Top 5 budget video cards to use with Sony Vegas?

[Dave Osbun]

I’d go with the Radeon HD7870 LE. You should be able to find one for around $230. Radeon cards work fine with Vegas 12 Pro (I personally use the Radeon HD7850). I can’t tell you how well it’ll take advantage of the GPU rendering since I don’t know how CPU you’re using, nor the rest of your system specs.

In the $180-$280 price range, the performance of the Radeon cards are better than the nVidia cards (i’m not saying that nVidia is a bad choice, however).

Dave

[Norman Black]

Of course GPU accel will work with AMD. Why would it not? The not working things are really Vegas bug issues and that applies to all video cards on all brands.

I have had an AMD 5850 and recently upgraded to a 7950. You can get a 7950 for under 300. Most will be right at 300. It is probably the fastest “compute” card for that price unless you can get an older top 500 series for a better price. The AMD 79xx cards were a new design optimized for general purpose compute. Nvidia optimized the 600 series cards to game like render performance verses the 500 series. Still, the 600’s are no slouch to be sure.

You can look at benchmarks for real world verses paper, at places like Anandtech but that shows 90% games. Anandtech does have a set of “compute” benchmarks they reliably run on all cards they test. Not much but better than nothing.

Exactly how all that all that measures to Vegas who is to say. Just like the game benchmarks, one card “wins” this and another wins that. Is Vegas, this, or is it, that.

[David Alfredo]

I agree with Norman, the 7950 is a fantastic computing card, specially for OpenCL, the Nvidia GTX 600 series was also so good at computing that Nvidia crippled their performance in order to prevent these cheaper cards to compete against their more expensive Quadro cards… but still when running CUDA-optimised code or DirectCompute instead of OpenCL their performance is higher than that of ATI/AMD cards…

but this is the point you need to look at, you need software specifically designed to support the new computing architecture found in GTX 600 series Kepler cards… and Vegas doesn’t support them yet, I doubt very much we will ever see an update to Vegas taking full advantage of these cards, most probably we will have to wait until Vegas Pro 13.

With this said I’d go with either some cheap NVIDIA GTX 570 if you can find one or ATI/AMD 7950 instead, using CUDA for the former and OpenCL for the later… though personally I’m not very fond of OpenCL implementation in Vegas… perhaps I’d seek that GTX 570 for maximum compatibility and performance in Vegas 11-12.

[Norman Black]

“but this is the point you need to look at, you need software specifically designed to support the new computing architecture found in GTX 600 series Kepler cards… ”

What the heck is this supposed to mean?

CUDA and OpenCL are more similar than different in general compute and neither one cares about Fermi/Kepler/GCN or whatever. And they never will. That is the whole point. Hardware independence. They are programming languages. Both AMD GCN (79xx) and Nvidia since Fermi are designed around thread parallel execution, so optimizing an app for “that” optimizes for both AMD and Nvidia these days.

Kepler did one thing. It left out execution units/features which games typically don’t use. This sacrifices general purpose compute performance, that CUDA/OpenCL apps do. There is nothign bad about its compute performance. It actually does very well as it is more efficient than Fermi.

I agree Nvidia is creating segmentation with the 600 series visa vie Quadro.

Why do this? To make a smaller chip, which increases yields and profits and saves power. Or to look at it another way, by saving power you can make a card with the same power demands as before but more execution units optimized to typical game performance and then really kick arse in that arena.

Titan (GK110) is Kepler but puts back all the “compute” features the 600 left out. GK110 also adds 64-bit performance which does not apply to our world here. Also it is just a bigger chop overall. More everything.

“though personally I’m not very fond of OpenCL implementation in Vegas…”

What does this mean?

[David Alfredo]

[Norman Black] “”but this is the point you need to look at, you need software specifically designed to support the new computing architecture found in GTX 600 series Kepler cards… ”

What the heck is this supposed to mean?

it means that the Kepler CUDA architecture is not backwards-compatible with Fermi and older CUDA architectures, apart from a different computing pipeline it features a hardware decoder/encoder not found in previous GPUs… let’s take this example: Folding@Home has been using CUDA acceleration for years, pioneering GPU computing acceleration… but with their standard GPU client the issue was that Kepler cards were only using 25% of their CUDA processors due to Nvidia breaking backwards-compatibility. Solution ? Stanford University had to develop a Kepler-specific CUDA client to use 100% of Kepler CUDA computing power and it did show that, even with the performance crippled, computing was faster than that of the Fermi architecture… same goes to video encoders such as the one built-in in Sony Vegas Pro 11-12, not compatible with Kepler and not accelerating rendering at all. Solution ? Sony will have to update Vegas to GTX 600 cards, no drivers can fix that, only software developers… it’s up to Sony to contemplate whether adding support for GTX 600 series cards is financially viable or not.

