[Greg Estes]

Yep.

A Q4000 gets you faster/better SDI I/O capability and the ability to do stereo editing with 3D Vision Pro. It’s also the go-to card for some other apps you may care about in your shop. Like Avid or Maya. So if you care about those things, you may want to think about that. But if you don’t, then yes I think you have it right.

[Greg Estes]

We are supposed to have some new benchmarks next week that include K5000. If I forget to post, ping me.

In general the answer is going to be different if you mostly use PPRO or if you do a lot of AE. AE uses both GPUs in a system and works best when they are the same (GPU ray tracing with OptiX needs all GPUs to have the same frame buffer memory size, so it will default to lowest if they are different). OTOH, Premiere Pro uses only one GPU at a time, so for primarily PPRO use, a little Quadro and big Tesla is the best price/performance.

So dual K5000’s would solve both problems, but if you mostly do PPRO then you might be over-spending.

And, it must be said, Adobe does not officially support Kepler in CS6 ;-(

[Greg Estes]

Mixing Kepler and Fermi is not tested, supported or recommended.

It might work if you try it, but that would be very application dependent and when I checked with the product team they gave me the “not tested, supported or recommended” answer.

It’s the technical dissimilarity, not a random marketing decision.

[Greg Estes]

Understood.

We design GeForce primarily for gamers and exclusively for consumers and there are a handful of features in Quadro that make a difference for pros. Totally get it that many users will compare raw performance of a GeForce with that of a Quadro and find GeForce is as fast (or faster due to overclocking, etc.) In other cases Quadro will be faster.

The value of Quadro isn’t necessarily speed. It’s features, reliability, consistency (we build and test Quadro ourselves; GeForce is made by something like 32 different manufacturers each with their own approach) and, yes, sometimes speed (due to chip-level feature differences like dual copy engines and such).

And if you don’t need or care about those things, then GeForce might very well be a great choice for you. No argument there.

[Greg Estes]

Sorry – that was unintentionally misleading. I meant the Q6000 for both the performance being the same and the price being less

Quadro 2000 + Tesla C2075 would be street price of about $2700.
Quadro 6000 has a street price of around $3250 (varies a lot by reseller)

Those two will perform about the same for Premiere Pro. And you could downgrade the Q2000 for a Quadro 600 and save another couple hundred bucks but I don’t recommend it, just because the Quadro 600 won’t be the right choice for almost any other pro app you want to run.

The Quadro 5000 is around $1750, so costs less and performs less, too, relative to either a Quadro 6000 or a Maximus (any config), but it will perform better than a single Q2000 or Q4000.

Hope that all makes sense. Remember this is for Premiere Pro we’re talking about. Other apps might be a different story. Assimilate tested Maximus for Scratch and said a single Q6000 performed better, as one example.

Net-net: Best bang for the buck for many users is Q4000. Best price/performance for serious pros is Maximus (with modest Quadro plus Tesla C2075) unless you are doing a lot of After Effects then upgrade the Quadro to be beefier because AE uses both Quadro and Tesla together for multi-GPU rendering.

[Greg Estes]

I’ll try to dig up some good benchmarks. Basically, here is the way I think most users should make a good choice:

1. Are you running hard enough problems so GPU acceleration will make a difference? If you are just doing a couple of layers with minimal effects, then the GPU acceleration won’t change your life. If you are working with ARRI or RED files and lots of layers, then you’re crazy to not have a GPU.
2. Each Quadro provides better performance than the one below it in the product line. Q5000 outperforms a Q4000 which outperforms a Q2000 and so on. Adobe has done a very good job of using CUDA and it shows when you throw more CUDA cores at a given problem size.
3. Once you get to the point where a Q5000 or Q6000 might make sense, flip over to Maximus. The reason is a Tesla C2075 will perform (for Premiere Pro) almost exactly the same as a Q6000 but it costs on the order of $1000 less. So save your money and buy a Maximus with a small Quadro plus a C2075.
4. This logic stops being true if you do a lot of After Effects, because AE uses *both* GPUs and now a small Quadro and a big Tesla won’t make as much sense as a more balanced system.
5. For most users, a Quadro 4000 with a street price of $750 USD makes the most sense. It will accelerate Premiere Pro, After Effects, SpeedGrade and Photoshop not to mention Nuke, Maya, 3ds Max, etc etc.

