Just watched this video and it certainly puts things into perspective:
https://www.youtube.com/watch?v=JHWxAdKK4Xg&ab_channel=TechNotice
Everyone has been making fun of Intel for their power hungry CPUs, but it seems there’s more to that story than meets the eye!
I’m pretty sure the sub-par real world power usage for AMD can entirely be thrown at the feet of their adoption of chiplets.
It’s a matter of perspective, “Real world” isn’t specific. I apparently am not living in the real world by your definition 
And power-hungry cores during load are more visible because of cooling requirements, temperature and noise. 30W vs 50W idle is the same regarding cooler and noise.
I know that AMD loses on low/idle over Intel. And I agree with @twin_savage that chiplets (especially IO Die) are the ultimate source for that. iGPU also contributes to that equation…they are not free in terms of die space and power.
Intel with 15th gen will have to face this as well, it’s basically chiplets…so I wouldn’t trust Intel to keep that benefit once they go chiplets next year.
Problem with idle consumption is that you can’t change much about it…TDP limits are just upper limits, not lower limits. So you can e.g. limit Intel to 80% the power and still have lower idle consumption.
My CPUs usually have something to do, so idle isn’t that important, but TDP and PPT are. And AMD is leading there. A 13900k@125W can almost compete with a 7950x in that bracket, but most SKUs just don’t have the amount of E-cores necessary to compete. P-cores on full load are just gas guzzlers. And you need 6-8 really fast Cores for gaming benchmarks.
I like the Video, but benchmarking the same old creator benchmarks over and over again as if there were no other benchmarks to get more than 4 cores running…Phoronix should be used more often as it covers many different things.
I have been trying to get @GigaBusterEXE to make a video for the channel on low power system building tips / tricks and comparison
It’s a topic we never payed much attention to…because I remember passively cooled CPUs and GPUs. And “low-power” home server is the flavor of the year, energy prices aren’t going down any time soon.
A nice compendium/overview on hardware/software is a good thing. Good hardware is one thing, but BIOS settings and locking CPU to higher P-states, CPU governor, general OS power settings,etc. are important too.
Yeah I mean my how lab etc is like 600w doing not a lot of work, I can afford it but its kind of crazy to think that its burning that power all day. (3 Host cluster 2 JBOD shelfs, 1 Enterprise 24 port 2.5/10g poe switch, Smart PDU do eat up a bit of power)
2 intel Machines (N100 and Xeon D1541)
2 AMD Machines both 5950X
Yeah it adds up. And usually old CPUs are fine in terms of performance for most things…but power consumption from a 14nm node compared to modern 5nm manufacturing node is staggering.
I scrapped my old Haswell Xeons mainly for power reasons…my new 7900x can do more with a single CCD at 1/5th the power.
Power draw is a major part for anything I buy, not just IT stuff. I do the calculations and often find the expensive more efficient products are cheaper in the long run.
The average desktop/gaming PC doesn’t run 24/7 so power isn’t that critical…if you game for an hour or two after work, getting a Titanium PSU is utterly pointless. And idle power draw figures aren’t useful either. That’s why we have 300W GPUs…it often doesn’t matter that much. It matters a lot if you run them 12h a day or 24/7
if your computer, server, TV, (insert random electronic device) is idle, do 1 of 2 things.
Monitors draw a surprising amount of power
Having CPU idle power optimized and leaving your 150W screen on isn’t really that much difference.
My 49" Samssung Odyssey makes noises after powering off because the plastic contracts when it is hot and reverts back when power is off. It’s quite the heater.
Mom taught me that. I plug out stuff every day. Every bit counts. I hope to get some home automation soon so I can flip wall plugs on/off with a click.
careful, the “home automation” stuff has standby power usage. in some cases it could be more than what you are “controlling”
another reason why i assumed Z-WAVE would win out over time, compared to ‘wifi’ devices.
Something about this video rubs me the wrong way. I don’t know what it is, but when you jump over to the laptop side, you can directly compare 12/13th gen to AMD on the framework laptop. They’ve got drop in motherboards that use almost identical cooling in the same enclosure. At the end of the day, the AMD variant gives you 3 hours more battery life while browsing the web on the 55wh pack, which is a direct indicator of mixed load perf per watt.
I don’t know what’s going on with this test, but something doesn’t seem correct, when you consider that both manufacturers use very similar CPU architecture across desktop and laptop skus.
I think alot more OEM power profile tuning goes on with laptops which muddies the comparison.
On the test that arstechnica did, the Intel framework 13 laptop could last almost 2 hours longer than the AMD framework 13 laptop’s 7 hour run time on a pcmark battery test. It could be argued the AMD 7840U was more powerful than an Intel 1370p though.
This will be interesting because the way Intel says they are organizing the 15th gen chiplets, they should have as good power consumption as monolithic since entire chiplets will be shut down when not in use.
Interesting!
Agreed. That’s why I was bummed when Sapphire Rapids weren’t on a new lithography.
The problem with this is as the video was alluding is that most of the time we’re not actually doing anything. When I’m reading a .pdf, the computer is idle, but I can’t read it with the computer off. So low idle is certainly not a useless thing. And the problem with cycling servers is you’re using up cycles of your hard drives. If you’re running $50 Barracuda drives, then replacing them when they break isn’t much of a problem. When you’re buying 20TB drives, you want them to last as long as possible.
CPU architecture, yes, but not the CPU itself. AMD laptop CPUs are still monolithic, while most of their desktop counterparts are chiplet. The interconnect, the I/O die, etc. consume power, which is why the desktop counterpart are idle at higher power consumption, despite Zen 3/4 cores being more efficient than current Cove cores. (And why there’s disparity between desktop and laptop results.)
BTW I’m surprised that desktop CPU Intel i3-13100 (60W TDP) in mainboard Asus Pro H610T D4-CSM (+1x RAM + 1x m.2 SSD) is able to achieve only 
2.3 — 2.8 W idle power consumption (it was unbelievable for me 
)
see my post with details here: Nutral's HomeLab blog - #36 by xnd if you don’t believe
It is? When i am reading a pdf i have 42 other tabs open, music or a YouTube video playing, and at least one thing disassembled next to me.
And it was not really meant to say to turn off your server, more to say find it other tasks.
This is a really important point that I didn’t realize until you pointed it out.
Good point.
I’d still happily trade idle consumption for peak.