The process does not start when multiple mp4 files are selected and the export button is pressed.
5.1.2
Mac
1 Like
The discussion about āmulti-GPUsā is way far off, you have always been able to use multiple and no, you donāt need two instances. I used to run a pair of watercooled 1080TIs, and older versions of TVAI would burn those things all day. Running 2 processes or 1, both GPUs would roughly double the performance of 1. Then I sprung a leak and had to upgrade my cards I just ran a single 4070 which was effectively about the same as 2 1080tiās using much less power, but the card is only like <1/2 utilized. I picked up a 2nd 4070, and in the case of 1 or 2 processes it is just a bit less than double the performance of a single 4070. Either way it wouldnāt knock anyoneās socks off cuz TVAI only ever used 1/2 of the 4070 at best, cards barely warm up.
So the problem isnāt that TVAI doesnāt do multi-GPU right, it does scale at least with 2. The problem is it only ever used a certain section of the cards. The only thing I really got with 4070s over 1080tiās was power efficiency which is huge donāt get me wrong, my power bill looks alot better now. Iād still like to see TVAI hotbox my GPUs though as even if I maxed them both out Iād be about 1/2 of what a set of 1080tiās ran power wise.
I just updated to the new version because the release notes said there was all this improved performance on multi-GPUs. I havenāt tested it yet but it sounds like it was just hype for the āproā version, I donāt expect it to be any better than the previous version but we will see.
Edit: Just started testing and realized they definitely messed up the UI implementation on multi-GPU. 4x upscale on a DVD, āAll GPUā, Max Process=1, running 1 video, results in 1 card @45% usage and the other one doing absolutely nothing. āAll GPUā, Max Process=2, running 1 video, results in both cards at about 36% and roughly 40% faster on the single videoā¦how does that make sense? āAll GPUā, Max Process=2, running 2 videos, results in both cards still at about 36%ā¦this is exacly how it was before!
I thought this update was supposed to improve as it was but itās exactly the same as before. Obviously they must have meant the improvement was only for the upcoming āproā version but I canāt understand how anyone would want to pay for features of pro if nobody can prove it actually works. No idea how everybody elseās system runs but someday Iād like to use more than 1/3 of my GPU here, been promising better performance for as long as I can remember.
Would you be more precise with settings to these ādoubleā?
My purpose is to encode one video but fast
At the moment done my tests done with GPU-Z shows a load balancing between and not at all paralelized workload
I have 15 of Aorus Xtrem
I would be happy to make them run as you suggest
In other what was the last version that perform as you firstly described
Blockquote
Would you be more precise with settings to these ādoubleā?
What are you referring to? As I posted (it was an edit)
āAll GPUā, Max Process=1, running 1 video (720x480 interlaced DVD, 4x iris)
about 6.5fps, only uses about 40% of GPU-1, nothing on GPU-2
āAll GPUā, Max Process=2, running 1 video (720x480 interlaced DVD, 4x iris)
about 11fps, only uses about 36% of GPU-1 and GPU-2 equally
The above is clearly some bug with the UI logic, max process should not matter if you are only running a single process, but it wonāt offload to the 2nd GPU unless max process=2
āAll GPUā, Max Process=2, running 2 videos (720x480 interlaced DVD, 4x iris, both same)
about 5.8fps, only uses about 36% of GPU-1 and GPU-2
I understand all of that, and make same obesrevation so my question is simple : what was the latest release that gave you the most parallelized worload treatment ?
I am not trying to encode 480p, 576p nor 720p, I want to scale up 1080p to 2160p with top quality using brute force GPU power.
v5.1.3.1b give godlike result in term of quality but at a so huge price : job is splited all around the GPUs so they run 100% workload / GPU number, whatever the number you got.
I got a 4090 runing @ 6% with 0% PCIE bandwidth used, GPU almost enter sleeping P state ā¦
Lol the new one does exactly that, just as the previous one did. I explained exactly that 2 you.
2 cards gives roughly double the performance as one, minus the odd UI settings bugā¦that is a āparallelized workloadā. I canāt say what happens above 2 GPUs, I donāt even know what CPU can serve that many GPUs.
