I think the TDP figures in general are pretty bullshit, intel should be mandating a stricter adherence to PL2 and not have a PL2 that is of order 4x the PL1/TDP rating. They should separate more between low power SKUs and high power SKUs and motherboards should be forced to list supported power. Same goes for AMD, I really don't see why they need to come up with this bullshit formula for their TDPs when they clearly have a power limit number that is actually being used.
AMD's TDP formula was explicitly created as a reverse-engineering of Intel's formula so that cooler manufacturers and SIs could treat them equivalently in their design processes. Of course this has been undermined by Intel since being on a path of making TDP ever less meaningful of a metric, but still - that's not AMD's fault.
To be honest, imo Watts meaning Watts should be mandated either by the industry or by law (or both). A R3 3100 that eats 50 W under full load and reaches 70 C with the crappy boxed cooler on low revs in a slim case surely can't fall into the same TDP category as a R5 3600 that maxes out the 88 W PPT with a blink of an eye and can't be cooled without a tower cooler and lots of airflow. That's just a straight out lie from AMD.
This goes to both you and
@GorbazTheDragon: you're approaching this from the wrong angle, which either stems from a fundamental misunderstanding or from wanting something that doesn't exist. The issue: TDP is not a consumer-facing specification denoting power draw. It never has been. Historically it has been roughly equivalent to this, but this is more coincidental than intentional. TDP is a specification for SIs and cooler OEMs to design their cooling solutions and system designs around. If TDP was meant to denote power draw directly, it would for example be a guide for motherboard makers in designing their VRM setups - but it's not, and there are specific specifications (dealing with the relevant metrics, volts and amps) for that. You can disagree with how TDPs are used in marketing with regards to this - I definitely do! - but you can't just transfer it into being something it isn't.
TDPs serve as broad categories for SIs and cooler makers to design for, and is explicitly directed at large OEMs. This is where the 2/3-tier (historically ~95/65/35W) TDP systems come from - they're guidelines for designing cooling systems for three tiers of CPUs. There has always been variance within these tiers - just as there is with laptops, where a 15W i7 always needs more power than a 15W i5. Treating TDP as an absolute number for power draw has always been wrong. It's just happened to be roughly accurate at times. But it's also typically been far too high - like the R3 3100 you mention, or the i5-2400 in my modded Optiplex 990 SFF (nominally 95W, yet I've never gotten it past ~78).
This is where the current issues stem from - TDP used to be
reasonably close to
normal power draws, with non-high-end CPUs often coming in noticeably below that number. As technology has progressed, competition has tightened, and Intel has been stuck on 14nm(+++++++++++++) yet has needed to increase core counts, Turbo - which has
always been explicitly temporary, variable and potentially above TDP in power draw - has become more important, and has started pushing the silicon closer to its limits. Turbo clocks have divereged much further from base clocks than ever before (that aforementioned i5-2400 has a 300MHz Turbo on top of its 3.1GHz base clock), while the definition of TDP has stayed the same, and the categories have also stayed the same - largely due to Intel not being able to change these due to their OEM partners (if they changed the 65W class TDP to something more realistic like 105W, this would necessitate every OEM out there completely redesigning their SFF business systems to maintain base performance).
(Of course, we also need to take into account that stock (including stock boosting behaviour) power draws for CPUs are
much higher today than 5-10 years ago. An i7-7700k stuck pretty tightly to its 91W tDP in terms of idle-load delta power draw, and only scaled to ~120W when OC'd. These days AMD's 105W CPUs boost to 138/144W, and Intel's 125W CPUs boost to 170-250W.)
The reason for these issues is that Intel is using an OEM-facing design class denomination in consumer products without changing it or otherwise informing users what it means. This of course leads to a lot of confusion. But it also makes no sense for them to change those classes in the OEM world - which is easily 10x the size of CPU retail. A more sensible solution would be a consumer-facing "power class" or some such to denominate something closer to power draw. But that would ultimately look like they're suddenly saying their CPUs use far more power, which means that such a move would never be sanctioned by corporate and PR.
Of course this is only tangentially related to the issue at hand here - it's one root cause of it, but indirectly. The gap between base clocks (and power at those clocks) and turbo clocks (and the power at those clocks) is now large enough that due to Intel not enforcing their PL1, PL2 and tau specs with motherboard manufacturers, we now have a situation where the same CPU can perform very, very differently depending on the motherboard you put it in, which is not how things are expected to work. Intel could easily do this - but it would also make their CPUs look worse in reviews, so again, corporate and PR would never accept that. So instead, we get this quasi-sanctioned motherboard-dependent not-quite-auto-OC situation where the ultimate performance of a system is far more variable than ever before. Which of course sucks for end users and DIYers. But Intel (and AMD, though potentially a
tad less) ultimately doesn't care about us - they care far more about the OEMs and laptop makers that represent the majority of their sales.
Very true.
As a SFF maniac, I would much rather have no PL2 at all. I believe most, if not all motherboards let you customise these things, so you can set a PL2 that's the same as your PL1, but then who knows how much less performance you're getting out of your CPU. I'd love to see benchmarks that cover this, so I could decide where to upgrade. Knowing how much faster X CPU against Y CPU is on full power is of no use to me.
That would be great! Completely agree if reviews would cover this. At least TPU does test at both Intel official spec as well as unlocked power limits. But depending on just how SFF you go, there'll always be tuning (and the related stability testing) needed, which would drastically increase the reviewers' workload. And of course binning/silicon lottery outcomes dramatically affect this. So it's not very likely to happen.
But giving up boost isn't happening. CPU boosting represents
massive performance gains in everyday tasks such as web browsing and office work - even in very thermally limited systems. Which is why most OEMs let their CPUs constantly bounce off the thermal throttle point of the CPU - it doesn't harm the CPU or system in any way, but allows for far better responsiveness and performance. We as enthusiast DIYers tend not to accept this, nor the performance loss inherent to it (with the commonly accepted wisdom being that if you're bouncing off the throttle point in a DIY system, replace your cooler and gain performance!). This is doubly true if we also want silence - another factor most OEMs don't care about.
So as SFF enthusiasts - which is a niche and extreme approach to DIY PC builds, after all - we need to accept that a) we might not get peak performance, b) we'll need to tune our systems more, and c) there are no official specs denoting the information we need. That's life. And it's not going to change. Luckily SFF case designs are progressing at a rapid pace, allowing for much, much better cooling in smaller volumes than ever before (the number of <15l cases fitting 240 or even 280mm radiators today compared to 3-4-5 years ago speaks to this), so the tuning and compromises are shrinking, or at worst keeping pace with how power draw is deviating from the expectation created by TDP classes. But we need to accept that our use case is non-standard, and account for that in our builds. (On a related note, are you on the smallformfactor.net forums?)
Btw, did you test your 3600 when you had it at its 45W cTDP/Eco Mode setting? That might have been a better fit for your case/cooling.