You know what.
It's, I think it's about time for an upgrade.
So I've been sporting an EVGA GTX 960 four gigabyte model for just over three years now,
and now that we're two generations removed from that GPU, I decided it was time to upgrade.
This video will look at how I did with my upgrade choice, as well as make some comparisons
to the GTX 1080 at the end.
To start off with, though, why did I choose the card I did; the RTX 2070?
Well for starters it fit in my budget of around $500, but it also offered several significant
improvements over my existing card, the main one being four gigabytes of VRAM rather than
four, which is a bigger improvement than you might think.
See, with system RAM, not having enough generally just means your applications will run slower
as they have to swap data from system storage more often, but that isn't usually the case
for GPU applications.
If you utilize more VRAM than your card has, the application using that VRAM will straight
up crash rather than just run slower.
I've had plenty of Premiere Pro renders where I had to disable GPU acceleration as I didn't
have enough VRAM to meet the demands of that particular render.
Moving up to a much more powerful card with double the VRAM, and, perhaps more importantly,
that matched my PC build's color scheme, was a very useful upgrade for me.
With that said; benchmarks.
For the benchmarking we'll be using PCMark, Cinebench, Blender, a 4k Premiere Render,
and VRMark.
The last one might seem a bit out of place since we're testing workstation performance,
but things like architecture and engineering are moving towards using VR, so it may be
a useful benchmark to look at for overall workstation performance when thinking about
potential future needs.
With PCMark, which is testing general PC use, we saw a solid 900 point increase between
our two GPUs.
That increase isn't significant, but it's actually bigger than I was expecting since
faster storage, more RAM, and a faster CPU have a bigger impact on the type of things
that PCMark tests, such as web browsing and video chat.
Either way it's a welcome improvement.
Cinebench is a 3D rendering test and, well, our significantly more powerful GPU actually
performed worse, and I tested it multiple times to be sure the slower number wasn't
just an anomaly.
The difference is within what I would consider the margin of error, but I really don't know
why we didn't see noticeably higher numbers.
If anyone has any ideas what might've gone wrong, please let me know in the comments
and I might follow up in a future video.
Whatever though, on to Blender.
With Blender, another 3D rendering benchmark, the GTX 960 scored in at 21 minutes and 37
second.
Not that matters though, because the RTX card didn't even complete the benchmark.
The reason is that Blender doesn't yet support the version of CUDA on the newer RTX cards;
CUDA being the programming interface used to control the cards.
So that's a DNF for the RTX card.
With Premiere we finally get to see a different story.
The RTX card managed to render my 4k composition about 5% faster than the 960; a small but
noticeable jump in performance.
The composition was a compressed 4k file with Lumetri Color corrections, Red Giant Cosmo,
and FilmConvert applied, being rendered as a 4k DNxHR file.
The difference isn't huge, but again remember that this newer card can render things that
will crash on the 960.
Lastly, the RTX 2070 finally gets a chance to totally trounce the GTX 960 with VRMark,
where it scored over double what the older GPU managed.
That means going from just barely being able to render 3D VR games at acceptable framerates,
to being able to handle VR rendering without breaking a sweat.
Pretty sweet.
So what does this all mean?
Well, to answer that let's talk about the start of this new GPU's name; RTX, changed
from the GTX branding on previous generation cards.
RTX is a new technology Nvidia developed to allow for real time ray-tracing in video games.
This allows for much more accurate lighting, reflections, and shadows, but it comes at
the cost of being very demanding.
To compensate Nvidia included specially designed Tensor Cores in the RTX cards to handle the
ray-trace rendering.
Those cores were previously only available in Nvidia's Tesla GPUs, which are designed
for AI and deep-learning.
That means the new cards are capable of running this RTX technology, but it also means general
performance is cut back.
The GTX 1080, which is right now retailing around the same price as an RTX 2070, has
2560 CUDA cores compared to the 2070's 2304.
The 2070 also has a reduced core clock speed, but it does have significantly faster and
wider bandwidth memory, which makes a really big difference in GPU performance.
But how do all of those specs translate to actual performance?
Well it depends on what type of applications you're looking at.
For things that are primarily CUDA based like Premiere accelerated rendering, the two cards
will trade blows, but when it comes to 3D rendering the newer card can pretty consistently
crush the 1080.
That is, if your 3D application supports rendering on the 2070.
And yes, I'm atting you Blender.
That said, that leads us nicely into my conclusion for this video.
The TLDR is; get the GTX 1080 for tried and true performance, and get the RTX 2070 for
future performance.
The GTX 1080 is a very solid card, and you can get them very cheap used, or even new.
Issues with RMA have been sorted out, unlike RTX cards which have had issues as significant
as reportedly spontaneously combusting.
The 1080 sports more CUDA cores which are supported by essentially every application
that utilizes GPU acceleration.
On the other hand, the RTX 2070 is looking ahead with its addition of Tensor Cores.
Those Tensor Cores are untapped hardware that could be utilized to consistently give the
RTX cards an edge over the previous generations in workstation applications.
There are already developer tools available to take advantage of those Tensor Cores, so
it's just a matter of time before we start seeing updates in popular GPU accelerated
applications like Blender and Premiere Pro to utilize that hardware and get additional
performance boosts.
Either way though, you can't really go wrong, as both cards are great options, and either
one is certainly a fantastic upgrade from an older card like my GTX 960.
Anyway that's it for this video.
If you liked this video, if it was useful, if it was an interesting comparison and analysis,
hit that like button.
If not feel free to hit that dislike button.
If you have any questions or comments feel free to leave those down in the comments section
down below.
And of course, if you wanna see more videos like this one, please go ahead and hit that
subscribe button.
No comments:
Post a Comment