Quadro or GTX for new system?

[Paul J Steinberg]

I'm building a new PC that will be running ScratchLAB and Adobe CS6 suite. One thing I can't nail down is the GPU. Almost all of the setups I have seen listed for higher end systems all use the Quadro line of cards (I was thinking of the Quadro 4000 or 5000) but the newer GTX line are now PCI3.0 enabled, have A LOT more CUDA cores and are a fraction of the price. I've gone through the Adobe forums and the resident hardware guru (Harm) over there seems to favor the GTX line but then again it seems like most folks over there are working on h.264 and AVCHD files.

What would the ideal card for working with R3D media in ScratchLAB/Speedgrade/Premiere (already have a Rocket card for the machine)?

[Jeff Kilgroe]

GTX. Done.

You only need to spend the extra on a Quadro if you actually need a few of its unique features. And as of this time, the only unique Quadro features supported by Adobe would be the SDI output option and that's a huge upgrade expense to the Quadro card. You're much better off to use a DeckLink, Kona or similar system. I would talk with Assimilate and see what their current thoughts are on GPUs for Scratch as I'm not up on Scratch that well lately. I guess I should probably pay them a visit at NAB in the next couple days. Personally, I would avoid the Quadro 4000. The 5000 is just OK and the 6000 is really nice. But the new Quadro cards arrive here in about 2 months...

As for the GTX cards, the new ones do have more CUDA cores, but they are a vastly different design from the previous series. The GTX680 is showing to only be about as fast, as a GTX580 for CUDA acceleration at the moment. I'm sure it will improve a bit with driver updates, but don't expect it to be much different/faster in the end. The GTX680 is going to be superior for OpenGL acceleration and other aspects though. It has double the bandwidth for inbound data and nearly 3.5 times the bandwidth for outbound data, due to the PCIe 3.0 spec and the improved design on the output path for the PCIe bus. So for GPU-assisted tasks like H264 rendering and other functions it provides, it should be a good deal faster than the GTX5xx series.

I would recommend one of the 4GB GTX680 models which haven't hit the market just yet. Probably within the next 2 or 3 weeks and they will be hard to get ahold of at first. Cards like the EVGA GTX680 4TW Edition. If you absolutely must buy now and don't want to wait, the GTX580 with 3GB is a great choice still and prices are dropping.

[Paul J Steinberg]

Brilliant! Thanks a ton Jeff. I'm not pulling the trigger until after NAB just to see what new stuff may come out.

The GTX580 seems to be a real beast both in terms of size, power and heat. The GTX680 seems a little better. Maybe a little time will allow Adobe to certify the 680 card. I'm currently running a GTX480 thinking that it would be certified and it never was that has meant applying the 'hack' to get the MPE to run with the CS5/5.5 suite.

[Danno Anderson]

http://ppbm5.com/MPE%20Charts.php

Benchmark results should answer your questions.

[Les C.]

GTX. Done.

You only need to spend the extra on a Quadro if you actually need a few of its unique features. And as of this time, the only unique Quadro features supported by Adobe would be the SDI output option and that's a huge upgrade expense to the Quadro card. You're much better off to use a DeckLink, Kona or similar system. I would talk with Assimilate and see what their current thoughts are on GPUs for Scratch as I'm not up on Scratch that well lately. I guess I should probably pay them a visit at NAB in the next couple days. Personally, I would avoid the Quadro 4000. The 5000 is just OK and the 6000 is really nice. But the new Quadro cards arrive here in about 2 months...

As for the GTX cards, the new ones do have more CUDA cores, but they are a vastly different design from the previous series. The GTX680 is showing to only be about as fast, as a GTX580 for CUDA acceleration at the moment. I'm sure it will improve a bit with driver updates, but don't expect it to be much different/faster in the end. The GTX680 is going to be superior for OpenGL acceleration and other aspects though. It has double the bandwidth for inbound data and nearly 3.5 times the bandwidth for outbound data, due to the PCIe 3.0 spec and the improved design on the output path for the PCIe bus. So for GPU-assisted tasks like H264 rendering and other functions it provides, it should be a good deal faster than the GTX5xx series.

I would recommend one of the 4GB GTX680 models which haven't hit the market just yet. Probably within the next 2 or 3 weeks and they will be hard to get ahold of at first. Cards like the EVGA GTX680 4TW Edition. If you absolutely must buy now and don't want to wait, the GTX580 with 3GB is a great choice still and prices are dropping.

Don't assume the GTX680 will be faster for opengl. They crippled opengl (for non-game apps) after the 285, so that the 580 as 10 times slower than the 285. Unless you are running a game, opengl is most likely going to be very slow.

CUDA in the 680 will vary depending on whether you need 32 or 64 bit float. 64 bit is going to be slower.

[Paul J Steinberg]

http://ppbm5.com/MPE Charts.php

Benchmark results should answer your questions.

Thanks Danno. I'm surprised how well the Quadro 4000 is doing on those tests.

[Jeff Kilgroe]

Don't assume the GTX680 will be faster for opengl. They crippled opengl (for non-game apps) after the 285, so that the 580 as 10 times slower than the 285. Unless you are running a game, opengl is most likely going to be very slow.

Where did you get that info? They have done no such thing. I practically live in OpenGL accelerated applications -- Maya, XSI, Lightwave, Modo, Pro/E and more. The only thing that is crippled in nVidia land is OpenCL and it's not so much that it's crippled intentionally, but rather as a byproduct of the current implementation as it takes a back-seat to CUDA. I also write my fair share of OpenGL code too...

