-
Picking a card: GTX 550 vs GTX 570
I’m building my first dedicated editing/VFX rig. I’m a VFX artist so I’m looking for a rig that will give me the best realtime performance. Render performance is important as well, but I’m more concerned about being able to see what Particular is doing right now than what it will be doing when I render it. I’ve built tons of computers in the past but I’ve never had to really compare video cards like this since I don’t have a very good handle on how AE CS6 utilizes video cards. Right now I’m looking between two cards:
Here’s what I’ve gathered so far:
Support
According to this page, AE CS6 supports the 570 but the 550 isn’t listed. This right here may be all I need to know, but I’m not sure if the page is really up-to-date.
CUDA Cores
From what I’ve read, lots of CUDA cores = good. The one obvious discrepancy is here, where the 570 boasts 480 CUDA cores, whereas the 550 only has a measly 192. Using SLI (which would allow me to buy two 550s for the same price of one 570) I’d be rocking out with 384 CUDA cores.
According to this page, AE will utilize the CUDA cores of all installed GPUs, so as long as AE actually supports the 550 I’m not doing too shabby.
GPU
GTX 570
– Core Clock: 732MHz
– Shader Clock: 1464MHz
GTX 550:
– Core Clock: 951MHz
– Shader Clock: 1903MHzThis definitely confuses me. The 570 boasts more than double the number of CUDA cores that the 550 has, but it has lower clock speeds. Maybe those two things aren’t mutually exclusive but from what I thought I knew about CUDA technology, higher clock speeds means support for more CUDA cores.
Still, experience has taught me that a higher clock speed is always better. Is that the case here? Or would I be better off going with more CUDA cores instead?
VRAM
GTX 570
– Total VRAM: 1280MB
– Clock Speed: 3800MHz
– Interface: 320-bit
GTX 550
– Total VRAM: 1024MB
– Clock Speed: 4356MHz
– Interface: 192-bitAll these numbers really mean to me is that the 570 has a little over 1GB of VRAM and the 550 has right at 1GB. However, I’m putting 16GB of RAM in my machine. How much is the VRAM going to matter in that case, and how much of a difference are the clock speeds going to make?
Also, keep in mind, I’d be using SLI if I went with the 550, so it would actually have 2GB of VRAM. Would it be more beneficial to have more RAM or faster RAM?
Like I said, all of this research is moot if AE CS6 doesn’t support the 550. However, there’s a great deal going on right now where I could get two 550s for the same price of one 570. If anybody could clear up any of these issues for me, I’d greatly appreciate it. Thanks!
-
6 Replies
-
[Trezy Peebles] “Render performance is important as well, but I’m more concerned about being able to see what Particular is doing right now than what it will be doing when I render it.”
Upgrading your GPU won’t help you at all with this. Normal 2D and Classic 3D rendering in After Effects is done exclusively on the CPU.
Ae CS6’s new ray-tracing renderer is CUDA-accelerated, and greatly benefits from an approved NVIDIA card.
Some third-party effects like Sapphire or Neat Video may use CUDA for processing, and some like Optical Flares, Element 3D, ShapeShifter Ae, Magic Bullet Looks, etc. process on the GPU via OpenGL. The Ae interface itself is also accelerated by OpenGL.
Particular is not GPU-accelerated.
[Trezy Peebles] “Maybe those two things aren’t mutually exclusive but from what I thought I knew about CUDA technology, higher clock speeds means support for more CUDA cores.”
Nope. Core count and clock speed are independent (just as they are with CPUs).
[Trezy Peebles] “Still, experience has taught me that a higher clock speed is always better. Is that the case here?”
Not necessarily true. A 12-core Xeon workstation system at a lower clock speed outperforms a 4-core desktop system at a higher clock speed at many — but not all — computational tasks. It all depends on how parallel the task is.
[Trezy Peebles] “Also, keep in mind, I’d be using SLI if I went with the 550, so it would actually have 2GB of VRAM. Would it be more beneficial to have more RAM or faster RAM?”
