Comparable Benchmarks with Power Limit?

[subj3nn4]

Is there any source of information that compares newer and older CPUs under the same power limit, say 100W?

 

I am on a 9700K and I am struggling to find out if the performance gain of newer CPUs isn't completely or mainly related to their increased power draw. Ultimately, I want to know more about efficiency. I think this topic is very much overlooked or handled very inconsistently in hardware testing.

 

Thank you.

[Eigenvektor]

11 minutes ago, subj3nn4 said:

I am on a 9700K and I am struggling to find out if the performance gain of newer CPUs isn't completely or mainly related to their increased power draw.

While power draw is part of it, there are definitely clock speed and architectural improvements.

 

I've seen several reviews from HuB and GN that include numbers like perf/watt or perf/joule. Example from Gamers Nexus (timestamped): https://youtu.be/8KKE-7BzB_M?t=595

 

Not sure you'll still be able to find number for the 9700K, but if you have a watt meter you could run the same benchmark they did to get a comparable Watt hour number (basically watts x runtime).

[Tetras]

This is just opinion that I can't show you any benchmarks for, but I would say: it depends.

 

Intel versus Intel: the skylake architecture is capable of operating with much lower power consumption when gaming than 11th - 14th gen, but it's also not capable of pushing the kind of FPS that 11th-14th gen CPUs can. If you're using the GPU in a situation where it is operating @ max and the CPU can keep up, you're better off with skylake, but if your CPU can't keep up (e.g. 4090 @ 1080p/1440p), then 12-14th gen will produce more FPS per watt.

 

In terms of: which CPU is better at a given power limit. I'd say that 12-14th would be better because of their overall higher performance which gives them more headroom for efficiency gains.

 

Intel versus AMD: the X3D is just better in every way, performance, power consumption and efficiency.

[PDifolco]

1 hour ago, subj3nn4 said:

Is there any source of information that compares newer and older CPUs under the same power limit, say 100W?

 

I am on a 9700K and I am struggling to find out if the performance gain of newer CPUs isn't completely or mainly related to their increased power draw. Ultimately, I want to know more about efficiency. I think this topic is very much overlooked or handled very inconsistently in hardware testing.

 

Thank you.

Reviews are done using each chip TDP, it'l make no sense to review a 14900K at 65W  

Anyway, base power (per core) of chips didn't change much , your 9700K is 100W TDP for 8 cores, a 13700K is 125W but has 8 ecores in addition to pcores

AMD 7700 is 65W for 6 cores, so overall it's always 10W-15W /core (unless OCing)

So overall all the gain in on clocks and IPC, with OCing allowing to gain 5% for +40% power (on Intel chips!) but it's not representative

 

 

[Mark Kaine]

no that's not how it works, fabrication nodes keep getting smaller and more efficient all the time, while using about the same power  = they *are* faster and more efficient... AM4/5 often uses just 65w for example and most of them smoke your ancient and inefficient little intel with ancient architecture for breakfast  = )

 

28 minutes ago, PDifolco said:

Reviews are done using each chip TDP, it'l make no sense to review a 14900K at 65W

 

oh, of course it does, sounds like something DDTS would do!

 

[PDifolco]

30 minutes ago, Mark Kaine said:

no that's not how it works, fabrication nodes keep getting smaller and more efficient all the time, while using about the same power  = they *are* faster and more efficient... AM4/5 often uses just 65w for example and most of them smoke your ancient and inefficient little intel with ancient architecture for breakfast  = )

 

 

oh, of course it does, sounds like something DDTS would do!

 

Yeah and he'll do that gaming with a gt530 and a 200W PSU 

 

[subj3nn4]

2 hours ago, Eigenvektor said:

I've seen several reviews from HuB and GN that include numbers like perf/watt or perf/joule. Example from Gamers Nexus (timestamped): https://youtu.be/8KKE-7BzB_M?t=595

 

Thanks, that is helpful.

 

2 hours ago, Tetras said:

Intel versus Intel: the skylake architecture is capable of operating with much lower power consumption when gaming than 11th - 14th gen, but it's also not capable of pushing the kind of FPS that 11th-14th gen CPUs can

That's what I was assuming. Have you got a source for this?

 

29 minutes ago, Mark Kaine said:

AM4/5 often uses just 65w for example and most of them smoke your ancient and inefficient little intel with ancient architecture for breakfast  = )

Are you referring to the 9700K? Because this one idles at 10W and plays everything at 70W max. I just wondered if an upgraded CPU would even go below that.

 

Mind I play at 75fps with a 6900XT.

[Eigenvektor]

1 minute ago, subj3nn4 said:

Are you referring to the 9700K? Because this one idles at 10W and plays everything at 70W max. I just wondered if an upgraded CPU would even go below that.

You can look at sites like TechPowerUp, they usually also include power consumption and efficiency tests: https://www.techpowerup.com/review/intel-core-i7-9700k/16.html

 

If a CPU is more efficient at maximum load, it should generally be more efficient at a fractional load. I know that HuB/GN also often include numbers like frames per watt to measure efficiency, though that's more typically done for GPUs rather than CPUs.

 

1 minute ago, subj3nn4 said:

Mind I play at 75fps with a 6900XT.

It's unlikely you'll find benchmarks for this specific scenario. You'll need to find some general efficiency benchmarks and extrapolate from these.

