View Full Version : nvidia maximus
11-16-2011, 03:25 AM
This ma already be posted. i thought it was a good read.
11-16-2011, 03:31 AM
Yes, a few people have posted about it. From what I have read this won't help much with r3d debayering and the workstations will start at $8,000 to $10,000 USD. If it's possible to use these cards in a custom built PC then that could be very interesting for a Windows based Resolve box.
11-16-2011, 04:42 AM
So, is this sort of a way to get the power two CUDA video cards into one system? We know that PP is not capable of using two GPU cards in SLI. In fact, it cannot even use all the cores on a GTX 590 because that card itself is SLI. So is Maximus essentially like putting two beefy cards into a system and having them share the GPU computing tasks (Mercury, etc)?
Someone at Adobe told me that this system is blazing fast for Premiere. I'm very curious to see some benchmarks or some kind of clear demonstration of exactly how much faster Maximus is than a non-Maximus system. This could be extremely useful for 4K editing, I would think.
11-16-2011, 05:21 AM
What was interesting in the article in reference to RED
"If you're working with a lot of uncompressed RED files with lots of layers, Maximus starts to make a lot of sense."
This led me to believe there is some magic happening in the debayering.
11-16-2011, 06:40 AM
Yeah it was already posted...
Maximus I believe requires a Tesla card which obviously doesn't come cheap, around 2K for one,
It off loads certain elements like graphics and so forth to the Tesla card and the GPU card works the rest, I don't think the Tesla card would be used to power a display nor would it be something in combo with SLI,
PP5.5 is the only program able to use it right now but the benchmarks look pretty damn good...
To say the least, it maybe worth it if you do a lot of graphics, R3D and other stuff in PP5.5 and potentially the rest of the adobe suite. The thing i'm a bit lost about is they discuss OEM providers and don't seem to mention purchasing the Tesla separately which you can do...
Makes me wonder if this is specific to the motherboard for it to be activated, like 'maximus' approved boards or what...
11-16-2011, 07:05 AM
Still does not accelerate R3D decompression or debayer.
As Imran points out, this is a Tesla-based workstation approach. nVidia is trying to push their Tesla system into a more mainstream position. It comes at a steep price as you need the Tesla approved GPUs, which are all Quadro cards. IMO, it's more a marketing stunt at the moment, but hopefully will grow to a lot more than that. Hopefully becoming a truly accessible performance product. It seems that since OpenCL is coming of age and dependencies on CUDA are starting to wane, that Maximus is going to be nVidia's next major performance push. None of that is bad, it's just nVidia's typical way of pushing the market forward. At some point in the future, CPUs and GPUs are going to work intimately together and tasks will be distributed across the types of processors that are most appropriate.
IMO, Premiere could see similar benefits if Adobe were to just access multiple GPUs via SLI. That would be a lot more enabling to a great many existing users.
11-16-2011, 08:13 AM
Having Maximus generally gives your machine the ability to direct graphics rendering actions (the actual render to a pixel which will be directly output to a display) towards a Quadro with higher priority, while compute code is pushed towards a Tesla with higher priority.
I am setting up a Quadro 6000 w/ a Tesla M2090 in a Quadro Plex 7000 chassis at the end of December and will be able to run some performance tests then using CS 5.5 and SCRATCH.
11-16-2011, 09:00 AM
FYI: you are correct, this is a duplicate thread with EXACT same title...
One thing to note about Maximus, all the benchmarks are against their 1 yr old
legacy Generation Quadro GPU lineup, which is about to change in the next 3-6 months with a
new Quadro based on upcoming GPU.
So, the more I think about it for the next 6 months , nVidia is trying to sell you a $2500 math coprocessor that's
about to be outdated very quickly by a new lineup of GPU with supposedly double to triple the performance anyway.
If you can get the payback now, it's cutting edge. If you have a low end quadro and don't want to pop for a Quadro 5000
or Quadro 6000, it makes sense because you can use it with a next Gen Quadro anyway.
AMD just released a pro Firepro $200 card that's 2X faster than the Quadro $200 card shown in the study.
Also, Intel is now releasing a math coprocessor with 50 cores that should boost Open CL performance
If you can get by for now and wait 3-6 months, I think the benchmarks then will look drastically different.