|
|
|
|
|
|
by kllrnohj
1404 days ago
|
|
|
Why are you using peek instead of average? The average power over the test is what matters, not the instantaneous spikes. And did you miss the whole "configurable TDP" thing? The OEM gets to change the power targets. They aren't fixed. Saying the 4800U in a NUC uses power X, therefore all 4800U's use power X is flat out wrong. That's not how it works, different usages will set different power limits. |
|
|
If the TDP is only configurable up to 28w, but they're pulling almost 70w total system power (when there's simply no other real power users in the entire system), then there's a serious problem with their default "average" power usage not respecting TDP in any real way.
A lot of AMD laptops have historically dropped 30+% of their performance when you unplug them (a notable example would be the MS Surface 4s) so they can look better than they really are as reviewers tend to be lazy about this and test performance while plugged in then do a rundown test to push out a "review" within the couple day window they can still make ad money.
Comparing to other Zen 3 chips (I've gone into that elsewhere in this thread), top-binned EPYC uses 4-4.5w per core to do 2.45GHz base clocks (that's 32-36w for 8 cores at ~2.5GHz while they claim base clocks of 2.7GHz). If you lower that max TDP even further, those clocks go down even farther.
I disliked when Intel started going with TDP as a recommendation rather than a limit 15 years ago. I disliked when AMD followed suit so they wouldn't seem to be at a disadvantage (well, at less of a disadvantage as they were getting crushed at that time). I dislike the M2 Air which throttles back rather quickly.
The only good thing in recent times has been Intel's 12th gen adding a peak power consumption metric (no doubt because the difference between that and their normal TDP would have basically guaranteed a lawsuit).
> Why are you using peek instead of average?
Many (most) benchmarks are short-lived, but people infer long-term performance from these.
If a chip is getting N marks at 60w, long-term performance where power adjusts down to 30-40w isn't going to be anywhere close to that number.
M1/M2 when actively cooled can run at peak clocks/performance all the time. This means that performance expectations are in line with what reviews show. This is good for consumers (and they need to be more upfront about the air thermal throttling).