General Category > Hardware

Threadripper 5975wx VS 3990x

<< < (2/3) > >>

Nejc Kilar:

--- Quote from: Robin Walker on 2023-01-09, 17:12:54 ---Thanks guys, some interesting stuff above.

I’m really just looking for a direct comparison between the 5975wx and the 3990x when it comes to long rendering times in corona.

If i take the middle of the pack for corona benchmark results and compare them, I come to conclusion that the 3990x is about 1.5 times faster than the 5975wx

BUT if i look at these cinebench multicore results here it tells me that the 3990x is only about 1.2 times faster https://www.cpu-monkey.com/en/cpu_benchmark-cinebench_r23_multi_core-16

For both I just divided 3990x score by 5975wx score. Cinibench does not give results in seconds and it's obviously not corona render engine, but surely the same logic applies as it's a multithreaded render benchmark? maybe I'm wrong here?

Bottom line - If the 3990x is only 1.2 x faster then i'll probably turn a blind eye since its quite a bit of hassle to source all the parts for a 3990x build, not to mention  the volatility of the price.

But if it’s 1.5 faster then i'll go out of my way to do it, since i'll be getting more than one build for the office. Oh and 5995wx are completely out of the question, £7000+ for a build when they appear to be only marginally better than a 3990x…according to the benchmark results.

Just waiting for that magic person who owns both to give me a comparison benchmark :)

--- End quote ---

By and large each renderer has its own approach and specifics which does mean it might hit the CPU differently. As a personal opinion, I would encourage you to focus on the benchmarking tools of the software that you are using - that'll probably the best indicator for the performance you'll get although that said there's scenes that might have a different hit on the performance and so even software specific benchmarks are supposed to be taken as a ~value.

As for the 3990x vs 5975wx debate, I'd probably also look into a 3990x vs 3970x comparison and to get to the 5975wx performance levels I'd just subtract another 10-15% of the 3970x times. It isn't scientific no but should be indicative enough imho.
Also, my impression is that the performance class didn't really change between Zen 2 and Zen 3. Sure, there is that 10-15% boost + whatever clocks get you but a 5975wx still occupies a similar spot in that stack that the 3970x did.

danio1011:
We have both 3990x's and a 3970x.  I used to think the 3970x was snappier due to its slightly higher clock speed for single core tasks, but I think I was just fooling myself, they're really about the same for workflow.  The 3990x, though, is way faster when set loose on our farm.  I don't have exact numbers but when watching them all work away in our render manager the 3970x always seems about 30-35% slower.

Robwan:
Thanks guys,

Yea we actually also have a 3970 too :) So I'm aware of its speed compared to the 3990x.  But adding a 10-15% speed increase to give an idea about the 5975wx seems like a good idea. Ill run another test this week and see how it aligns with my other research.

For me it's all the about render times and having better cpu takes hours off. I've used max and corona on a variety of devices - laptops, i9 pc, older dual xeons and I've never noticed any difference with anything other than raw render time. If I was looking for it, I'm sure I would see that some cpus fire the render up a bit faster but I'm more interested in cutting long render times.

Only if we could put two 7950x on the same motherboard 🤯 that would be awesome.

TomG:
Putting two on one board - that would make them Epycs then? :)

Robwan:
Well, that's food for thought 🎂. If I could get two epyc chips totalling at less than the price of a 3990x, but could match the render time of the 3990x.

How would I work out how well they perform though? There's so many models to choose it seems. I guess if I knew how well one performed, I just slice the the render time in half for two right?

Navigation

[0] Message Index

[#] Next page

[*] Previous page

Go to full version