I think my hard upper limit is about a kilowatt PSU, but sized that large for spikes. (my current one is 850, and working fine.) I'm probably not willing to go past 500W sustained for the whole system.
I never checked what my 5700X/3060Ti is using, but my system is always whisper quiet and i barely have any fans installed. Something to check for the weekend for fun at least.
He means the jumps that happen around (or a bit over) 10W, so at the low end. Something notable happens there, although its only a single-digit jump.
Aside from the first part, which "Anonymous Chicken" answered, on the second part you're totally right, these are designed with an energy efficiency sweet spot somewhere in the vicinity of 1600 to 2200MHz. But I am not searching for the lowest energy operating point, even though I think that is an interesting and worthwhile goal. For my use case I just want my software to build faster. For that goal I am trying to investigate the marginal benefit of higher E-core clocks, compared to their cost. The cost of letting the E-core clusters rip at 4500MHz is that it reduces P-core clock speeds by several bins.
How does that affect performance on the P-cores? Would nerfing the E-cores actually make the system faster?
Sorry, i made an error. I would have actually liked to disable the P-cores for sessions where battery-life is the most important, but i know that this is not possible as one P-core must always be available. The best idea then seems then to be to find a sweet spot in terms of frequency and power, like @jwbaker mentioned but to find the lowest energy consumption while retaining usability.
When the E-cores are allowed to run at 4.5GHz they heat up the adjacent P-cores so much that they are bouncing off the 100º thermal throttler, which isn't great for performance at all. I strongly prefer to have the system run at an open-loop operating point that never hits 100º. Right now I have a long-term power envelope of 180W and short term peaks of 225W, and stability. I don't like my builds to crash.
What I meant was, can't you just disable the e-cores? That would let you test if there's any difference with the P-cores. If there is, then you can look into e-core throttling.
Ah, you said you were compiling code, two channels should do fine.
The 14900K hits 100º just doing an all-cores build of a big C++ project like Abseil. This is with a Noctua NH-D15. I don't really have any benchmarks laying around, except
openssl speed for what it's worth. Actually openssl speed ghash is a bit of a power virus, but doesn't reflect anything I would actually do.
Tom's Hardware said:
I am using a Thermalright contact frame, but I don't think it can alter the reality here. The part can generate more heat than the cooler can dissipate in the steady state.
Well, going back to my March 16th comment #363, I don't see the point of the higher power consumption. I have the CPU configured for total steady state power within the capabilities of the air cooler, and the AIOs I have owned all make an irritating sound at low power levels, which has turned me off the whole genre.
What irritating sound? I've been using AIOs for around ten years and don't know what sound you're hearing.
Most motherboards will let you set the pump to 100% all the time, so the noise will never change.
All AIO pumps make a low but audible pure tone sound in the vicinity of 100-1000Hz, which is just like having a tuning fork being struck constantly under your desk. Review sites that only talk about the weighted SPL are completely overlooking this issue. For example here's a "be quiet!" — irony alert — pump measured on a spectrogram. You can see the fundamental and overtones. It's the worst kind of noise a PC could make. Anyone would prefer a much louder SPL if it was broad spectrum noise. This is why I have gone back to preferring an air cooler even though their power dissipation potential is less.
My system is not under the desk - maybe your desk is acting like a speaker box amplifying the sound. Anyway, all I hear is the AIO pump gurgling on startup and then whatever sound the pump makes blends into the 5 cooling fans in the case, which grows to 8 when the GPU fans come on. I never hear anything resembling a tuning fork being struck. But some are more sensitive to certain sounds and only you can say what annoys you.
There are apparently CPUs where that is not the case. They are either degrading rapidly, or the binning process isn't catching them at the factory.
Yeah, today you should buy the vanilla model, in my humble opinion.
