Discussion The 64 core 200W EPYC 7702 runs cooler than the 3950x by 15C+....interesting results.

[traderjay]

The two rigs have identical coolers, Dark Rock Pro 4 and after one week of testing, the EPYC settled at a nice 65 to 70C during 100% mining load while the Ryzen 3950X is blasting away at 80C ++. The ryzen 3950x is running stock, with PBO and voltage offset of -0.025V. The 3950x during full load draws close to 170W of power for comparison sake while the EYPC draws the full 200W.

I guess the much larger IHS and the 8 chiplets spread apart really help with heat dissipation and cooling performance.

 

[Topweasel]

That extra 30w is spread over 4x the chips, so a lot is at play. Now keep in mind much like when analyzing the 10900k and what Intel did to help it transfer heat quicker to the HSF. It still is a "hot CPU" where hot is the amount heat it generates. Just because he runs cooler under the same cooler doesn't make it a "cooler" cpu. It will still heat up the room 30w quicker than the 3950x.

 

[Markfw]

Also, the 3950x runs at about 4 ghz, while the 7702 runs at 2 ghz. My 7742's also run nice and cool and efficient.

 

[StefanR5R]

To add to the previous responses:

1.) As long as a CPU cooler's radiator is not severely undersized, the temperature readings from modern processors depend much more on their clock speed and Voltage than on cooler size relative to socket power.

In other words, as these temperature readings generally come from little spots, they are very much indicative for heat flux density, but not all that much indicative for heat flux.

2.) Not only has EPYC 7702 four times as many Core Complex Dies (CCDs) as Ryzen 3950X, it also has got a four times as large and functional I/O die (IOD). That is, one 7702 is pretty much exactly the same as four Ryzen 3950Xs.

7702's default package power tracking target (PPT target) is 200 W. 3950X's default PPT target is 142 W. (If you see more actual sustained power usage, then this processor is not being configured at defaults.)

Reconfigure the little 3950X to 1/4 of 7702's PPT, i.e. to 50 W, and see what happens to clocks, Voltages, and ultimately temperatures.

3.) Modern desktop processors are configured at completely ridiculous points of their Voltage-frequency relationship. And desktop mainboard vendors typically go out of their way to make it all even worse. Such shenanigans are very uncommon in the server space.

 

[thor23]

The Epyc is running much lower clocks/voltages so heat density is much less.

 

[DrMrLordX]

In other words, as these temperature readings generally come from little spots, they are very much indicative for heat flux density, but not all that much indicative for heat flux.

Bing bing bing! We have a winner. The closer you push Zen2 to its optimal voltage/clockspeed range, the less-problematic are the hotspots.

 

[alexbirdie0]

The two rigs have identical coolers, Dark Rock Pro 4 and after one week of testing, the EPYC settled at a nice 65 to 70C during 100% mining load while the Ryzen 3950X is blasting away at 80C ++. The ryzen 3950x is running stock, with PBO and voltage offset of -0.025V. The 3950x during full load draws close to 170W of power for comparison sake while the EYPC draws the full 200W.

I guess the much larger IHS and the 8 chiplets spread apart really help with heat dissipation and cooling performance.

Turn off PBO for 3950x and compare again. PBO on IS OVERCLOCKING and NOT STOCK.

 

[myocardia]

Turn off PBO for 3950x and compare again. PBO on IS OVERCLOCKING and NOT STOCK.

Not only that, but everyone so far that I have been able to find who has compared PBO-on vs PBO-off, has found that turning PBO off results in either higher performance, or equal performance. The differences are between different softwares. IIRC, the equivalent performance were with single or lightly threaded software, and the lowered with PBO-on was with higher threaded workloads. These CPUs have been designed to run at as high of a clock speed as they are able to run, as long as they are kept cool. Since PBO raises voltage, it also raises heat production, lowering the speed at which the CPU runs more often than not.

 

[DrMrLordX]

PBO only got me extra performance in purely ST benchmarks. And even then, not by much.

 

[KentState]

My 3970X runs hotter when a few cores turbo up over all them being loaded. I was wondering how they kept heat in check, but the first time I ran Prime95 I got my answer.

 

[traderjay]

I get higher hash rate in mining with PBO on vs OFF. I get close to all core clock of 4Ghz vs 3.85.