Friday, June 5th 2020
Intel Posts 10th Gen Core Power Limit and Tau Values
Intel today updated the public data-sheet of its 10th Gen Core "Comet Lake-S" desktop processor to reveal precise power limit and tau values of each specific SKU. PL 1 or power level 1 is interchangeable with the processor's TDP as a power value. PL 1 is sufficient for a processor to sustain its base frequency (nominal clocks). For example, a processor with 65 W TDP has PL 1 at 65 W. PL 2 is what affords the processor the power to seek out boost frequencies. This value varies with between model to model, with the unlocked K/KF SKUs getting higher PL 2 values than the locked ones. The company also disclosed Tau. This is a timing variable that tells the processor how long (in seconds) can it stay within PL 2, before having to retreat to PL 1.
Source:
Intel 10th Gen Core Datasheet
49 Comments on Intel Posts 10th Gen Core Power Limit and Tau Values
should I have used numbers ?
I forgot to write to take a look at package power readings,sorry.
It's odd, but the 8 core 3700x got a lower power consumption than the 6 core 3600/x, while having slightly higher clocks and more cores. Probably having something to do with AMD using higher binned chip as you go up in the line-up.
I think how 3000 manages ccx's in gaming is to maximize the load on fastest cores within one ccx.the reason why so many times differences between 3600,3700x and 3900x are fractions in gaming is not actually bad core count scaling but the fact that each ryzen 3000 sku is a higher bin.
on 3700x you just have 2 more cores sitting at low load,it's not gonna contribute much to power draw especially at ryzen's efficiency levels.the main cores being more efficient make up for that.
I'm sorry Intel, this is not good enough for me.
I can’t say I’m shocked. If I did stick with intel I could cool them at least, since it doesn’t look too far off of the numbers I see with my first gen.:laugh:
wh per 60 sec idle
pclab.pl/zdjecia/artykuly/mbrzostek/2020/cpu_energy/idle_60sek.svg
wh per 25 minut run of blender - big scene.
pclab.pl/zdjecia/artykuly/mbrzostek/2020/cpu_energy/blender_goose_25m.svg
energy cost of a big blender scene.the difference between 10900k wo.power limits and 3900x is same as 6 min of idle time power draw.
pclab.pl/zdjecia/artykuly/mbrzostek/2020/cpu_energy/blender_goose.svg
small blender scene
pclab.pl/zdjecia/artykuly/mbrzostek/2020/cpu_energy/blender_hair.svg
3 min run of x264 encoding
pclab.pl/zdjecia/artykuly/mbrzostek/2020/cpu_energy/x264.svg
90 sec run of 7-zip
pclab.pl/zdjecia/artykuly/mbrzostek/2020/cpu_energy/7zip_small.svg
4 min run of stockfish
pclab.pl/zdjecia/artykuly/mbrzostek/2020/cpu_energy/pychb.svg
whole pc in witcher 3
pclab.pl/zdjecia/artykuly/mbrzostek/2020/cml_s/wykresy/power_3d.svg
whole pc during video playback - ridiculous
pclab.pl/zdjecia/artykuly/mbrzostek/2020/cml_s/wykresy/power_video.svg
Gee,I hope you've got a good place to put all those power savings.
whole article in Polish,use a translator of your choice
pclab.pl/art84541-2.html measuring peak power draw is wortless in the face of tests like this one.
peak power draw is lower on amd
typical power draw is same or lower on intel due to amd's boost
I mean,look at video playback power draw values.that 3000 is probably boosting like crazy.you'd need manual tuning,and end up on same level as intel in the end (maybe).their cpu resource management is way too aggressive - it ruins arch/node efficiency for normal tasks.
pclab.pl/zdjecia/artykuly/mbrzostek/2020/cml_s/wykresy/power_video.svg