TVAI 3.1 performance on older PC (i7-6700) + 3060 ti GPU

This is a good news story. Decided to move my RTX 3060 ti temporarily from its usual Ryzen 7 5800x 8 core / 32GB memory home to my old i7 6700 (2016) / 16GB memory system - just to see how it performed on TVAI 3.1.4 - trial version.

The speed results were excellent! On Artemis HQ, an old 640x480 2 times upscale ran at 23 fps on the 5800x but on the 6700, it still ran at 22 fps. I got 85 fps on Chronos Fast 2x slomo, not much slower than on the 5800x (95 fps)

This was only a small snapshot but the moral of the story is, for anyone needing faster throughput on upscaling many old videos, don’t spend £1K or more on a 40 series upgrade without first checking the performance of your existing GPU on your older system if you still have it.

So I can achieve a far greater throughput than even a 4090 on one system would provide, by buying one extra ‘cheap’ 3060 ti for my 7 year old system. The licence seems to allow this, so food for thought, all you potential upgraders out there.

Has anyone else tried this?

Well, CPU and GPU have to be in balance.
I don’t have to PCs, but I experimented a bit with tuning my Ryzen 7 5800X and the RTX 3060. Overclocking CPU from 4.25GHz to 4.5GHz resulted in 10% more power draw while TVAI speed went onyl <1% up. Same effect with GPU overclocking from 1800 MHz to 2200MHz, that did also not increase processing speed.
Therfore I run my system with CPU @4,25GHz and GPU @1800MHz, both >90% load.
I can process comfortably 3 videos in parallel (Proteus, 2x upscaling) with in total ~10fps speed, and have almost no negative impact on other work done in parallel.
Usually I load a sequence of up to 9 videos, activate them all for export, and let TVAI do the job 3 after 3. Off course I care that during that processing I don’t need to reboot the PC (no need to worry for overheating)

hmmm, could you try oc your memory?

Oc the memory of my Threadripper resulted in a increase of about one frame. 8.8 → 9.8

2666mhz → 3200mhz

Eh, didn’t look right, you have a Ryzen.
OC a Ryzen doesnt matter, but whats matter is oc its memory.

Thank you, but I don’t know how to oc memory.
By the way: memory available is 32GB for CPU and 12GB for GPU while both load of memory is very low.

P.s. I’ve set allready memory to 3666 in BIOS (is that what you declare oc?)

I always thought that but what I’m suggesting, at least for the kind of upscaling and slomo that I do, the CPU and GPU do NOT have to be in balance (within reason).

Because with the same modern GPU, I got 95% of the same speed with a 2015 released i7 6700 as with a 2020 released Ryzen 7 5800x. That surprised me - and as I still have the 6700 system I no longer feel the need to upgrade to an RTX 4000 series and bigger PSU for faster throughput, I can just buy a much cheaper 3060 ti for the old PC, use the same PSU and run both computers as and when I need to.

I can run two instances of TVAI too on the 5800x system, not sure about the 6700 setup yet as I haven’t tested it for that.

lets say yes, you could lower it to see if TVAI gets slower, but it will by 100%.

What is the memory speed on both systems?

Since both are DDR4, im curious.

2400 and 2600MHz on the 6700 and 5800x respectively.

That’s really low for the 5800x. 3200 would boost it quite a bit, I think.

I’ve never much been interested in overclocking or fast memory tbh but I may try it now.

Incidentally, the 6700 system has its original GTX 1070 ti back in it now. It runs the same 640x480 video at 2x upscale - Artemis High - at 10.5 fps, compared to 22 fps with the RTX 3060 ti. I have an even older i7-4770K system with 8GB RAM and a 1070 GPU, that does it at 8 fps but I’m not planning to test it with the RTX 3060 ti.

I just enabled XMP, (That’s what my BIOS calls it, even though it’s supposed to be something else for AMD…) but that’s what my sticks are rated for. I hear RAM overclocking it the most tedious.