The New Chip#

AMD recently released their Ryzen 9 9950X3D2 Dual Edition. Just like that suffix, my purchase of this CPU was unncessary. Following in the footsteps of AMD, I did it anyway.

But what is this chip? Basically the high core count Ryzen chips with 3D vcache have only had that cache on one of the CCDs because AMD says it’s not worth it to put it on both. But they listened to the people that want it anyway, and released this $900 chungus of a chip.

You can find a plethora of reviews online for this already, by the people that do reviews for a living. They all basically have the same conclusions:

  • Yes, this is the fastest desktop out right now.
  • Only very specific use cases can take advantage of this extra cache.
  • Don’t buy it, you fucking idiots.

I read those conclusions, really internalized them. Then, because I can’t read, er because bigger numbers mean I’m having more fun, because I’m one of the elite few that can take advantage of the specific use cases like heavy compiling and virtualization, I decided to spend too much money.

Let’s do some benchmarks to try and measure what I get out of this update, and if it was worth it. Hypothesis: it probably isn’t…

BIOS Updates#

Had to update [my motherboard’s](https://www.gigabyte.com/Motherboard/B650I-AORUS-ULTRA BIOS to have support for the new CPU. I at first updated to the newest, shiniest version; F42a. But, the mobo Displayport port wasn’t working in the OS. It worked in the BIOS and the bootloader, but not in the OS. Rolled back to F41 because of that, and it works just fine now. ¯\_(ツ)_/¯

Had to do some other tweaking and disable Preboot DMA protections to get shit to get my VMs to work. There was some speculation online that this is for RIOTs root kit anti-cheat. I don’t play RIOT games, so I don’t give a shit.

Other Notes#

The 7950X3D was cooled using PTM7950 phase change thermal pad, good coincidence in naming. I bought a thermal pad from a local computer store, it was not PTM7950. No idea what it actually was other than just junk. Ended up using Thermal Grizzly Duronaut for the 9950X3D2 since that’s what Micro Center had.

Obviously this will make for at least a little bit of difference when it comes to the thermal performance between the two.

Benchmarks#

These benchmarks will be the Tctl temperature and average clock speed after 3 minutes of stress in s-tui for both vanilla and the TKG Kernel. I also tested compilation times for the Zen 4 & 5 versions of the TKG kernel, though while booted under the vanilla kernel. Lastly, I did a couple of runs of the Cyberpunk 2077 benchmark in a Windows gaming VM with the GPU passed through.

I will also avoid doing any updates or changing any other variables to keep it as apples-to-apples as possible.

7950X3D#

Stock Arch kernel - 6.19.11.arch1-1#

The idle temp in s-tui was ~42.6 C

  • s-tui running with the s-tui stress for 30min:

    • Tctl temp: ~78.0 C
    • Avg Freq: 4640-4658 MHz

  • TKG compiliation. Version 7.0.1 with the EEVDF scheduler

    • znver4:
      real  11m1.361s
user  295m33.769s
sys   28m45.858s
    • znver5:
      real  12m11.795s
user  330m41.457s
sys   30m35.542s

TKG kernel - 7.0.1#

An idle temp of ~42.2 C

  • s-tui running with the s-tui stress for 30min:
    • Tctl temp: ~77.5 C
    • Avg Freq: 4669-4680 MHz

In the gaming VM, the Cyberpunk benchmark had these average FPS values: (43.63, 44.14, 43.50)

9950x3D2#

Stock Arch kernel - 6.19.11.arch1-1#

  • s-tui running with the s-tui stress for 30min:

    • Tctl temp: ~86.1 C
    • Avg Freq: 4932-4942 MHz

  • TKG compiliation. Version 7.0.1 with the EEVDF scheduler

    • znver4:
      real  10m0.917s
user  276m26.829s
sys   19m52.707s
    • znver5:
      real  10m5.018s
user  277m44.686s
sys   19m57.543s

TKG kernel - 7.0.1#

  • s-tui running with the s-tui stress for 30min:
    • Tctl temp: ~ 84.8 C
    • Avg Freq: 4960-4967 MHz

This is where things get really interesting.

Cyberpunk benchmark without changing CCD/core pinning: (41.93, 41.66, 41.66)

But this is lower than before?! Let’s try the other one: (41.55, 41.34, 41.45)

Nope, maybe lower ambient temp: (42.06, 42.17, 41.66)

A little better. But for funsues, lets recompile the kernel and try again: (42.35, 41.92, 41.94)

Yeah, those are still lower than before.

Conclusion#

Well, less of a conclusion and more of just some obvservations.

This chip has a 40% higher TDP, and so it runs hotter:

  • 7950X3D TDP: 120W
  • 9950X3D2 TDP: 200W

I’m running a 280mm radiator for cooling, and AMD recommends a 360mm rad. So I’m already out of spec. 280mm is a lot closer to 360 than 240 is, but still a bit small. I don’t exactly have much room for upgrades in the NCase M2.

I also forgot to check the other CCD while I still had the 7950X3D installed, so I can’t do a full suite of tests. Kind of a bummer that my immediate virtualization use-case suffers from this “upgrade.” I also forgot to test the native framerate for the 7950X3D, but playing natively within Arch under the 9950X3D2 is still pretty good: (47.10, 47.03, 46.65)

At least compiling the kernel did have some decent benefits. My package update cycle will be so much faster now , which is a huge benefit since I update more than game now anyways. Thanks {employer}

I should spend more time to tune the system and see if I can make the upgrade worth it in some tangible way…