Techtechpotato Youtuber dude's 3DPM beta benchmark

[igor_kavinski]

This is a benchmark doing some commonly used (in the scientific community) algorithmic calculations. The author (PhD so he knows what he's doing) shared beta code on Anandtech which I was able to compile. It seems to scale very well with the AVX-512 capabilities of CPUs despite it not being hand tuned super optimized code.

Binaries can be downloaded from here: https://www.overclock.net/posts/29526304/

It's a quick 2 minute multicore benchmark that can aid in validating CPU/RAM tuning efforts.

Now for some scores!

First up is my sad score (9950X3D 85C temp limit 6000C28 non-PBO FCLK 2167):



i9-14900KF@P8x57- 2x59/E45/R48 RAM 32GB DDR5 7200 CL34 @ 8200 CL36-48-48-2N



Tuned 9800X3D:



domdtxdissar's 9950X3D2 (he holds most HWBOT records with his custom cooled MORA setup):



Anandtech cases and cooling Supermod's watercooled stock Threadripper:

 

[igor_kavinski]

Core i3-1115G4 LPDDR4-3200

Hyperthreading OFF



HT ON

 

[igor_kavinski]

Core i3-1125G4 DDR4-3200



Core i3-1215U (not sure whether it's DDR4 or DDR5)



Yup. Intel shot itself in the foot by disabling AVX-512 in Alder Lake.

 

[igor_kavinski]

Alder Lake memory technology comparison using the same CPU:

 

[igor_kavinski]

Yes. AMD FX-8320 is actually not that bad! But the power consumption is.

 

[igor_kavinski]

Even Ice Lake with single channel RAM delivers performance close to Core i3-1215U. Intel, will you ever learn????

 

[igor_kavinski]

Epyc Rome ES (max 64 threads only due to Windows limitation)

SMT OFF



SMT ON



SMT hurts performance.

 

[igor_kavinski]

Epyc Milan ES SMT off



Both Epycs have a lot of knobs in the BIOS that may lead to better performance. NPS1 and SRAT disabled setting causes Milan to lag behind Rome.