Subject: Software that works with Nvidia Maximus setups, best drivers for Maximus systems, Maya doesn't work with Maximus as advertised. Information before buying a Tesla
Read this before buying a Tesla K20, Tesla C2075 (to supplement your Quadro card) or an Nvidia Maximus workstation ( Quadro 6000 plus Tesla C2075 or Quadro K5000 plus Tesla K20)
I wanted to pass on some information I received from Nvidia today which may save you some time and money if you are thinking of buying a Tesla to boost the performance of your Quadro card or a new Maximus system.
The key point is that Maximus does not work for most of the commonly used 3D modeling and animation programs (like Maya), as advertised.Also, forget any information you got from a google search. There is a lot of false information out there. The information on the Nvidia website is not reliable and is misleading. For example, there is a Maya case study talking about how Maximus helped the company but the reality is that Maximus does not work for Maya at all. It almost looks like it is deliberately misleading.
I was especially surprised by this because many of the Maximus press releases, put out by Nvidia, specifically name Maya as an example of a program that will benefit from the technology. Today they told me that it only works on Maya with a plug in that the company wrote specially for the demo and is not available to the public. Google Nvidia Maximus and Maya and you'll see what I mean. There is plenty to make you think it does something that it doesn't.
Maya would be the obvious application for Maximus as the advertised benefits included being able to visualise simulations in real-time which requires heavy compute and displaying high quality imagery at the same time.
Nvidia said it works with applications like Adobe Aftereffects. That's great, I use that, but.... it works fine with a quadro 6000 alone and doesn't need a Tesla too. Simulations in Maya however, brings the viewport to a painful 0.2 fps and would have really benefited from a Maximus approach. Extra speed in After effects in no way justifies the cost of a Tesla - $3000+ for the Keplar generation.
There is no reliable list of applications that work with Maximus on Nvidia.com. As I found out, A case study showing the benefits does not mean your application works with the technology. If you are considering the investment in a Maximus platform, I would contact Nvidia directly to ask if it will work with your software before you buy. The new Tesla K20 is a considerable investment. Avoid the disappointment of finding out that you may as well have bought a brick when it doesn't work for any of the programs you use or only works for programs that never needed it.
I thought the Maximus drivers directed compute that that the Quadro would have handled to the Tesla using additional tools outside the application. I didn't know that it is 100% reliant on 3rd party software companies that don't seem to all be jumping at once. My guess is that the number of Maximus platforms out there is too small to warrant most software companies spending money and resources on.
I take my hat off to the head of marketing and PR at Nvidia. They had me believing the hype for a while. I really believed it would improve Maya's performance in simulations like they said. Great job. I really fell for it.... The reality is that Maximus is a pointless waste of time and money. I was excited when My Tesla C2075 arrived but performance was no different to the Quadro 6000 alone and very little if anything benefited from the presence of my new friend.
It's all about software support in this market and 85% plus use Quadro cards so AMD and Maximus platforms don't get a look in, despite their technical merits. The message seems to be; Great technology but nobody cares. The Fire Pro W9000 is technically superior but at $3000, the $1800 Quadro K5000 beats it across the board for this reason.
Also, In the interest of saving somebody else the time I wasted, the driver Nvidia mentions in the Maximus System Builders Guide actually doesn't work for a Maximus platform (why would it???). Nvidia told me to use: driver version 311.15. This information is not on their site for some reason. Not all drivers work for Maximus (most don't) and it took me 4-5 days of going back and forth with questions to find that out. My simple question about which is the best driver for a Maximus setup for Maya was not as simple as I thought. Most Autodesk products don't work with Tesla's.
Again, check before you buy. There is a lot of inaccurate information out there on the best spec for Maya and other similar 3D modeling programs. Plenty of people suggest a Tesla should be added to the mix in high end platforms without knowing that there is a very high chance that you will never use a program that can utilise the Tesla in any way.
Worse still. People are advising that it is better to buy a cheaper Quadro like a 2000 to pair with a Tesla as if you get the same performance for less money. Most pro users are better advised to buy the best GPU they can afford and spend the extra on better CPUs or duel CPU setups etc. That $3000 can do at lot for you if you don't waste it on a Tesla. You've been warned! For most software, one GPU. Best you can afford.
I have asked the guy from Nvidia to double check all this for me because I want it to be wrong. It's especially strange to me as Nvidia's strategy with the Keplar Quadro cards of selling them without a strong compute capability almost makes the Tesla K20 a necessity if high levels of compute is a requirement. However good viewport performance is with the K5000, it seems like I'm still better off with a Ferni generation card as a good all rounder when you take into account the fact that most software only uses one card at best.
The same is true BTW for SLI setups. Very few pro apps can use Quadro SLI and it only works on a small number of Motherboards. A geforce SLI motherboard will not work for Quadro SLI. Also, only the Quadro 5000 and up can SLI (you can't SLI 2 Quadro 4000's). There is similarly little info on the Nvidia site about which software benefits from SLI although there is a small list of motherboards that are certified for Quadro SLI. I don't know if it is up to date. It's like trying to get blood out of a stone.
It seems like Nvidia has every incentive to be vague about all this. I would be too if I invested tons of money developing and marketing a new technology that nobody can use. The less we know the better. We'll all find out after we buy it...
I'm mad at Nvidia for misleading me but more disappointed with the software companies that stop us taking advantage of the new technology. Hardware development may as well stop for a while as it seems to make no difference to performance in the real world. It's great that I have 4 times as many Cuda Cores. Shame that doesn't help me in any way.
Right, I'm off to sell my Tesla C2075 before anyone reads this. Talk about a hot potato....
If anyone has any experience of getting Maximus to work with their software, I would love to hear about that experience too.