Just bought Video Enhance 2.0, and would like to install it on a dedicated Win 10 PC. Can anyone recommend a pre-built, or custom-built PC that meets all the recommend(or more)system requirements(CPU, GPU, RAM, VRAM). I’ve used nVidia cards for years, and would probably stick with them. Thanks!
Welcome.
In all honesty, A DIY custom build is out off the question unless you want to pay scalpers 200-300% above retail for the CPU and GPU.
With that, given the GPU shortages your best bet is a custom order from the gaming pc maker like Origin, Digital Storm, iBuyPower. Getting a Ryzen 5900x/5950x coupled with a RTX3080, 6800xt or 6900xt is your best bet for best performance and you can very likely order one today through these builders. These system builders have these hard to get cards in stock typically at MSRP or barely above as they are a priority, the catch is you have to order the pc through them.
In terms of RAM I would buy minimal ram as they over charge on that. I’d order from Amazon and Newegg. AMDs CPU’s like 4 sticks vs 2 sticks of ram (performance). It is cheaper to buy 2 2x16gb memory kits then 1 4x16gb memory kit. Dont ask me why just look it up yourself. Try to get ram that is 3600mhz or faster at CL16. 3600mhz and CL16 is the price to performance sweet spot . Hope that helps.
1 Like
I am going to start my own thread in a second, but… I have exactly this setup: Ryzen 5950x, RTX 3080, 32Gb RAM @3600.
Gaia High Quality v5 is giving me 6.34sec -per frame-. I’m upscaling my 5.7K 360 video to 8K for good use in VR. (this means 36 hours for a 12min video… )
Is something wrong or is this as good as I should expect…?
1 Like
Nothing is wrong here I afraid. You’re trying to upscale a really high resolution footage to a super higher resolution one. So the result you get is correct. If you want it to be faster, you need to lower the original footage to 1080p or 720p, then upscale to 8K then. Or 36 hours are what you get for the output.
How does that make any sense though… if I take -out- details first, then there are fewer details to put in, so the model would have to do --more-- work… presumably. At least, more guesswork.
Can you at least assure me that with the higher input quality (I’m using ProRes → ProRes, will start doing ProRes → Tiff once my new SSD arrives) I’m presumably more assured of much higher fidelity in the upscaled result?
And any way to leverage the CPU in this action also? I was previously getting 13sec/frame with a 3900x. This 5950x has 50% more cores and much better IPC… so should almost reach a 6-8frame/sec mark alone. Combining them would surely be the best bet?
VEAI uses tensorflow to run. CPU has no tensorflow, so it will be much much slower. I think you have a typo overthere (6-8frame/sec => 6-8sec/frame). You cannot take out the details, you can only reduce the details. Reducing it or increasing it makes no difference, because it still has to run those filter anyway, even it’s almost nonexistent.
Only TIF and TIFF run fastest because these formats are uncompressed. It means CPU does not need to compress the images or video (ProRes or MP4). But you need a good GPU. Running VEAI using CPU alone will take so so much time.
Yes indeed sorry, should have been secs/frame.
As I said though, the 3900x alone was also pulling 13-14secs/frame when not using the GPU. Which is not terrible, compared to 6secs/frame with the GPU. Which is why I’m surmising that a combined effort should somehow be ideal (and even more so with the 5950x). Even if it’s only a small contribution. Or maybe they can each focus on their ideal tasks, like Resolve does… and actually accelerate over one-or-the-other in isolation.
OK if reducing or increasing the difference doesn’t make a difference then why did you say earlier that to have it run faster I should feed it 1080p instead of 5.7K?
A related question if I may: How linear is the perf. improvement with a second GPU? And if the 2nd GPU is not the same spec as the first (let’s say a 3080 + a 3070) is this a big disadvantage in terms of maximising secs/frame?
VEAI focuses on GPU more than CPU (about 80%).
More cores don’t mean faster here, but better single core performance will speed up the process (just around 5-10% only).
VEAI runs faster or slower depends on the source. Lower resolution sources will upscale faster. Higher resolution will be slower.
480p => 4K : 0.14 sec/frame
1080p => 4K : 0.32 sec/frame
You know what I mean.
More GPU = faster. But you should use 1 GPU only if you don’t use VEAI in hardcore mode 24/7 like me.
Let’s say you have 2 different GPUs, each GPU will do its job. GPU 1 will work on even frames, the other will work on odd frames. But if one GPU is slower, it will affect the faster one. That’s why you need 2 identical GPUs for optimal performance.
OK. So (let’s say) a 480p → 4k will be roughly 2.2x faster than a 1080p → 4k upscale. Will the 480p → 4k upscale look the --same-- as the 1080p → 4K? Or will the 1080p → 4k result be at least 2.2x as good in terms of eventual picture ‘quality’ ( in the viewer-relative, intangible sense)?
Understood on the GPU matter - thanks.
Unfortunately, 480p => 4K is not as good as 1080p => 4K. 2x is the best quality you can get. Don’t do 4x if you don’t know what you’re doing.
RIght, I’m not intending to of course, I have the 5.7k source already.
Does PCie bandwith make a difference for the frame performance? I could potentially fit up to three cards in this machine but the 3rd slot’s a 4x only. Would that hinder performance from the other two slots?
I’m using PCIE 4.0 x4 as my second slot. No problem or difference at all. Don’t worry. But you should worry about your power supply because VEAI is very power hungry at 4K and above.
