I try to use 4080pc and 3060ti pc to convert a 4K video to 60fps,
but they use the same time in tvai
(1.9fps/sec)
can someone tell me whats going on?
I use the tvai 3 and h265 mkv output
Which CPU do you have, TVAI is CPU limited.
60TI PC using 12400
4080 PC using 13700kf
eh ok, whait for the 3.1.0 release.
Any indication when that will be publicaly released? What kind of improvements are you seeing? Thanks!!!
It is released now. And the improvements typically range from 1x performance (no improvement) to 2x performance compared to the previous stable release.
Is there a performance difference in version 3.1.0?
IDK,i am working now, maybe try when I go back home
Oh my fking god.
I test already，
The speed is more than twice as fast as before
But I think the problem is not fixed
because my 3060ti have the same speed with 4080 again
My theory is that your cpus are not able to feed the gpus fast enough.
I’ve a 24 core 3960X and its not able to saturate my W6800 (rx 6800 XT) because my intel nvidia system does outperform it by 38%. (7820X + Quadro RTX 5000).
Or its because the AMD 6000 gen has no Martix (Tensor) hardware. (The 7000 gen has.)
Are you using the 13700 with DDR4?
Did you try to activate GPU Scheduling in Windows?
my full spec
I7 - 13700kf
Z790 tomahawk DDR4
D4 16GBĂ—2 3200 C16
tuf 4080
but I never checking GPU Scheduling, I will try it later
Checked.
I have activated that already. I don’t think TOPAZ has used up the performance of the 4080 has much to do with GPU scheduling
Maybe maybe not.
What did help with my config was setting the AI Processor to Auto instead to the GPU.
I have done what you suggested, but the usage of 4080 is still only half, and the temperature is not high
1 Like
It would now be interresting if the performance will rise with a DDR 5 kit.
because a single channel ddr5 has 39GB/s, ddr4 has 26GB/s.
There is also a memory speed bottleneck.
I did go from 1333 Mhz to 1600 Mhz an the performance did grow from 2.5 fps to 2.8 fps.
1080p → 4K
With my Threadripper 3960X and a W6800.
So, conclusion, you need fast RAM and CPU to get the performance up.
I want someone to do that very same test on an Intel CPU. I’ve heard that AMD CPUs process faster with better RAM frequency.
Intel I9-12900KF, overclocked to 5.1 GHz with all cores active, AIO water cooled
EVGA RTX-3080Ti Hybrid AIO water cooled
32 GB DDR5-5400 in 2 16GB sticks
Upscales reading from and writing to Micron ATA SSD
Windows 11 22H2
Topaz VEAI 3.1.0
Upscale from 480p DVD rip (~700MB files) to 960p using Proteus for detail recovery, sharpening and noise reduction, and running 4 simultaneous upscales, manages about 12 frames per second each. The four upscales finish in about 90 minutes.
I then take the 960p files and upscale those to 2160p using Artemis HQ. Again, running 4 upscales at once, runs at 7.5 frames/second each, and the total duration is ~4 hours.
FYI, both my CPU and GPU are clocking in at about 60°C, I have consumed 20GB of system memory and 10GB of VRAM.
All GPUs have some form of “Matrix” or Tensor core - even most modern CPUs have them. The ones in GPUs are generally much more numerous and powerful than CPUs. The newer and more powerful the GPU, the more numerous or more capable are the tensor cores (or both).
With regard to a RTX-3060Ti vs a RTX-4080, I wouldn’t expect much difference except in the number of simultaneous FFMPEG threads that can be run (due to more VRAM in the 4080). While the 40XX series has more Tensor cores than the 30XX series, they aren’t very different in capabilities. Once you have all the AI code running you may not max out all the 3060Ti cores, so having more on the 4080 may not make a difference.
1 Like