Still waiting for the 4GB GTX680 cards here, but my GTX560 and GTX580 cards are smokin' compared to the GTX285. Yes, in OpenGL. It's not like there's a little leprechaun in there that shifts the card into low gear if they think you might be running something that qualifies as a game. Maya and XSI typically prefer the Quadro cards because of their support for the additional hardware buffers, but these days that factor is relatively insignificant. Where the Quadro cards shine is in precision for edge and point rendering. You get far fewer gap anomalies and artifacts with the Quadro series in their higher precision modes than you do on the GTX cards. But unless someone is running Pro/E or similar on a system on a regular bases, I don't know that I would recommend the Quadro card. I own plenty of Quadro cards too... I often question why I've bought them when I get better performance on other systems with a regular GTX card.

There are indeed some OpenGL functions that are optimized differently between the driver sets of the Quadro and GeForce families. And yes, this can lead to different levels of performance with those functions.

CUDA in the 680 will vary depending on whether you need 32 or 64 bit float. 64 bit is going to be slower.

Yes, the 64bit performance will be quite a bit slower than 32bit on the 680.

[Subhadip Sen]

As Jeff points out, GTX has its merits. However, consider waiting for benchmarks before jumping in. The official preliminary recommendation for Speedgrade is Quadro 5000/6000. Quadro comes with extra features - including 10-bit monitoring, better warranty/support, SDI outs [select models] etc. - and drastically better performance in some cases (certain OpenGL vector functions and double-precision compute). For everything else GTX is the better choice. As for GTX 580 vs GTX 680, it depends on the nature of the compute, varies from app to app. As of now, GTX 580 3GB is a fine choice - prices are dropping drastically (clearing out inventory + the $350 HD 7870 is faster) and for under $400 it's a good deal. As I posted in the other thread, here's some compute benchmarks, GTX 580 is still faster in most situations. It wins in the ComputeMark benchmark but that is DirectX 11 based, and given the multi-platform nature of Adobe CS I doubt DirectX will be leveraged. Even so, there's no Adobe benchmark on that site, so that's something worth looking out for.

As for the GTX cards, the new ones do have more CUDA cores, but they are a vastly different design from the previous series. The GTX680 is showing to only be about as fast, as a GTX580 for CUDA acceleration at the moment. I'm sure it will improve a bit with driver updates, but don't expect it to be much different/faster in the end. The GTX680 is going to be superior for OpenGL acceleration and other aspects though. It has double the bandwidth for inbound data and nearly 3.5 times the bandwidth for outbound data, due to the PCIe 3.0 spec and the improved design on the output path for the PCIe bus. So for GPU-assisted tasks like H264 rendering and other functions it provides, it should be a good deal faster than the GTX5xx series.

Unfortunately the only chipset GTX 680 supports for PCIe 3.0 right now is Z77 for LGA1155. The notable omission here is X79, where it would default to PCIe 2.0. Hopefully it will be enabled by a driver update in coming months. The H.264 rendering is faster due to the NVENC fixed function encoder. AMD and Intel have them too with VCE and QuickSync respectively. Otherwise, the primary bottlenecks for massive amounts of data - memory interface and render output units - were 50% greater on the GTX 580 clock-for-clock. Though offset by higher clocks on GTX 680 and a more efficient ROP array, it only just matches GTX 580, with no advancement.

[Steve Cee]

Best of both worlds?


http://www.nvidia.com/object/maximus.html
NVIDIA Maximus-powered workstations solve this challenge by combining the visualization and interactive design capability of NVIDIA Quadro GPUs and the high-performance computing power of NVIDIA Tesla GPUs into a single workstation. Tesla companion processors automatically perform the heavy lifting of photorealistic rendering or engineering simulation computation. This frees up CPU resources for the work they're best suited for – I/O, running the operating system and multi-tasking – and also allows the Quadro GPU to be dedicated to powering rich, full-performance, interactive design.

I'm building a new PC that will be running ScratchLAB and Adobe CS6 suite. One thing I can't nail down is the GPU. Almost all of the setups I have seen listed for higher end systems all use the Quadro line of cards (I was thinking of the Quadro 4000 or 5000) but the newer GTX line are now PCI3.0 enabled, have A LOT more CUDA cores and are a fraction of the price. I've gone through the Adobe forums and the resident hardware guru (Harm) over there seems to favor the GTX line but then again it seems like most folks over there are working on h.264 and AVCHD files.

What would the ideal card for working with R3D media in ScratchLAB/Speedgrade/Premiere (already have a Rocket card for the machine)?

[Paul J Steinberg]

Hmmmm...so, correct me if I'm wrong, but it looks like a bit of a guessing game as to how to best set up for Adobe's new pipeline/workflow. Currently I see ScratchLAB using the GPU when working/grading and then rendering out to Quicktimes/MXF seems to go to the CPU (this works well in the sense that you can be rendering and working at the same time without working on a slug of a machine). Premiere uses a lot of CPU power but then uses the GPU to work on realtime effects and video decoding but I've never seen my GPU even hit 40% load, usually it sits around 5%-10%. When rendering out I've seen Premiere use maybe 10%-15% of the GTX480 I currently have but Adobe Media encoder (that I typically use) just pushes every thread on the CPU to 100% and the GPU sits idle.

As far as over all system performance I currently have an i7 960 @3.2Ghz, 12Gb RAM, GTX480 with a Red Rocket (3 year old machine). I can playback, edit, grade DSLR, C300 and Red media all in real time no problemo. What I'd like to get is increased render performance . It seems to me that the place to spend money is on CPU(s) unless the Maximus configuration is going to take over render duty.

NAB opens tomorrow so I hope Adobe will post system spec's or even better, suggested system configurations.