• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Should I upgrade my TR3960X to a TR3970X for an extra 8 cores/16 threads

Associate
Joined
22 Oct 2006
Posts
1,081
I would be interested to hear from any other felloe TR users if they have gone form a 3960 to a 3970 and if they found any major performance differences.

Most of the heavy lifting i do is in Premier Pro / Handbrake from a hobby perspective. Mainly work in 4K

I know the TR39XX series is getting a bit long in the tooth but it would be a major cost to upgrade to the later TR series and I don't think I could justify the cost.
 
I would be interested to hear from any other felloe TR users if they have gone form a 3960 to a 3970 and if they found any major performance differences.

Most of the heavy lifting i do is in Premier Pro / Handbrake from a hobby perspective. Mainly work in 4K

I know the TR39XX series is getting a bit long in the tooth but it would be a major cost to upgrade to the later TR series and I don't think I could justify the cost.

You'd like be better off selling the TR system and getting a 9950X unless you need the PCIE lanes.
 
From the few benchmarks and reviews I've seen on the net, even though it has 33% more cores/threads, it doesn't seem to work out that way in reality. Is the all-core boost clock lower?

The average increase is around 10-15%, with some apps being very small (like 3-5%) and more rarely a little larger (in the region of 20%).

In the review on THG, the increase in Premier Pro overall was just 5%, though export was over 8%.

For handbrake, PCMag have a review and it was about 6% faster for their test.

Puget have some numbers here:
 
Last edited:
From the few benchmarks and reviews I've seen on the net, even though it has 33% more cores/threads, it doesn't seem to work out that way in reality. Is the all-core boost clock lower?

The average increase is around 10-15%, with some apps being very small (like 3-5%) and more rarely a little larger (in the region of 20%).

In the review on THG, the increase in Premier Pro overall was just 5%, though export was over 8%.

For handbrake, PCMag have a review and it was about 6% faster for their test.

Puget have some numbers here:

Thanks Tetras - Excellent link and benchmarking data

Think I'll stick with the 3960X for the time being
 
I would be interested to hear from any other felloe TR users if they have gone form a 3960 to a 3970 and if they found any major performance differences.

Most of the heavy lifting i do is in Premier Pro / Handbrake from a hobby perspective. Mainly work in 4K

I know the TR39XX series is getting a bit long in the tooth but it would be a major cost to upgrade to the later TR series and I don't think I could justify the cost.
Have you tried using GPU codecs with handbrake. I’ve mucked about with the AMD ones and had some good results. Conversion time was much quicker.
 
Have you tried using GPU codecs with handbrake. I’ve mucked about with the AMD ones and had some good results. Conversion time was much quicker.

I've messed about with the GPU codecs but have never really been happy with the quality of output compared to the CPU codecs.
 
Does anyone know why the difference exists between GPU & CPU encoding. Why is the CUDA encoding quality sub par compared to CPU encoding.
 
I can't go into specific technical reasons between the reason why the GPU encoding can appear inferior to CPU encoding for say h264 or h265 encoding because I don't know, not my field.

But I do know from my own testing, the difference is in efficiency. The software encoders like x264 and x265 are more efficient than GPU encoders NVEnc or Intel Quicksync, they need a lower bitrate for an equivalent quality in video. If you compare say <20Mbit 4K video h264 if you look closely you may spot the difference between software and GPU encoding. At higher birates of say ~30 Mbit you would be hard pressed to see a difference. Even though for archival purposes and for video editing where quality is paramount I would not use NVEnc or Quick sync. The advantage with GPU encoding is it's a lot faster, and probably more power efficient. Issues really arise when people use GPU encoding at low bitrates like 16MBit/s or lower for 4K h264 and the quality looks bad.

As for the difference between 3960X and 3970X, I wouldn't bother. 8 extra cores of Zen 2 isn't going to make a big difference. Working in 4K, encoders like x264 or built in ones with DaVinci resolve or premier won't scale very well to 24 cores let alone 32. Usually have to run parallel tasks to get CPU utilisation up.
You are better off considering a mainstream platform like the Ryzen 9950X or 7950X. If you can't get enough of Threadripper, the PCIE lanes, more cores and you're feeling thrifty Threadripper 7000 is legit. Expensive as hell but for stability and performance you cannot beat it. I hope AMD gives us an excuse to spend more money on Threadripper Zen 5 soon.

I can't recommend a Intel 14th Gen 14900K. I had one for just encoding video and ffmpeg.exe threw exceptions after 2 weeks of usage (very performant has to be said). The issues these CPUs have are real. QuickSync encoding is useful if you don't mind compromising on quality but doesn't sound like it in your case. Where Intel genuinely has an advantage is in the hardware video decoding space. The iGPU can decode h265 10bit 4:2:2 video easily, my threadripper has to do this in software. Nvidia cannot support decoding this video either.
 
Last edited:
Back
Top Bottom