Silicon lottery - how much difference is acceptable?

I recently got my hands on TR5965wx workstation with ASRock WRX80 Creator motherboard and RTX3080 gpu (Gigabyte Turbo).

While i understand why my GPU scores may fall below average quite a bit since it’s Gigabyte Turbo blower design and thermals are far from perfect (but I really needed true dual slot GPU so it doesn’t block my pci-e slots that I need for other cards), it’s not all that clear to me why CPU scores fall behind the pack quite a bit as well.

I’m on dual fan config of Noctua NH-U14S TR4-SP3 so according to internetz and benchmarks for 3990x i should be getting around 70 deg under load, maybe less because it’s 64 core and I’m on 24 cores. But I don’t. It stabilizes at 83 deg in open air setup so it sounds quite high. I perforned Blender benchmark and got 680 points while other scores reach 710. That’s quite substantial difference.

Does it fall into silicon lottery error margin or should I look for issues with my setup?

Is that setup physically similar to yours? [Open Air, etc]
Was the thermal grease reviewed [in case of uneven pressure/application]?
You do have apple vs. orange scene [lower core / high freq vs high core ct. / lower freq]

Have you reviewed, how those benchmarks were performed? [Specs, PBO, etc.]

What are your ambient temperatures like? Most of the tests are climate controlled at around 20-25 deg C.

There’s zero silicon lottery involved into running a CPU at stock settings. If there was nobody could segment CPUs by performance. I don’t know how much 30 points in Blender benchmark really matter. But you should do more runs to make sure it’s not a fluke. Those small differences, sometimes, may be also due to the software configuration, if you’re running more background tasks compared to other users.

You might be running at an higher room temperature compared to other people testing. Don’t get fooled by the Youtube reviews because even GN runs tests at 20°C environment temperature, more or less, which is unachievable for most of the world, especially during summer/spring (unless you live in the US and run the AC constantly like a madman to keep that temperature). You should also check the thermal paste application.

2 Likes

I’m at around 26 deg in room. Blender benchmarks scale linearly since it’s just based on render times so 680 vs 710 points account for around 4% render time difference.

I don’t know what PC used guy who got second Blender benchmark (unfortunately there were only two benchmarks for 5965wx - it seems to be the same guy on two Blender versions so I ran benchmark on the same two versions and got nearly identical % delta thus I believe it’s the same dude) but considering it’s TR PRO I doubt he was on LN or anything ridiculously crazy xD I mean it’s workstation CPU, he most likely also had workstation-ish build. Worst case he was on WC and gaming RAM while I’m on standard 3200 ECC RDIMM but still (x) doubt. Also idk if it’d even make difference in Blender bench.

It’s quite hard to find any reliable, real-world benchmarks for 5965wx. It’s not like every second kid around the block uses that cpu… And people who do post benchmarks often make crazy OC builds not just standard workstation builds that’s why I kinda focused on Blender benchmark since it sounds like something people who use workstation for work would run…

It’s stock settings but CPU boosts to around 4.25 ghz (while base clock is advertised to be 3.8 for this cpu and single core boost is stated to be around 4.5 ghz). So I believe his CPU could maybe just boost higher due to better thermals?..

1 Like

okay I just took a look here:

And I think it might… just might be related to fact that I’m running dual channel ram config because I didn’t know if sticks I chose will work so ordered only two for test and remaining 6 are still in delivery…

1 Like

That’s it, you got it. If you’re behind just 4% while using dual channel memory you’re gonna pass the 710 points with 6 channel memory for sure!

3 Likes

But you are also correct about thermals - zen boost higher the colder it is, and yes 5-7c ambient difference can be a visible difference in a heavy workload.

2 Likes