Ae’s ray-tracing renderer doesn’t use SLI.
It will, however, use all available CUDA GPUs on your system. Note that VRAM usage will be capped on all cards at the lowest amount on any one card.
[Trezy Peebles] “I’m building my first dedicated editing/VFX rig.”
If you’re considering Premiere Pro as your editor, I’d go with the 570 hands down — Premiere only supports multiple CUDA GPUs in a Maximus (Quadro/Tesla) configuration. It will not support multiple GeForces.
Walter Soyka
Principal & Designer at Keen Live
Motion Graphics, Widescreen Events, Presentation Design, and Consulting
RenderBreak Blog – What I’m thinking when my workstation’s thinking
Creative Cow Forum Host: Live & Stage Events -
[Walter Soyka] “Upgrading your GPU won’t help you at all with this. Normal 2D and Classic 3D rendering in After Effects is done exclusively on the CPU.”
Thanks a lot Walter, that definitely helped me out. I find it strange that AE uses the CPU exclusively but I suppose it makes sense since the CPU is almost always more capable than the GPU.
[Walter Soyka] “Some third-party effects like Sapphire or Neat Video may use CUDA for processing, and some like Optical Flares, Element 3D, ShapeShifter Ae, Magic Bullet Looks, etc. process on the GPU via OpenGL. The Ae interface itself is also accelerated by OpenGL.
Particular is not GPU-accelerated.”
I guess I chose the only plugin I use that doesn’t utilize the GPU as an example, go figure. Most of what I do utilizes either Trapcode Form of Particular, though I also use MagicBullet for coloring and I’m just getting into using Element 3D so the GPU will still be vital for me.
[Walter Soyka] “Ae’s ray-tracing renderer doesn’t use SLI.
It will, however, use all available CUDA GPUs on your system.”
So what’s the technical difference here? I thought SLI was just a technology allowing your machine to utilize all cards connected through SLI, so how is it not doing that by using all available GPUs and VRAM?
[Walter Soyka] “If you’re considering Premiere Pro as your editor, I’d go with the 570 hands down — Premiere only supports multiple CUDA GPUs in a Maximus (Quadro/Tesla) configuration. It will not support multiple GeForces.”
Premiere Pro has been my editor for years so I don’t think I’ll be changing anytime soon. However, I use Premiere Pro only for editing in my workflow. No effects, no transitions, no stabilization (though the addition of stabilization to CS6 is tempting). Hence, I’m not particularly concerned about it’s performance on this new rig. It’s still something I’ll keep in mind.
-
[Trezy Peebles] “Thanks a lot Walter, that definitely helped me out.”
You’re welcome!
[Trezy Peebles] “I find it strange that AE uses the CPU exclusively but I suppose it makes sense since the CPU is almost always more capable than the GPU.”
It’s not so strange when you consider Ae’s heritage, and the fact that numerical computation on GPUs has only recently become practical.
[Trezy Peebles] “I guess I chose the only plugin I use that doesn’t utilize the GPU as an example, go figure. Most of what I do utilizes either Trapcode Form of Particular, though I also use MagicBullet for coloring and I’m just getting into using Element 3D so the GPU will still be vital for me.”
OpenGL performance on these cards will probably be a wash. I’d look to cards with more VRAM, as Element 3D stores textures and meshes on the GPU.
[Trezy Peebles] “So what’s the technical difference here? I thought SLI was just a technology allowing your machine to utilize all cards connected through SLI, so how is it not doing that by using all available GPUs and VRAM?”
SLI is a technology for linking multiple GPUs for image processing. CUDA is a technology for general-purpose computation on the GPU. Simply put, SLI works on pictures, but CUDA works on numbers. The two technologies are not related.