I think what you are not realizing is that multiple GPUs require alot more CPU, at some point it will seem like the GPUs just keep dividing performance because they are all limited to a portion of the CPU.
If you wanna run a bunch of GPUs youāre gonna need the biggest threadripper or Epyc you can afford.
Man I do code paralized things. If job is done correctly the CPU must also sleep, the only thing he has to do is spread jobs around GPU.
So, again what release gave you best result for you, isnāt clear ?
THERE IS NO DIFFERENCE BETWEEN VERSIONS
THERE IS NO DIFFERENCE BETWEEN VERSIONS
THERE IS NO DIFFERENCE BETWEEN VERSIONS
2 Likes
I agree. I see better results using free super resolution models made for images than the output from even Proteus v4 (or Photo AI and/or Gigapixel for that matter). The only downside is the lack of temporal stability. But even Topaz isnāt really giving that, as Iāve tested all the models and combinations and there are issues theyāve never addressed.
1 Like
You can calm down, I am not resposible of unefficiency of the āmultiā much advertised but not effective.
To me Videoproc 7 with ability to tune GOP with huge values may help you.
Drawbacks AI models are not yet to the level of TVAI, even with impressive results in lats release.
TVAI updates are bugfixes, then when it starts to get decent they roll out a new UIs that will create more new bugs. A TVAI license is basically a season pass to a version that eventually will work OK until they break it again. Itās all a paid beta test until they can do exactly what everyone new they were gonna do and release a more expensive āproā version. The improvements are usually just a bunch of stuff nobody cares about.
Weāre at version 5 now, itās realistically no better than version 3 which we had to suffer a ton of headaches to get to. Keep hoping GPU support will get better or the performance will improveā¦nah, theyāve been saving that for pro.
8 Likes
Honnestly ā¦ no pro. will buy such product with such instalability or erratic problems.
To me, this is a buzz to catch attention and funds.
The website is awesome, the app. is a dice roll, all is said
Lol why the hell did I press the update button?
Previews disappearing? Are you kidding me? I search the forum and this has been happening on V5 since the beginning. Why the hell do I have to check a forum to see how bad the topaz guys broke the new version before I apply the update I PAID for?
Here this is still happening in the next beta releaseā¦you have got to be kidding me. Whoever took over Topaz AI after V2 is the most incompetent group of software people to have ever touched a keyboard. How many years later and you guys still break everything you touch? Iāve been pretty patient with this whole thing, itās mostly tolerable when a version is finally working half decent and I ignore the update button but man, I donāt think Iāve ever seen a button Iāve come to loathe so much as the TVAI update button. Dumb ass me, I thought when it said āimproved GPU performanceā that it might actually be the caseā¦HA, ya got me again. 15 minutes guys, thatās how long it took me to realize you broke it again. Thanks for wasting my time guysā¦of I go to roll backā¦again.
A single 4090 is able to saturate a 16 core 7950X.
Topaz Video AI 5.1.2.0 and interface bug still unresolved. Hereās how it tends to open at times.
Itās really very frustrating
1 Like
i love HDR 10 Template from Topaz AI will be absolutely great therefore Co Pilot is right even though it a forward looking statement at this moment in time.
I do also, but I would very please to learn how dev transform SDR to HDR by including meta data in the output file.
Thereās indeed some metadata, but we, community would be very pleased to listen how they do process this in order to know if we use it or not, without having to sample hours or days during the output result.
1 Like
A single 4090 saturate a 32 core / 64 threads CPU.
So what is the limit 128 ? 256 ? Or is it simply badly coded ā¦
Thereās absolutely no bandwidth on PCIE, so CPU is doing much more than what it should.
If CPU is used as a memory access, then VRAM should be used much more that it is the case.
And thereās other way to DMA if RAM is the true need.
Hello I had to roll back to version 504. On the last three updates the Nyx filter Enhancement there are no changes regardless of what I do with the settings. My rig, Ryzen 7 3800X, RTX 3060Ti, 32Gb Memory, 750 Watts PSU. Nvme storage OS and backups.
