The deceptive nature of such benchmarks is also troubling.
For all benchmarks, context matters. These days, most in-depth reviewers understand how to work with transient power and thermal limits with respect to time. For example, measuring the MacBook Air M1's peak speed/time vs. its lower sustained speed after it hits thermal throttling around the 5-7 minute mark.
If you limit power on current generation Intel/AMD processors, the perf/watt is pretty good; depending on the specific load, sometimes better, sometimes worse than M2. When you allow more watts, perf goes up, but perf/watt goes down as you see diminishing returns. (With some loads, you may also have a point where more watts reduces performance, whoops)
Geekbench isn't great as it doesn't even attempt to capture power usage, but that may not be available or accurate on all systems anyway.
Still slightly confused as to what exactly the “time window” means physically. With mine set to 1000us I still seem to get multiple seconds at full power. But it bumps down before the fan spins up so that seems good enough!
Looking forward to the inevitable realization tomorrow that this is going to make my computer unusable in zoom calls.
However we have many more layers now. I had a 80x24 serial tty exchanging ASCII characters with the computer back then.
According to https://vt100.net/heath/z19-om.pdf it can do 110 to 9600 baud. It was connected to some Z8000 based Onyx server. It was generally good but incredibly slow when more than four or five of us were compiling code for the tests at university class. Pascal, Berkeley p system or something with a similar name.
User problem: since I use Vim for everything, my expectations as far as responsiveness goes are probably not calibrated correctly. It is almost always very good, so the moments of badness are noticeable. The user (me) is unreasonably picky.
Technical: LaTeX syntax highlighting. It is I’m pretty sure, generally known to be a bit of a pain. Lots of nested environments that can go back dozens of lines and change how commands are interpreted. It is happy in a decent (not huge or anything) sized Fortran code, so I think config is mostly OK. Maybe I should look around and see if there’s a nice LaTeX plugin out there.
You have every right to have high expectations for performance with Vim
Depending on how hard you push it, Neovim can also be noticeably faster than Vim (notably Lua/LuaJIT being much faster than VimScript)
Furthermore there's the whole intentional delay on ESC exit from modes like an aborted search query... that can make it feel unresponsive but it's just a timeout. Try it, hit '/' in vim, then ESC to abort the search, count a full second of unresponsiveness.
For anyone looking for a good summary of the knobs available for AMD: https://github.com/FlyGoat/RyzenAdj/wiki/Renoir-Tuning-Guide
Or for Intel: https://github.com/lhl/linuxlaptops/wiki/2022-Framework-Lapt...
* https://www.kernel.org/doc/html/v4.12/admin-guide/pm/intel_p...
* https://manpages.ubuntu.com/manpages/kinetic/en/man8/x86_ene...
That doesn't sound right at all. Nothing in a PC system (absent water systems with huge reservoirs I guess) is going to buffer excess heat on minute-long timescales, not even close. There's literally nowhere for that energy to go; 125W for a minute would be melting the solder joints.
CPU throttling works on at most small-integer-second scales.
Based on the reported benchmark performance, it seems very unlikely that temperature-based throttling kicked in, and it's clear that the chip was operating well above 35W for at least a large portion of the benchmark run. So the PL1 and PL2 turbo power limits are the relevant controls at play here.
Again, the thermodynamic argument is fundamental here. You're saying that a "35W" CPU is "actually" drawing power equivalent to a 125W CPU for exactly the time of a benchmark, which is several minutes. That excess would have nowhere to go! There's no reservoir to store it. (Obviously the cooling system could take it away, but part of your argument is that the cooling system is only good for 35W!).
Most, if not all, subtests of GeekBench need less than 28 seconds, so it is quite possible for the entire benchmark to be run at an 105 W power consumption. Whenever a subtest finishes, the power consumption momentarily drops, which resets the 28 second timer.
If the computer has poor cooling, it may happen that when the CPU spends too much time and too frequently at an 105 W power consumption the junction temperature limit is reached, which triggers thermal throttling and the power consumption is reduced. This is a different mechanism, independent of the one that reduces the power consumption down to the nominal TDP after 28 seconds, or after another configured time.
Thermal throttling reduces the power consumption only enough to keep the temperature under the limit, so the power consumption may remain greater than the TDP until the 28 seconds pass.
I've done so, on many occasions, with actual power meters rather than trusting software power estimates. You really do commonly see a laptop's power consumption drop significantly ~28 seconds into a multithreaded CPU benchmark.
> (Obviously the cooling system could take it away, but part of your argument is that the cooling system is only good for 35W!).
I make no such claim that the cooling system is limited to 35W. I only claim that the default platform power management settings from Intel impose a 35W long-term power limit, unless the system builder has adjusted the defaults to account for whatever form factor and cooling choices they've made.
Perhaps you haven't realized that the turbo power limits will still kick in even if the CPU die temperature is not too hot—because they're not actually a temperature-based control mechanism?
Now I see where the disconnect was. You're right, if this was a laptop that could happen. It isn't, and it didn't.
The only relevance of the socketed nature of this part is that it is easy to put it in a normal desktop form factor where a big heatsink and possibly tweaked turbo limit settings can be used to generate misleading benchmark results. But it's not actually certain that this is what's happening; the Intel-recommended default behavior for this chip can plausibly produce the reported results—just not in any way that could be reasonably described as "35W".
[0] https://www.wolframalpha.com/input?i=%28100w+*+30+sec%29+%2F...
Since the benchmark in question (Geekbench 5) only runs for a minute or two, it does most of its work before the chip even attempts to throttle down to anywhere near 35W. An actual power measurement averaged over the entire benchmark run would yield a significantly higher value than 35W.