News Noctua SFF Cooler Roundup: Testing Small Coolers Against Intel’s Core i9-13900K

[hotaru.hino]

Yeah, not having ever run CineBench, myself, I wonder if it's fixed in the workload size or is it fixed in the amount of runtime?

The default test is a fixed workload size and however many can be completed in 10 minutes, completing anything partial when the time runs out.

You have to view energy usage in conjunction with performance. For a machine someone is using interactively, you can't afford to look only at efficiency.

This depends entirely on your requirements. If time is literally money or you need a task done in a time constraint, then yes, performance tends to matter a lot more than efficiency.

But if all I'm doing is say a Handbrake run or Blender render and I don't really care how long it takes, then I'd rather have more efficiency.

On the flipside with a time constraint however, ideally you should have something that has the shortest runtime completion so the CPU can have more downtime.

 

[bit_user]

I haven't had time to play with it much and try and undervolt, or do a foam mod,

I'm comparing against systems which also aren't undervolted or using a foam mod.

I'm typing this on a Sandybridge i7-2600K, sort of bumping along between 4% and 8% utilization, and it's about 45 degrees, using the boxed cooler and just a 90mm case exhaust fan. I don't know it's true idle, though. My linux box is a Sandybridge E5 Xeon and has a true idle below 40 degrees.

but my idle use is really typing/light browsing + Youtube video playing.

Okay, so you weren't really quoting its true idle temps?

 

[bit_user]

On the flipside with a time constraint however, ideally you should have something that has the shortest runtime completion so the CPU can have more downtime.

Since power scales nonlinearly with frequency, it tends not to be a winning strategy to juice a CPU's performance in a "race to idle", for the sake of efficiency. A long time ago, things were a bit different.

 

[Albert.Thomas]

The only thing I wish the article did was say how long each Cinebench run lasted, and the score doesn't make it obvious what that is. Power consumption needs a time component to get the metric that matters in all this: how much energy did it take to do the thing?

Like the 45W figures may look nice from an efficiency standpoint, but if it takes over three times as long to complete the task compared to max power, it's actually worse.

These scores were for 10 minute runs.

 

[cyrusfox]

Okay, so you weren't really quoting its true idle temps?

Guilty, I never leave it idle without sleeping, but the few where I attempted to idle it was sub 50°C, so likely 45-50°C for true idle. I don't have any exhaust fans, only fans exist on the PSU the CPU and the GPU which is sandwiched on the other side, it can get rather toasty (Optane 905P drive, GTX1080, and 2x 4TB SATA SSD hanging around).

 

[bit_user]

Guilty, I never leave it idle without sleeping, but the few where I attempted to idle it was sub 50°C, so likely 45-50°C for true idle. I don't have any exhaust fans, only fans exist on the PSU the CPU and the GPU which is sandwiched on the other side, it can get rather toasty (Optane 905P drive, GTX1080, and 2x 4TB SATA SSD hanging around).

Thanks. That explains a lot.

Update: I just killed Firefox and let the CPU sit at 1% for about 10 minutes. The package temperature settled down to 40 C and package power was about 9 W, according to the ThrottleStop utility.

 

[cyrusfox]

Thanks. That explains a lot.

Update: I just killed Firefox and let the CPU sit at 1% for about 10 minutes. The package temperature settled down to 40 C and package power was about 9 W, according to the ThrottleStop utility.

I left mine (13900 on H670 itx/ax mobo) idle with all applications closed for 15 minutes and came back to package power of 1.857W and idle temp of CPU package of 43°C(CPU core was 40°C). So some advancements have been made since Sandy Bridge. I downloaded throttle stop to find these fields, but as TS does not give an average I found the corresponding identical fields in HWiNFO and pulled from that.

 

[bit_user]

I left mine (13900 on H670 itx/ax mobo) idle with all applications closed for 15 minutes and came back to package power of 1.857W and idle temp of CPU package of 43°C(CPU core was 40°C).

That's lower power than I'd expect and higher temperature than I'd expect for that power dissipation.

I have a i9-12900 sitting in an air conditioned lab, at work. It's super idle, with no GUI even running. turbostat claims the package power is just 0.6 W and sensors tells me the package temperature is 28 C. The chassis is a compact desktop form factor Dell Precision series (which is also why I couldn't get a K-tier CPU). Think of a small mini-ITX case that's only tall enough for half-height PCIe cards. They appear to have a blower-style cooler that vents directly out the back of the case and it has no dGPU.