it was ~1100 for the cpu and motherboard.
the power usage on that epyc is only 155w tdp, so not horrible at all
New Homelab build, should I go 14700k or 14900k?
- Thread starter jkimmerling
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
the power usage on that epyc is only 155w tdp, so not horrible at all
I mean, compared to a 30w CPU which would run your NAS fine it's a pretty big difference
and that's CPU power max not CPU+MOBO+RAM+ETC... but either way, with 4x drives your cpu\moob\ram wil use same or more power for sure.I went with 14700K. My use case might be a bit different, since I run plex and do want Intel Quick Sync.
The benefit with Intel (as far as I have seen) is that you will have somewhat lower idle power consumption compared to some of the other CPUs. If you don't need QuickSync maybe it makes sense to go with another CPU.
Idle power consumption is for me at least important as well.
The benefit with Intel (as far as I have seen) is that you will have somewhat lower idle power consumption compared to some of the other CPUs. If you don't need QuickSync maybe it makes sense to go with another CPU.
Idle power consumption is for me at least important as well.
There's more to life than low power gear.
There's also the need to take into consideration the usefulness and efficiency of the gear for potential uses down the road. If it sits idle 99% of the time then it doesn't really make more than a couple of $'s year difference.
My setup is 24/7 w/ a 7900X CPU and an A380 GPU inside w/ a U.3 NVME that can hit 25W under load on its own. 99% of the time it's just pushing packets as a router / FW and under 1% utilization. When I need to transcode a video file though it ramps up a bit but, it's duration is a lot less than if it were CPU only or a lower end setup. Once dialed in the GPU handles it with only 2 cores and it's a brief process that pushes 1000 fps while converting a media file.
Having the potential to ramp up and complete a task more efficiently vs cheaping out to hit 10W total consumption most of the time. It all depends on the forecasted use.
The work around is to use a cheap $100 A380 to accomplish whatever you need QSV for. I use one to convert TS files to MKV to reduce space and not engage the CPU on playback.
Sure why not, but one more fan and more power consumption. (Guessing you will have more power consumption, have not actually checked)
It's just sitting there and it's the lowest max power of 35W GPU that can do QSV. Fan doesn't spin if it's not being used and power is minimal in stand by when not processing files. There are other 380's though that are 75W cards and require additional PSU cables to be attached.
The iGPU on the CPU uses more power than the card does most of the time.
For a 24/7 machine I would not look at TDP (which is only relevant when running at high load) but at IDLE consumption since this is what you have to pay for.
If you consider CPUs such as 14700K or 14900K or 7950X than Intel has an advantage here. As we all know, these Intel CPUs need quite a lot of power under load but the platform (!) is very efficient at idle.
Although I see AMDs Ryzen 7 as the better CPU, by own build (some month ago) is Intel based because comparable boards need appr. 10 W less in idle (around 20 vs. 30+ for the whole system).
Raptor Lake gives you the option to very easily change power limits at runtime in Linux with no additional tools (just accessing RAPL using /sys-filesystem). This way, I can operate my 14900K in a very efficient manner (and restrict it to PL1=PL2=125 W for more demanding tasks).
If you consider CPUs such as 14700K or 14900K or 7950X than Intel has an advantage here. As we all know, these Intel CPUs need quite a lot of power under load but the platform (!) is very efficient at idle.
Although I see AMDs Ryzen 7 as the better CPU, by own build (some month ago) is Intel based because comparable boards need appr. 10 W less in idle (around 20 vs. 30+ for the whole system).
Raptor Lake gives you the option to very easily change power limits at runtime in Linux with no additional tools (just accessing RAPL using /sys-filesystem). This way, I can operate my 14900K in a very efficient manner (and restrict it to PL1=PL2=125 W for more demanding tasks).
Are you sure about that? My googlefu gives me around 15W idle power usage on A380. I don't measure any additional idle watt on igpu (headless mode)
I'm headless as well and it still adds to the monthly costs if you want to get down to the penny.
Closest thing for measurement we could use is powertop to compare things. I don't have a physical power meter as costs are under $0.10/kwh here.
Code:
sudo powerstat -R
Time User Nice Sys Idle IO Run Ctxt/s IRQ/s Fork Exec Exit Watts
-------- ----- ----- ----- ----- ----- ---- ------ ------ ---- ---- ---- ------
Average 0.1 0.0 0.1 99.8 0.1 1.0 1023.9 905.0 0.3 0.2 0.3 35.52
GeoMean 0.1 0.0 0.0 99.8 0.0 1.0 1015.1 893.7 0.0 0.0 0.0 35.48
StdDev 0.1 0.0 0.0 0.2 0.1 0.2 143.3 149.5 0.9 0.8 1.0 1.67
-------- ----- ----- ----- ----- ----- ---- ------ ------ ---- ---- ---- ------
Minimum 0.0 0.0 0.0 99.2 0.0 1.0 829.0 670.0 0.0 0.0 0.0 31.88
Maximum 0.3 0.0 0.2 100.0 0.4 2.0 1565.0 1443.0 4.0 4.0 5.0 39.76
-------- ----- ----- ----- ----- ----- ---- ------ ------ ---- ---- ---- ------
Summary:
CPU: 35.52 Watts on average with standard deviation 1.67
Note: power read from RAPL domains: pkg-0, core.
These readings do not cover all the hardware in this device.