[Norman Black]“CUDA and OpenCL are more similar than different in general compute and neither one cares about Fermi/Kepler/GCN or whatever. And they never will. That is the whole point. Hardware independence. ”

sorry but that’s not how it works, even if CUDA and OpenCL are somewhat abstract to hardware configurations the Kepler architecture is a radical change from that of previous Nvidia cards, you really need to update your software to “see” all available computing (CUDA/OpenCL/DirectXCompute) cores in Kepler GPUs, software not optimised for Kepler will only “see” 25% of the cores or none at all, resulting in poor performance or software crashes (Vegas 11-12)

[Norman Black]“though personally I’m not very fond of OpenCL implementation in Vegas…” what does this mean?

it means that I’ve been using a lot of OpenCL and CUDA in Vegas since Vegas Pro 11 with a wide, different array of hardware and CUDA has always yielded better and faster results than OpenCL, I know that Nvidia don’t optimize drivers for OpenCL, they got CUDA and that’s where their R&D money goes, on the other hand AMD/ATI engineers have been tweaking their GPUs for OpenCL computing all these years, resulting in fantastic performance in software which extensively uses all of OpenCL advantages, but in our case, Sony Vegas, Sony engineers started GPU acceleration with CUDA in mind and OpenCL was added later, so it’s only normal that, for now, CUDA works better in Vegas than OpenCL.

Sorry if I’m not making myself clear, I didn’t take English at shcool but German and French, hope that doesn’t annoy you guys much.

[Matt Carlson]

David is right here… let’s sum up.

1) The 600 cards are crippled without licensing the new SDK. To my knowledge only GAME developers are even allowed to do so right now. Even if Sony wanted to implement the 600 series Nvidia may refuse to let them (welcome to the new world where lawyers rule.) A 500 series card runs slightly better than a crippled 600 series card.

2) OpenCL and Cuda are not interchangeable they just coexist because developers implement both. In the case of Vegas OpenCL is used for preview acceleration only. Cuda is the only implementation that uses render acceleration (and that is limited to the Sony avc codecs.)

[David Alfredo]

I’d like to hear how NewBlue got the “green light” from Nvidia since they support the GTX 600 series cards in Titler Pro 2, same for Adobe and their latest builds of After Effects… in my experience an ordinary GTX 660 Ti Kepler card is quite faster than the top of the line Fermi GTX 580 when it comes to editing and rendering with the latest build of Titler Pro 2… also games using PhysX (CUDA-powered phyiscs engine) run way smoother on a Kepler than on a Fermi, you can check this by using simple benchmarking tools like FluidMark 1.5.0 or Direct Compute 2.1

As Matt said, Kepler cards get “magically” fixed when developers license the new SDKs, remaining crippled to all the rest. I fear the Sony Vegas will never get to use Kepler gaming-GPU’s computing power, cheaper cards offering same computing speed as the expensive Quadro line… killing the sales of their overpriced professional cards won’t happen, anyways, as Nvidia commented on this situation, the GTX 600 series is designed for gamers not video professionals, even if the computer power is there it doesn’t mean you have the right to use it how and where you want, something along the lines “you buy our hardware but the way it performs through drivers and SDK still belong to Nvidia”

The one thousand dollar GTX Titan features a revision of the Kepler GPU, specially for the computing architecture, but the backwards compatibility remains broken.

[Norman Black]

Crippled compared to what. A 500 series. In ways the 600 beats the 500, but mostly game/render (OpenGL, DirectX), as designed. It is just different. No SDK can change that.

Kepler is not crippled and only game optimized. The Tesla K20 (GK110) is Kepler architecture and is a pure compute card. The K20X is Kepler and is in the Titan super computer, hence where Nvidia got the name Titan for the new consumer card (if $1000 can be considered consumer).

Kepler is an over all program and data flow architecture. Not a chip. The 600 series GK102 and Tesla K20 (Titan video card) GK110 are Kepler but implement it differently as they target different markets. The data and program flow are the same in all Kepler, and really not much different than Fermi. These differences applications really need not worry about.

I use GPU for Sony AVC and Main Concept AVC, on AMD, via OpenCL. So CUDA is not the only implementation for file renders.

Vegas GPU use is not used for preview only. File render uses it also as Vegas generates the video stream for both preview and file render. Only certain things in Vegas use GPU, so you may or may not see a performance boost.

[Stanislav Alekseev]

What else I have:
i5 – 2500k
4×4 DDR3 1333
Intel P67 1155

So seeing how things are just a tad more expensive around here and we don’t really get promotions like you guys (Black Friday, Christmas etc), I’ve narrow to these choices:

GTX 660
and
Radeon 7850

What are your thought on using effects in Vegas (my understanding is that “levels”, “color correction” etc are all considered effects); what will be better for those or is it the general rule, whichever is better for rendering will be better for rendering the effects?
Should I except a worse GPU acceleration that with my GTX 460 with any of the ones I shortlisted?
In computing power (according to AnandTech, thanks Norman!) 7850 is better that the GTX ?

[Matt Carlson]

Norman you are not understanding the “cripple” here. The new architecture has 4 banks of Cuda. Backwards compatability allows older implementations to see the first bank but that is it. The upper end 500 series has 400+ cuda cores and a even with a less powerful clock narrowly outperforms the 600 series in Vegas (300+ cuda cores with 900 of them being unavailable.)

OpenCL does not help with your rendering. That is your imagination. What it does is help with OFX manipulation of frames but the actual compression of the video itself is only helped by Cuda in Vegas.