Hi, i'm planing to move from my quadro fx 4800 to a new one or many new ones.

What i have to buy to get the best feedback on houdini 12?
One quadro 6000? 3 brand new Nvidia GTX 680?

What you guys can you recommend to me?

the new gtx 680 with 4 go of ram look pretty cool, but for approximately the same price of an “old” fermi quadro 4000, I dont know which is the best in H12…

Have fun,
Thomas

Hi thomas, thanks for your answer.

That's my problem actually, i'm spending my whole day on houdini 12 right now (half of the day that mean when i'm rendering on a project i'm working on houdini on my second workstation).

I don't know actually what is best for houdini but i know open CL do not work with Quadro “FX” series.
That's why i want to change but for what? I hope someone can help me on that!!

Nobody know what gpu or graphic card are the most interesting for houdini 12? Working with SLI? or not? Better with 3 cards or not? Quadro 6 better than new game cards?

SLI is not supported, nor is crossfire, for rendering. You can assign OpenCL sims to a second card, however, avoiding the problem where a large sim would swap out all of your textures and vertex arrays for GL rendering.

The new GEForce 680 is focused on graphics, so OpenCL computing has allegedly gone down in performance compared to a 580 (which are discounted a bit at the moment).

A Quadro 6000 and 5000 are very overpriced, unfortunately (if money is no object then they're very good performers). The Quadro 4000, 2000 and 600 are more reasonable, with decent amounts of VRAM (2,1 and 1GB). The Quadro 400 is not a good choice as it only has 512MB of VRAM, nor are any of the Quadro NVS cards.

You can use consumer cards as well, but they are not recommended for production. You would need to be very wary of updating drivers with consumer cards, as they can contain regressions for workstation apps like Houdini.

Thanks for your answer now it's clear for me i will get a quadro 6000.

I find this benchmark here [tomshardware.com].



As you say twod the gtx680 is not a good choice …
Unfortunately, I did'nt find the same benchmark for workstation graphics card.

Thomas

edit :if someone has a radeon hd 7970, you can give us his impressions on it with houdini 12 ?

Those review are interesting regarding quadro and ati:

http://hothardware.com/printarticle.aspx?articleid=1540 [hothardware.com]

http://www.cgchannel.com/2011/10/review-professional-gpus-nvidia-vs-amd-2011/ [cgchannel.com]

if you are looking at the last review:
on furyball you can see the only quadro card more insteresting than ati is the quadro 6000. This is interesting to know if you are interested in the opencl capabilities of those cards.

The 680 comes with 4GB of RAM though as an option, doesn't it?

To me that's more important than the raw speed if you're doing OpenCL (fluid sim) stuff…. 2GB isn't enough to do anything beyond some previews IMO.

Cheers,

Peter B

No the 680 4gb are not yet released (as i know) and will be release about a month or less.

That's a good thing to know that 2gb is not enough.

I have a Quadro 4000 with 2GB of VRAM, and I ran a brief test with OpenCL and was only able to sim up to about 220^3 voxels, which is not enough to really show off the GPU's power. So I agree, RAM is really important for simulation on the GPU.

It's rumoured that “Big Kepler” is due in August. The GEForce 680 is based off the same architecture as the GEForce 560, but 4x the shaders with simpler hardware schedulers. The successor to the 580 is a much larger die, and will likely power the next gen Quadros and Teslas - especially given that SIGgraph is in August, and Nvidia traditionally launches the Quadros around that time. It will be interesting to see how much memory those pack, as the 6GB, 2.5GB and 2GB of the Quadro 6000, 5000 and 4000 were very large upon their release.

What would be more ideal a quadro 4000 plus Tesla c2075 or just one single quadro 6000?

I tested a Tesla C2075 last week against a quadro 4000 in a Dell T7500 system. I used the optimized scene jeff put online for this in this forum.
When increasing the voxel-resolution i came to a point where the quadro 4000 just ran out of memory pretty fast and the Tesla just continued but up untill that point i have to say that i was not overly impressed by the tesla's speed.
True, it simmed a bit faster than the quadro but it was nowhere near as impressive as I was hoping for. I personally wouldn't fork out the money unless money is no issue at all.

As far as Ati goes, i would definitely stay away from the radeon boards. I've seen nothing but trouble with newer catalyst drivers resulting in extremely low viewport performance and glitches in the recent past. (crazy like “selecting a vertice taking over 30 seconds to refresh the screen” kind of crazy)

Hmmm….I received a fairly good deal, so I went with the quadro 4000 + tesla c2075 setup. It actually ended up being almost a $1000 dollars cheaper than one quadro 6000 and both were brand new. I kind of figured the tesla wouldn't blow through sims, but I was expecting to hear great reports about it. Nevertheless, this will be my setup. Hopefully I find ways to optimize scenes to really get the most out of the card. I'm also curious as to what role nvidia maximus will play.

I agree with RudiNieuwenhuis post, I have ATI firepro v7800 card and it's totally useless, you can't select points without the clip plane jumpig all over or the screen freezing, and the fix for this glitch leaves you without textures.

It states on the ATI site that this is one of the few cards that doesn't support Houdini (find out AFTER I buy it) so best to check compatibility - it's not really their fault.

Some info about next-gen Kepler and Tesla:
http://www.anandtech.com/show/6140/nvidia-announces-keplerbased-quadro-k5000-secondgeneration-maximus [anandtech.com]