Page 1 of 1

How do the new NVIDIA 30 series RTX cards look for Neat Video?

Posted: Wed Sep 02, 2020 4:20 am
by albino1
I've been holding out on upgrading my old NVIDIA 980 Ti for quite a while hoping for a big improvement with AMD's RDNA2 for a reasonable price, but the NVIDIA announcement today looks awfully promising. I know there's no real benchmarks yet, and I hope they'll be coming the blog here like from the 20 series, but I thought I would ask how good the numbers put out today look in terms of Neat Video performance, especially since I don't even play video games so the only FPS I would upgrade for is with Neat Video ;)

According to NVIDIA themselves the RTX 3070 MSRP is going to be $499 and is a 2080 Ti equivalent, which is a pretty great deal, but the numbers aren't quite the same. Here's a couple of pages with relevant info in nice table formats:

https://www.anandtech.com/show/16057/nv ... 0-rtx-3090
https://www.reddit.com/r/nvidia/comment ... egathread/

The anandtech one is particularly helpful as it compares the 2080 Ti directly to the 3070, 3080 and 3090.

The 3070 for example has a higher boost clock, faster memory, higher single precision performance, and more CUDA cores, than the 2080 Ti, but it also has less RAM, a lower memory bus width, and lower memory bandwidth. Also the 3070 will be PCIe 4.0, though I'm not sure that matters, and it will have RTX I/O, which might be useful somehow?

I'm not sure which of those metrics really apply directly to Neat Video anymore since it was kind of rewritten for Neat Video 5, and a lot of those things are so much faster than they were a few years ago anyway where it might be approaching a point of diminishing returns on things like memory bandwidth.

It seems like maybe Single Precision GFLOPS/TFLOPS might be the most relevant to Neat Video using that previous NV Blog post as reference, with the 3070 having 20.4 TFLOPs vs 13.4 TFLOPs for the 2080 Ti (11.75 listed in the original NV Blog test). For reference the 3080 has 29.8 TFLOPs and the 3090 has 35.7 TFLOPs. If those scale even close to linearly from the 2080 Ti then it looks like a massive NV FPS increase, so hopefully we get NeatBench results from them as soon as they release.

I fully appreciate this is really premature to ask about these GPUs, but I think it's more of a general question about which of those factors are really the deciding pieces of information that tell us about potential improvements FPS when using Neat Video. Especially because we're going to want to compare the RDNA2 cards as soon as AMD releases the relevant info as well, as they're likely to be priced very competitively with NVIDIA for similar performance gains.

Thanks!

Edit: A lot of edits at this point.

Re: How do the new NVIDIA 30 series RTX cards look for Neat Video?

Posted: Wed Sep 09, 2020 10:36 pm
by NVTeam
Thank you for your message and comments. We have also been following those announcements and even got a chance to run some preliminary tests on the 3080. Please see this article with out first observations.

Thank you,
Vlad

Re: How do the new NVIDIA 30 series RTX cards look for Neat Video?

Posted: Thu Sep 10, 2020 4:51 am
by albino1
Wow, amazing that you got that! The performance increase looks significant to say the least, and I love that you think you can squeeze even more out of it once you really start tinkering.

Thanks so much for that whole write up, it makes me very excited to see your future tests, and might already make the 3080 a Day 1 purchase for me based on that alone.