you can “forget” about the TDP limit and allow the processor to operate at maximum frequency: obviously, this greatly improves the responsiveness of the system in real-world tasks.
In numbers, it’s done like this: the so-called Turbo Time Limit usually lasts 28 seconds, and at this time Short TDP works, which can reach 30-50 W: this is guaranteed to allow the processor to use the maximum frequency even under severe load with vector instructions. After these 28 seconds, Long TDP comes into play – those same 15 watts, and the CPU frequency is seriously reduced. And if the cooling system does, then in this mode the laptop will work, figuratively speaking, forever.
Okay, everything seems to be fine: Long TDP works, the processor does not overheat – idyll? Unfortunately no. Most laptops have a common cooling system for the processor and discrete graphics card. And, obviously, quite often there are tasks (for example, games) that seriously load both components of the system. In this case, usually, the maximum temperature of the GPU is still lower than that of the CPU, that is, the video card starts throttling earlier: and this, of course, negatively affects the frame rate in games. Output? Since the cooling system of the processor and the video card is common, then why not slow down the processor – it is rarely when it runs 100% in games, so a slight decrease in its frequency and heat dissipation should, in theory, not affect game performance, and at the same time, the video card will not throttle.
This feature is called BD Prochot, and, unfortunately, it really is a miscalculation. The problem is that if it is activated, the processor responds to overheating of the video card in the same way as to its own – in other words, drops the frequency up to 800 MHz. Obviously, this leads to a sharp decrease in heat dissipation and temperature of the GPU, so that the processor frequency is quickly restored to its previous level of several gigahertz due to this. And this katavasiya begins to occur once every few seconds: it must be understood that the frequency drop to 800 MHz is felt only like a frieze. That is, games begin to stably slow down once every few seconds – as they say, a pleasant game. Fortunately, this feature is easily disabled in the free ThrottleStop utility: of course, the video card will overheat and throttle, but it usually does it more smoothly, reducing the frequency by only a small amount. So yes, it will lead to some drop in fps, but it is still more pleasant than constant juggling.
But then it becomes funnier and worse at the same time. Of course, processors have been able to operate in a huge frequency range for more than a dozen years, often lowering it below their native one for energy saving. It is also obvious that the lower the frequency, the lower the voltage can be applied to the CPU and it will remain stable, and it will consume less energy. The so-called frequency-voltage table for each processor compatible with the board is in its BIOS, and, fortunately, board manufacturers usually stick to it.
That is, voltages, powers, and frequencies are set rather rigidly – what is changing? That’s right, this is current (I recall that power is currently multiplied by voltage). And, of course, Intel also sets a limit on it: in the case of the i5-8250U it is 64 A (IccMax parameter). Given that the voltage at 3.4 GHz is of the order of 1 V, we get a maximum heat release of no more than 64 W: the processor, obviously, will not be able to reach it at all (Short TDP is usually below 50 W), this raises a logical question – well, why did Iccmax need to be introduced if it never limits the processor? Moreover, by the way, in desktop boards this parameter is often set generally to 255 A – at a voltage of 1.2-1.3 V this will give an astounding 330 W: obviously, to put it mildly, this is far from the actual consumption of desktop CPUs.
And then even more interesting. I wondered which MOSFETs are used in the power circuit of my i5-8250U. It turned out that this is Sic634 – and they have a maximum current of 50 A and a peak current of 55. That is lower than what is required by the Intel specification. Of course, I decided that this saved Xiaomi, but then I found that exactly the same transistors are used in expensive Dell XPS 13 with exactly the same CPU. Of course, 50 A at a voltage of 1 V will give us as much as 50 watts – this is slightly more than Short TDP, which in my case is 44 watts, and many times longer than Long TDP of 15 watts, but still that manufacturers “forget” about the specifications Intel is scary when planning the power part of the boards.
Well, the cherry on the cake: of course, we will talk about Intel Turbo Boost 3.0. Modern processor engineering can be very accurately described with one phrase: “the third grade is not a marriage.” Top solutions for the LGA2066 socket can have as many as 18 cores and even everything in one crystal. Of course, the chance that all the cores will accelerate equally well is extremely low – there will always be 1-2 more successful cores capable of taking higher frequencies. So if Turbo Boost 2.0 accelerates all the cores or any of the frequencies specified in it, then after installing a powerful HEDT processor on the board, Turbo Boost 3.0 technology determines the best cores in it and only allows them to “boost” more than others. The difference can hardly be called significant, it is usually at 200 MHz, but we still see how Intel launches “semi-rejected” crystals, where different cores are accelerated in different ways.
So how does Intel auto-acceleration actually work?
Honestly – as the motherboard manufacturer decides, and, I think, you already understood this yourself. But on average, everything happens like this:
- There was a load – the processor’s heat dissipation exceeded Long TDP. The board starts using the Short TDP limit and starts the Turbo Time Limit. If at the same time the IccMax limit is exceeded, the processor will begin to reduce the frequency in order to keep within its scope, but in reality, this happens a little more often than never. If there is support for Turbo Boost 3.0, then a couple of cores “boost” stronger than others.
- Heat dissipation became less than Long TDP before 28 seconds of Turbo Time Limit operation? Well, reset this timer and wait for a new load jump. If not, the Long TDP restriction is triggered, the processor loses its frequency and starts working like this forever.
- Oh no, it’s not forever – our video card has warmed up pretty well, it needs to be cooled. BD Prochot turns on and the processor frequency rushes to the level of 800 MHz, you are greeted by lags. Fortunately, as I wrote above, this function does not work on all devices and is easily disabled.