Code:
-------- ----- ----- ----- ----- ----- ---- ------ ------ ---- ---- ---- ------ ---------
Average 2.4 0.0 2.2 95.3 0.1 2.3 150771.0 67514.4 11.1 5.5 10.9 58.18 N/A
GeoMean 0.0 0.0 0.0 95.1 0.0 1.8 12533.4 8510.3 0.0 0.0 0.0 53.18 N/A
StdDev 2.9 0.0 2.8 5.7 0.1 1.8 193495.9 84990.1 14.4 7.2 13.9 24.79 N/A
-------- ----- ----- ----- ----- ----- ---- ------ ------ ---- ---- ---- ------ ---------
Minimum 0.0 0.0 0.0 85.8 0.0 1.0 806.0 700.0 0.0 0.0 0.0 33.17 N/A
Maximum 7.2 0.0 7.3 100.0 0.4 7.0 473503.0 207734.0 48.0 23.0 47.0 97.76 N/A
-------- ----- ----- ----- ----- ----- ---- ------ ------ ---- ---- ---- ------ ---------
Summary:
CPU: 58.18 Watts on average with standard deviation 24.79
Note: power read from RAPL domains: pkg-0, core.
These readings do not cover all the hardware in this device.
Code:
sensors
amdgpu-pci-5f00
Adapter: PCI adapter
vddgfx: 1.46 V
vddnb: 1.10 V
edge: +45.0°C
PPT: 58.23 W
(crit = +84.8°C)
i915-pci-0300
Adapter: PCI adapter
in0: 0.00 V
power1: N/A (max = 43.00 W)
energy1: 40.81 kJ| Sparkle ELF Arc A380 6 GB Video Card | 18W - 75W |
The numbers are all over the board when it comes to guesses across different sources w/o a physical power meter to measure the current passing from the PCIE slot to the card it's more of an observation of clues. Messing with different attributes that control the power schedule make differences too. Default handbrake w/o QSV properly enabled will engage the CPU in addition to the GPU and push the power needs up from 2 threads to 16-20 depending on the file. So, 8-10X as much power needed theoretically. The biggest advantage regardless is the time to process files is 1/8th that of CPU only.
I would consider $1100 is quite as lot of money for a 16 core system, even if the cores are pretty capable unless you need PCI-E 4.0. Why would you spend $1100 if $400 can meet your requirements.
Can a LGA 2011-3 or LGA 3647 based system meet your performance needs? It seems to me, you need lots of memory and PCI-E lanes but that's pretty much of it.
155W is not the power usage but TDP of the CPU. Overall system power usage will depend on the number of hard drives, memory etc. My guess is that you will end up somewhere between 100watt to 140watt at idle. That's pretty high power usage for a NAS.
Last edited:
Why 14700K for just quick sync? Isn't this overkill even if the idle power usage is minimal? Do you game or run business workloads?
Is this only for transcoding/NAS or is this also the set up for productivity/gaming?
For me it's kind of an all in one of collapsed devices into a single box that started back in 2015. I rolled a bunch of things into a single box for convenience and performance instead of continuing to buy consumer junk. I've rebuilt it several times to what it is currently as tech changes over the years. It has the ability to be churned for a premium when it comes time and all someone needs to do is add their GPU of choice and upgrades to other stuff if they want to. I tend to build with the plan of churning as new stuff comes out not for holding onto it for a decade or more like some do.
If you measure at the wall with a power meter I'm saying that you should be a static +15W as compared to without having the card plugged in. For me power consumption is a factor, the +10 for ipmi is questionable since i we very seldom use it.
Otherwise i agree with you and i guess it comes down to what idle power consumption you are comfortable with and how much idle time you have. I would probably be better off with an i3kf, but I do like the extra amount of muscles from time to time. A fair trade off imo
Not just for quick sync, more multi-purpose platform and somewhat future proof. I do most things but gaming. Most of the time it's idle. If idle consumption is 55w vs 90w it will make a difference looking at 5-10years lifetime.
Still ihave other equipment with more power draw(switches), but harder to replace those.
Going off of others who have epyc homelabs, the idle draw is not horrible
And that Epyc has a TDP of 225w, so it would seemingly draw more power than the version I bought.
oof, did you buy the Epyc already? The 7313 is a full 2GHz and one major generation (~15% IPC deficit) behind the 7950X in single core performance and ~1 GHz behind in multicore. It's not a good use of money unless you are looking for RDIMM and lots of PCI-e lanes - the target use case is high capacity flash based storage where you need the I/O.
I need to be able to easily support 16sata drives and 4 nvme, qling with multiple gpus. i also need to be able to hit at least 512gb ram.
the 7950X falls short on a lot of those
Turbostat power readings of 14900K in idle mode with two active monitors (X11 + WindowMaker) driven by iGPU:
PkgWatt CorWatt GFXWatt
6.15 2.01 0.01
-> iGPU seems to be powergated nearly all the time. Running applications such as glxgears increases this up to 6.25 W so this reading seems reasonable.
No way an external GPU comes near these values.
My experiments with an A380 showed a rather high idle consumption running linux (appr. 20-25 W more than iGPU measured at wall for the whole system).
PkgWatt CorWatt GFXWatt
6.15 2.01 0.01
-> iGPU seems to be powergated nearly all the time. Running applications such as glxgears increases this up to 6.25 W so this reading seems reasonable.
No way an external GPU comes near these values.
My experiments with an A380 showed a rather high idle consumption running linux (appr. 20-25 W more than iGPU measured at wall for the whole system).
heh, that's a big change from "the ram usage will be pretty low" a few posts up. if your requirements are 512GB RAM and 16 lanes of nvme then yes, you are forced onto an enterprise platform. insofar that you accept performance is bad anyway compared to the desktop platforms, Skylake-SP is not a bad choice - peak performance will be about 20% worse than Milan, but the monolithic dies help idle power and you get x16 GPU, x16 nvme, and x8 HBA all at full speed