We would love for everyone to buy Maximus of course, but we’re actually more interested in having people get the most value for the money, and for 75% of PPRO users, that’s probably just a Quadro 4000. For power users, Maximus will make more sense than a single Q6000 or Q5000 because it performs the same, costs less and gives you the interactivity I mentioned in my prior post.

[Greg Estes]

Hey Jud and Todd and others,

I can maybe save you some money. When you run a Maximus system (i.e. a Quadro and Tesla together in a single workstation) and Premiere Pro, the Tesla is doing all the work. Maximus technology routes all the CUDA functions to the Tesla, and all the OpenGL stuff to the Quadro. MCP in Premiere Pro is all CUDA (or maybe OpenCL on a Mac but that’s another story). The simple benefit of Maximus technology is you can absolutely slam the Tesla with a render or encoding or whatever, and the Quadro just hums along so your user interactivity remains nice. If you had a single Quadro doing all the work there would be some contention and your interactivity will suffer. So in general a Quadro 2000 + Tesla C2075 will perform the same for Premiere Pro CS6 as a Quadro 6000 + Tesla C2075, but cost about $3,000 less.

Now, After Effects CS6 would put both a Q6000 and C2075 to full use, because AE uses our OptiX technology which is multi-GPU. But Premiere Pro is really a single GPU app, which is running all effects just on the Tesla and the GUI on the Quadro. Other Maximus apps like Quantel Pablo Rio and DaVinci Resolve work in a similar way, so go heavy on Tesla and light on the Quadro is best value for you.

And as far as too much geeky non-media/entertainment stuff on the NVIDIA site. Sorry about that. Always working to improve but we’re a big company. Any and all can email me with questions and I’ll route to people smarter than me and get you clear answers. gestes@nvidia.com

Cheers,
Greg

[Greg Estes]

Hey there.

Be careful comparing Fermi cores to Kepler cores. It’s a different architecture so it’s not an apples and apples comparison.

Greg

[Greg Estes]

Some quick points:
– CS5.5 isn’t multi-GPU in the sense that it takes advantage of two GPUs by spreading a task over both GPUs at the same time.
– CS5.5.2 does support an NVIDIA “Maximus” configuration with both a Quadro and a Tesla in the same workstation. The Premiere Pro CUDA work gets routed to the Tesla board, freeing up the Quadro for other stuff. The combination of a modest Quadro and Tesla C2075 (typical Maximus config) will cost less than a Quadro 6000 but offer about the same performance, so that may be interesting to high end users.
– Only a small portion of AE capabilities are GPU accelerated in CS5.5; that’s correct. Same for Photoshop. Premiere Pro is the heavy hitter in terms of GPU acceleration today.

[Greg Estes]

Obviously we can’t discuss futures, and so even if he would have asked me about that, I wouldn’t have been able to answer. And one shouldn’t read anything into a non-answer – no public company can comment one way or another on futures.

In terms of some of the other comments or questions:
– We are very interested in hearing from COW users on what you would like to see in future NVIDIA products (for Mac or PC). You can ping me directly at GregEstesME@nvidia.com. (Don’t expect comments back on future products 😉
– The reason to buy Quadro vs. a GeForce is always a hotly debated topic on forums. I won’t debate it here, but it comes down to Quadro is built, tuned and tested by NVIDIA for professional use by professionals in workstations. GeForce is built by a number of manufacturers and tuned for consumers doing gaming. There are ramifications to each approach and the net is that we – and our software partners with whom we work very closely to tune performance – recommend Quadro for professional use.
– We just launched a new newsletter yesterday specifically for the media and entertainment community. You can expect us to publish GPU-acceleration-related information from our partners like Adobe, Avid, Autodesk and many others going forward. Our goal is to have it be useful and not be a bunch of marketing BS about how wonderful we think we are. I am sure you will tell me if we meet that goal or not. If that’s interesting to you here is a link https://www.nvidia.com/object/io_1320921041194.html

– Greg E.