[Trezy Peebles] “Premiere Pro has been my editor for years so I don’t think I’ll be changing anytime soon. However, I use Premiere Pro only for editing in my workflow. No effects, no transitions, no stabilization (though the addition of stabilization to CS6 is tempting). Hence, I’m not particularly concerned about it’s performance on this new rig. It’s still something I’ll keep in mind.”
Scaling, deinterlacing, and color space conversions are all processed via CUDA. If ever work with frames of different sizes, different field orders, or different frame rates, CUDA can help accelerate your renders.
Walter Soyka
Principal & Designer at Keen Live
Motion Graphics, Widescreen Events, Presentation Design, and Consulting
RenderBreak Blog – What I’m thinking when my workstation’s thinking
Creative Cow Forum Host: Live & Stage Events -
It is very easy to get confused. I have been looking into this for a while now and some things still elude me.
Small contribution to unraveling this mystery… as far as I know Classic 3D in AE CS6 does benefit from the GPU. Not just the U.I.but also in interaction and rendering. Try the following. Use a layer extruder script with Classic 3D (AKA advanced 3D in CS5.5 and earlier). Then try the same technique in CS6 Classic 3D. Night and day. In CS5.5 you had an option to use openGL in the AE prefs as well is in the render queue. In CS6 it is no longer there, as I understand it, this is because it is on all the time now.
The new AE raytracer is basically NVIDIA’s Optix technology, which if I am not mistaken is Mental Ray ported to the GPU (simplified explanation).
When trying to make a decision on a GPOU there is much to consider, CUDA cores are one thing but many a developer (us included) opted not to do what we do with CUDA. Rather we decided on openGL + openCL. So the trick is try to find a balance. Follwoing is one site that may be of assistance: https://www.videocardbenchmark.net/
HTH
Cheers,
Chris
mettle.com -
[Chris Bobotis] “as far as I know Classic 3D in AE CS6 does benefit from the GPU. Not just the U.I.but also in interaction and rendering. Try the following. Use a layer extruder script with Classic 3D (AKA advanced 3D in CS5.5 and earlier). Then try the same technique in CS6 Classic 3D. Night and day.”
You should see differences in performance here in interaction, specifically on things like drawing the layer boxes, etc.
There is no Classic 3D OpenGL renderer. Any difference in rendering here must be attributed to something else (perhaps the global performance cache or other optimizations)?
[Chris Bobotis] “In CS5.5 you had an option to use openGL in the AE prefs as well is in the render queue. In CS6 it is no longer there, as I understand it, this is because it is on all the time now.”
In CS6, the OpenGL renderer from CS5.5 and previous is gone (because it was so problematic).
https://blogs.adobe.com/toddkopriva/2012/05/gpu-cuda-opengl-features-in-after-effects-cs6.html
Walter Soyka
Principal & Designer at Keen Live
Motion Graphics, Widescreen Events, Presentation Design, and Consulting
RenderBreak Blog – What I’m thinking when my workstation’s thinking
Creative Cow Forum Host: Live & Stage Events -
[Chris Bobotis] “When trying to make a decision on a GPOU there is much to consider, CUDA cores are one thing but many a developer (us included) opted not to do what we do with CUDA. Rather we decided on openGL + openCL.”
Chris, I had no idea you were using OpenCL — is that in both FreeForm and ShapeShifter, or just one or the other? I’d love to add the information to my standard reply about third-party effects that render on the GPU.
Trezy, in case you’re unaware, Chris’s company Mettle [link] develops some very cool 3D plugins for After Effects: FreeForm Pro and ShapeShifter AE. I know you mentioned you use Element 3D, and while there is some overlap between Element and ShapeShifter, I consider them complementary products with different strengths. They are worth a look.
Walter Soyka
Principal & Designer at Keen Live
Motion Graphics, Widescreen Events, Presentation Design, and Consulting
RenderBreak Blog – What I’m thinking when my workstation’s thinking
Creative Cow Forum Host: Live & Stage Events
Reply to this Discussion! Login or Sign Up