We do work with clients to produce content; however, we retain all rights to present accurate observations, testing, and conclusions. Consumers should have all the facts, and we’ll deliver that.
UPDATE 12/15/2020: Get the RTX 30-series instead… if you can find one.
Hey what’s up everyone. So let’s talk about upgrading your GPU for Premiere or Blender. This topic is pretty important to me as both a video producer and content creator, and several projects it was time to look at how to optimize production.
Both of these Nvidia GPUs are very good cards, right in the sweet spot for price vs performance, which we think most buyers have or will consider buying. As with most video cards these days, they aren’t exactly cheap, so what kind of performance improvements can you expect if you take the plunge? We hope to answer that today, with some benchmarks, and let’s get to it!
RTX 2070 Super performance, RTX 2070 premiere
So this is the continuation of our previous Intel vs AMD Creation episode, and if you haven’t seen that yet, you can check it out here, don’t worry I’ll wait for you. Those results saw a very decent gain with the new Ryzen 3900X, though if you’re using CUDA and have a decent amount of GPU workload while using Premiere then you’ll be looking for a new graphics card, I know we were.
Full disclosure, I do test and host episodes for MSI’s Pro Cast, and while I do have access to some hardware due to testing, both of these cards were purchased. At Techspin, we always give our honest reviews and opinions, so let’s get started.
Our Build with the 2070 Super Premiere
For the Intel build we have the Intel 9900K on an MSI MAG Z390 Tomahawk, Samsung 970 Evo 500 gig NVMe m.2 and 64 gigs of Corsair Vengeance DDR4 3200. AMD Ryzen 3900X we had on the MSI MEG X570 ACE, with an XPG 8200 PRO 256 gig NVMe m.2, which is on-par with Samsung, and 32 gigs of Corsair Dominator Platinum RGB 3000. Cooling was provided by a Corsair H150i 360 mil AIO cooler, which is best case scenario for both of these CPUs.
Stock settings were exactly that, with no tweaks in BIOS, so XMP is disabled. Overclocks had XMP enabled, only adjusting Core ratio in BIOS. All tests were run three times, best of three was taken. Both the X570 ACE and our Z390 Tomahawk have the latest BIOS-es installed, and the latest drivers, and the latest NVIDIA drivers. Premiere testing was done with a 2 minute 26 second 1080p project for Advantech, with some green-screen throughout, but no LUTs applied. We rendered with only CPU and then with CUDA enabled, and due to the green-screen ultra key used, the GPU gets a VERY good workout with CUDA enabled. Lower times are better.
GTX 1660Ti Premiere i9-9900K, 3900X tests
So starting with our MSI GTX 1660 Ti, last time we saw stock Intel CPU only time from 5 minutes 7 seconds(307) and 4 mins 12 seconds(252) at 5 gigahertz vs Ryzen 3900X stock at 4 and a half minutes(270) and 3 minutes 20 for an overclock of 4.2 gigahertz.
Testing with the GTX 1660 Ti and CUDA enabled cut down the stock Intel times to just 1 minute 39, and the 5 gigahertz OC gets it down to 1-37. For Ryzen the 1660 Ti posts a 1 minute 32 second time, at 1-31 with the 4.2 overclock.
With 100% GPU for the majority of the render, the bottleneck is of course now the GPU, so we ran out and grabbed the a new MSI RTX 2070 Super Gaming X Trio, a huge, huge card. This ended up being a good thing though, as without a case now swapping the Corsair H150i Pro to the Intel rig was much less of a pain and consequently, we got better and fairer results with premium CPU cooling.
RTX 2070 Super Premiere i9-9900K, 3900X tests
With our new 2070 Super installed, the Intel CPU at stock posts a great 1 minute 16 seconds, and with the 5 gigahertz overclock we hit 1 minute 12 seconds, the magic mark for me as we’re rendering in less than half the time of the actual clip. Ryzen at stock with the 2070 Super did 1 minute 8 seconds, with 1 minute 3 at overclock, shaving 5 seconds off the total time. That is pretty incredible, no matter which platform you’re looking at.
Because Premiere can utilize Intel’s on-board UHD 630 graphics, we did enable that and the results were worse than using just the 2070 super, adding an extra 22 seconds, making one minute 34 seconds.
Just for fun we tried Intel UHD graphics, the 2070 super in the main PCIe slot and a 1660 Ti very bottom, however Premiere rendering favored the 1660 Ti and UHD maxing them both almost 100 percent, while the 2070 super sat around 30 percent for the most part, dropping to 0 occasionally.
It also blue-screened my first attempt with a WHEA un-correctable error, but the second attempt completed- not the most stable setup. I’m going to speculate that the workload was split three ways, and the 2070 was waiting for the other two to catch up, with a unimpressive result so we didn’t record it.
Mixed GPUs is a bad idea
Turning off onboard graphics, last we tried the 2070 as main and 1660Ti bottom, and the result was terrible, first time blue screen, second time finishing at 2 minutes 35 seconds, almost a full minute behind just the 1660Ti by itself, which finished 1 minute 37 seconds at 5 gigahertz overclock. Premiere’s scheduler barely touched the 2070 at all.
RTX 2070 vs GTX 1660Ti, Conclusion
Okay, time for conclusions. Within this year I’ve actually tested an older GTX 970, an ASUS GTX 1070 for Premiere before running out and grabbing two new GPUs. Looks like a European vacation is on the shelf for this year heh. Thought this might be interesting for you if you’re running older hardware and looking to upgrade, with CUDA enabled, the MSI GTX 970 I have on hand posts 2 minutes 31 seconds(151), and the ASUS Strix GTX 1070 crosses the line at 1 minute 47 seconds(107). The 1070 is still a fairly decent card, with 30% better time than the 970. The 1660Ti at 1-39(99) is 34% faster than a 970, 7% faster than the 1070.
The 2070 Super coming in at 1 minute 16 seconds(76) is not the same price point as either the 970 or 1070 even with inflation, but is 49% more powerful than a 970, 29% faster than a 1070, and 23% better than a 1660Ti.
Given that this 1660Ti goes for 300 bucks and this new 2070 Super is sold out everywhere, but evidently runs about 540, that’s a 44% price increase for only 23% more performance. This isn’t a knock against MSI, as with all GPUs, you pay to get better performance, with higher cost for top-end cards.
The takeaway here is if you’re not needing the most powerful card right now, the 1660Ti is a very decent upgrade from older hardware. If you need better rendering times or are playing 1440p titles on Ultra or even 4K on medium to high settings, the 2070 Super is a very good choice for the money.
As we talked about in our last video, if your Premiere timeline uses the GPU a fair amount, you should be able to get a better boost from a simple graphics card upgrade rather than a whole motherboard and CPU bundle. And for gaming it’s a no-brainer, even with hardware in the last 5 years, a new graphics card will see the greatest FPS gains.
Now I’m really dying to see NVLink results with the 2070 super, information on this is pretty scarce online for Premiere with testers focusing on games, so it’ll be interesting to see if it actually works for Premiere, and most importantly, if there’s a substantial benefit to this configuration, or if it’s just better to save up for a top card like a 2080Ti. If you live in the Taipei area and have a 2070 super I can borrow for a day, hit us up on Facebook or Twitter!
We really appreciate you watching this far, and now we want to ask you, what kind of setup are you using for Premiere, or what GPU are you looking at now? Let us know in the comments. If this video was useful, be sure to hit that THUMBS UP button, or tell us how we could improve for next time! Please take a second to hit Like, subscribe, and the bell to get notified when we put up a new video! We always check the comments and respond to most, so if you have a question or if we missed something, please tell us down below.