- Oh, the manufacturer saved on the cooling system, we have CPU overheating – yes, as you already understood, the frequency drops again. But at the very least, 15 watts can be taken away by most COs, so this usually doesn’t come to that.
There was a load – the processor’s heat dissipation exceeded Long TDP. The board starts using the Short TDP limit and starts the Turbo Time Limit. If at the same time the IccMax limit is exceeded, the processor will begin to reduce the frequency in order to keep within its scope, but in reality, this happens a little more often than never. If there is support for Turbo Boost 3.0, then a couple of cores “boost” stronger than others.
Heat dissipation became less than Long TDP before 28 seconds of Turbo Time Limit operation? Well, reset this timer and wait for a new load jump. If not, the Long TDP restriction is triggered, the processor loses its frequency and starts working like this forever.
Oh no, it’s not forever – our video card has warmed up pretty well, it needs to be cooled. BD Prochot turns on and the processor frequency rushes to the level of 800 MHz, you are greeted by lags. Fortunately, as I wrote above, this function does not work on all devices and is easily disabled.
Oh, the manufacturer saved on the cooling system, we have CPU overheating – yes, as you already understood, the frequency drops again. But at the very least, 15 watts can be taken away by most COs, so this usually doesn’t come to that.
The result – in the mobile segment, instead of the maximum Turbo Boost frequencies of 3.5-4 GHz, most often with real work you can only see 2-2.5 GHz: of course, this is still higher than the native frequencies that Intel dropped below the baseboard, but, of course, this is far from the level that most users expect.
AMD Precision Boost Override – twist, torment, do not want to disperse
As we recall, after the 2011-2013 “bulldozer” FX, which was not able to compete with the Core i7, AMD decided to stop such serious experiments and return to the Zen core architecture with normal hyperthreading cores, which in this case is called SMT. Well, the idea, as we know, is a good one, and Ryzen processors are snapping up like hotcakes, quickly moving Intel from the sales leader (> 80% at the beginning of 2017) to the level of catching up (about a third of sales at the moment).
At the same time, AMD perfectly understood two things: firstly, it is not in a position not to sell a half-marriage (especially since Intel is engaged in this), so cheaper Ryzen without letter X operate at frequencies of 150-200 MHz lower than their “X” counterparts, even under acceleration. Secondly, with single-core performance, Zen and Zen + didn’t do very well, so it had to be lifted by all means. So the PBO technology was born, which, on the one hand, is very similar to Turbo Boost, and on the other, it is radically different.
In general, in the case of Intel desktop processors, only one limit is important – in temperature, all the others are either never reached, or are usually disabled by default in the BIOS, so that the processor works stably at its maximum Turbo Boost frequency for all cores, that is, on different boards, the CPU will generally show the same level of performance.
In the case of AMD, auto-acceleration technology has as many as four actively working parameters:
- PPT Limit (Package Power Tracking) – restriction on processor consumption of energy in watts.
- TDC Limit (Thermal Design Current) – limitation on the maximum current supplied to the processor. It is determined by the cooling efficiency of VRM on the motherboard.
- EDC Limit (Electrical Design Current) – limitation on the maximum current supplied to the processor. Determined by the VRM circuitry on the motherboard.
- Precision Boost Override Scalar – the coefficient of the dependence of the voltage supplied to the processor from its frequency.
The first parameter, in principle, is understandable: this is a restriction on TDP, everything is logical here. The second and third are already more interesting: if Intel simply has a current limit on the processor side, then the efficiency of the batteries on the board is also taken into account. The fourth parameter is even more fun: there is an internal coefficient, the principle of which AMD does not disclose.
As a result, we observe a somewhat unexpected picture: the Ryzen 7 3800X, which has a maximum frequency on paper of 4.5 GHz, can actually work both at higher (4.55 GHz) and much lower (4.375 GHz), and, what’s the funniest thing, There is no dependence on the VRM board, a cheap solution from ASRock is by no means at the very end of the list:
It should also be understood that such frequencies are achievable only with close to the single-threaded load. An attempt to load all the cores leads to the fact that the frequencies fall to the level of 4.1-4.3 GHz – as we remember, Intel has the same situation, but for one thing: in the case of desktop processors in the “blue” any core can accelerate to the maximum Turbo frequency Boost, so, by and large, in overclocking this frequency is almost always achievable for all cores – this is what we see if we do not take extreme benchmarks.
But AMD doesn’t do this: as I wrote above, in order to keep prices low, they decided to use the same trick that Intel uses in Hi-End processors – namely, to mark good cores, and only they “boost” to those indicated in high-frequency specifications at 4.5+ GHz. Obviously, in this case, all the cores will not get to such a frequency – this will require too high voltages, so this results in an unpleasant situation: if Intel has the indicated maximum Turbo Boost frequency, it is not achieved only in mobile processors that are not always used for a serious load, then at AMD this happens even on desktop processors, which makes theoretical maximum frequencies a fiction and nothing more.
Moreover, in the Ryzen Master utility, the best kernels are indicated with asterisks – as it turned out, this official program from AMD puts them randomly, in fact, completely different kernels turn out to be the best:
Personally, I can describe what is happening now on the processor market with just one word – tin. Lime frequencies, broken limits, non-fulfilling specifications – there is a serious feeling that the Indians are coding, and the Chinese are soldering. So when buying a new laptop or a motherboard with a processor, carefully study their actual performance and frequencies, because even in the desktop segment there can be an extremely striking difference when working, for a second, in completely default mode.