[subj3nn4]

11 minutes ago, Eigenvektor said:

If a CPU is more efficient at maximum load, it should generally be more efficient at a fractional load.

 

See, that's exactly the question here. Newer high core chips seem to have a much higher overhead that under the same (lighter) load they *seem* to actually draw more power. I think it's safe to say they generally draw more power at idle as a fact.

 

11 minutes ago, Eigenvektor said:

It's unlikely you'll find benchmarks for this specific scenario. You'll need to find some general efficiency benchmarks and extrapolate from these.

Yes, I thought so. Thing is, there aren't many. Would be cool if this was touched more in the reviews. Thanks for your help!

[Tetras]

41 minutes ago, subj3nn4 said:

See, that's exactly the question here. Newer high core chips seem to have a much higher overhead that under the same (lighter) load they *seem* to actually draw more power. I think it's safe to say they generally draw more power at idle as a fact.

Yes, it is true. The "overhead" (and idle/light load power) on 6th-9th gen is generally superior to 11th-13th and Zen 4. With 10th gen, it depends.

 

1 hour ago, subj3nn4 said:

That's what I was assuming. Have you got a source for this?

Not specifically, no, but if you search for YouTube videos, e.g. 9700K/9900K versus 13700K, then if they include the CPU power figures, you can often see the kind of thing I'm describing.

[Mark Kaine]

9 hours ago, subj3nn4 said:

I just wondered if an upgraded CPU would even go below that.

 

yeah well generally as said, newer cpus will use about the same power, while being faster due to literally better efficiency. 

 

with another caveat that a lot of intels have like "turbo power modes" where they will use a lot of power for a short time... basically so it looks better on paper, but often users have to turn that off because it produces too much heat and isnt all that great for performance because of that. 

 

ps: so yeah, if you specifically ask about intel.... they are not efficient at all and need a lot of tinkering to be less power hungry, in general. 

 

Quote

In terms of power consumption, instead of the usual 20W average and 50W maximum on idle, the overclocked Core i9-13900K started to regularly hit anywhere from 50-70W during normal tasks and very rarely stayed around 30W

 50 maximum in "idle" is unheard of for the Ryzens i had, especially if its a 65w variant...

 

(ps2: and of course these writers have no idea what idle means "light tasks" =/= idle lol ¯\_(ツ)_/¯) 

[Mark Kaine]

8 hours ago, subj3nn4 said:

See, that's exactly the question here. Newer high core chips seem to have a much higher overhead that under the same (lighter) load they *seem* to actually draw more power. I think it's safe to say they generally draw more power at idle as a fact.

yeah, but the question is also, maybe is that offset by overall better system efficiency compared to older systems?

 

for example my whole pc uses ~65w during idle... i suppose that could be even more reduced with a more efficient psu (by a couple of % at least)

 

 

i mean sure you can look at the cpu draw in a vacuum but does that really make sense?  And i dunno,  how much does your whole system draw at "idle"? (for me idle is literally 0-1% system usage,  otherwise it's not "idle")

 

[PDifolco]

9 hours ago, Mark Kaine said:

yeah, but the question is also, maybe is that offset by overall better system efficiency compared to older systems?

 

for example my whole pc uses ~65w during idle... i suppose that could be even more reduced with a more efficient psu (by a couple of % at least)

 

 

i mean sure you can look at the cpu draw in a vacuum but does that really make sense?  And i dunno,  how much does your whole system draw at "idle"? (for me idle is literally 0-1% system usage,  otherwise it's not "idle")

 

What software is that ? It's a pretty neat view, HWInfo64 is much more arid looking  

[Mark Kaine]

3 hours ago, PDifolco said:

What software is that ? It's a pretty neat view, HWInfo64 is much more arid looking  

yeah, it's pretty neat looking,  its corsair icue lol... need it to set fancurve of my rmi psu... but i do use this regularly to check temps, and especially power usage (the rmi has sensors for this built in)

 

 

 

by the time hwinfo64 would load im already done checking with icue

 

[PDifolco]

4 minutes ago, Mark Kaine said:

yeah, it's pretty neat looking,  its corsair icue lol... need it to set fancurve of my rmi psu... but i do use this regularly to check temps, and especially power usage (the rmi has sensors for this built in)

 

 

 

by the time hwinfo64 would load im already done checking with icue

 

Huh ok, ICUe is neat but it's a damn bloatware !  

 

[Mark Kaine]

8 minutes ago, PDifolco said:

Huh ok, ICUe is neat but it's a damn bloatware !  

 

its not really... uses like 0.5% cpu at all times

 

hwinfo64 actually does have some latency issues (especially when reading rmi data ironically lol)

 

 

 

 

318mb, i can live with that, its less than a youtube tab? 

 

ps: generally i have the most unexciting system usage ever... 164 processes...? And i love it! 

 

 

(163 now...)

[YoungBlade]

The closest thing that I can find is this TechPowerUp article that tested the 12900K at 50W, 75W, 100W, 125W, 190W, 250W, and with Tau (250W then 125W after 56 seconds)

 

What this shows is that, at 75W, the 12900K performs about the same as a stock i9 11900K or R7 5800X for 1080p gaming. At 100W, it barely loses any performance.

 

So yeah, if you get a newer chip and power limit it down to even 75W, it will still way outperform your current 9700K.