Nvidia GeForce RTX 50-series Blackwell reviews

I mean, is that something y'all are interested in seeing more of?

I'll benchmark my 4090 and a 5090 n my i7-3930k if you want, after I get a 5090 :D
It would be interesting IMO. Don't want to waste your time if it's too much trouble but I bet many modern games will run fine on that CPU if you're good with 30-60fps. My brother is currently playing Cyberpunk on a 3770K. Perfectly playable at medium settings.
 
I bet many modern games will run fine on that CPU if you're good with 30-60fps
I was CPU limited on Broadwell-E back on 3080 in 4K.
You could probably play some games on it but there's not a lot of reasons to test it with a 5090 IMO.
From the results we have it looks like even 9800X3D is barely able to keep up even in native 4K.
 
I was CPU limited on Broadwell-E back on 3080 in 4K.
You could probably play some games on it but there's not a lot of reasons to test it with a 5090 IMO.
From the results we have it looks like even 9800X3D is barely able to keep up even in native 4K.
Oh yea the 6700XT is massively CPU limited by the 3770K in that system. It would purely be a 3930k review in 2025. Which is interesting but it sounds like a lot of work for something so esoteric.
 
From the results we have it looks like even 9800X3D is barely able to keep up even in native 4K.
Do you have actual source for this thought? All the reviews I've seen have gained what you'd expect from OC (both factory and otherwise), which suggest is not CPU limited on topend consumer CPUs like 9800X3D at 4K
 
Do you have actual source for this thought?
Yes. It's the results of benchmarks we have. As I've said.

Here's a fresh example:
This one is CPU limited in native 4K for me on my 4090+5900X. Not for the whole benchmark but for a solid part, like 50% or so.
They are testing it in 4K with DLSS-Q which in my case was CPU limited at like 80% of the run.
Note the +15% gains 5090 shows on 4090 on 9800X3D there.

I'm pretty sure that a vast majority of modern games will have similar issues if the benchmark run isn't tested and selected properly to be fully GPU limited.
You will still have "gains" because the run isn't 100% CPU limited but they will be low because it is CPU limited to some degree.
 
The string of unexplained behavior continues. In PassMark, the 4090 is 18% faster than the 5090! The 5080 can't exceed the 4070Ti!!

Didn't Nvidia say that any app using tensor cores would have to be recompiled to see performance improvements? Could the same be said for straight CUDA applications as well?
 
Last edited:
Back
Top