strontiumDog

My Mac Pro 3,1 (2008) GeForce 8800 GT graphics card has just died. I didn't like it anyway. I have 2 x DVI-I (no MDP) 30" displays both of which must be supported simultaneously.

So, I need a new graphics card which supports the above displays. Given I have a 2008 3,1 MP and several hundred quid to spend (I'm in the UK) I'm thinking a pair of Radeon 5770 cards.

I am intending to do some hefty shader and GL processing so I need significant grunt. Given this, am I right in thinking that CUDA for jitter GL processing is a must? the nVidia cards don't seem to offer what I need unless I go to a Quadro card which will be significantly more money.

If max/jitter GL experts out there suggest CUDA is nice, but not necessary, the 2 x 5770 Radeons would probably be a better option given then I'd have separate frame buffers going through separate DVI-I connectors without the need for the mini display port converters (which I've heard horrible things about USB requirements etc). Having 2 separate graphics cards is OK with jitter isn't it?

My Mac Pro 3,1 (2008) GeForce 8800 GT graphics card has just died.  I didn't like it anyway.  I have 2 x DVI-I (no MDP) 30" displays both of which must be supported simultaneously.

So, I need a new graphics card which supports the above displays.  Given I have a 2008 3,1 MP and several hundred quid to spend (I'm in the UK) I'm thinking a pair of Radeon 5770 cards.  

I am intending to do some hefty shader and GL processing so I need significant grunt.  Given this, am I right in thinking that CUDA for jitter GL processing is a must?  the nVidia cards don't seem to offer what I need unless I go to a Quadro card which will be significantly more money.

If max/jitter GL experts out there suggest CUDA is nice, but not necessary, the 2 x 5770 Radeons would probably be a better option given then I'd have separate frame buffers going through separate DVI-I connectors without the need for the mini display port converters (which I've heard horrible things about USB requirements etc).  Having 2 separate graphics cards is OK with jitter isn't it? 

cuda-usage-vital-for-heavy-jitter-gl-work-choosing-new-gfx-card-for-mac-pro

I'd buy a single 5870 and not worry about the mini-Displayport connectors. I haven't had a problem with them, and the 5870 is a better card. Monoprice is a good source for cheap, quality adapters. You can buy two of their dual-link MDP-DVI adapter and run three displays if you like.

Not sure about the CUDA question. As I understand it the performance issue on the Mac is both the hardware and the quality of the card's drivers, and for the moment I think the ATI/AMD drivers are better.

I'd buy a single 5870 and not worry about the mini-Displayport connectors.  I haven't had a problem with them, and the 5870 is a better card.  Monoprice is a good source for cheap, quality adapters.  You can buy two of their dual-link MDP-DVI adapter and run three displays if you like.

Not sure about the CUDA question.  As I understand it the performance issue on the Mac is both the hardware and the quality of the card's drivers, and for the moment I think the ATI/AMD drivers are better.


Interesting - thanks for your input. Can I confirm you've got the same setup ie. a 3,1? I've had such mixed reports about having to use a 5870 usb port to provide extra power to get the res of mdp-connected displays to the maximum, and also then people getting unworkable interference on their displays; plus as I'm also running short of usb ports, it stayed my hand.

Anybody know about the CUDA advantages for jitter processing? I'm trying mildly to convince myself to justify a grand on a posh graphics card.

Interesting - thanks for your input. Can I confirm you've got the same setup ie. a 3,1? I've had such mixed reports about having to use a 5870 usb port to provide extra power to get the res of mdp-connected displays to the maximum, and also then people getting unworkable interference on their displays; plus as I'm also running short of usb ports, it stayed my hand. 

Anybody know about the CUDA advantages for jitter processing? I'm trying mildly to convince myself to justify a grand on a posh graphics card.


I am not sure CUDA will make it as an integral part of the jitter library. As far as i've seen jitter is a multi-platform app so it should run in either ati or nvidia based stations excluding any hardware specific parallel architecture. I think that if there is a gpgpu lang that might make it as part of the library i would be openCL. It would be amazing to see an official development supported by C74.

Of course you can still use CUDA and openGL but you must consider going for code( I've seen some xcode templates on internet).

I am not sure CUDA will make it as an integral part of the jitter library. As far as i've seen jitter is a multi-platform app so it should run in either ati or nvidia based stations excluding any hardware specific parallel architecture. I think that if there is a gpgpu lang that might make it as part of the library i would be openCL. It would be amazing to see an official development supported by C74.
Of course you can still use CUDA and openGL  but you must consider going for code( I've seen some xcode templates on internet).


I have a Mac Pro 5,1 which came with the 5870 built-in, so I did not go through an install. I would think the increased power of the card would be worth the trouble. What resolution are you running the displays at?

I have a Mac Pro 5,1 which came with the 5870 built-in, so I did not go through an install.  I would think the increased power of the card would be worth the trouble.  What resolution are you running the displays at?


I'm running two displays at 2560x1600 (30").

CUDA is just the name for nVidia's GPGPU SDK. AMD/ATI have what is called Stream. In any case, unless there's a really specific reason why you'd want to target one of their cards directly, OpenCL is the way forwards as it'll run on either (and in theory on Intel chips if they ever make the drivers).

However, if you're just looking to do shader work for OpenGL, then that's slightly different as GLSL is a different technology. But again either make of card is geared up to do this (the underlying architecture of modern cards do pretty much everything with shaders nowadays I believe).

CUDA is hardware dependent. The only app that i know that implements cuda directly is touchdesigner. They are very specific that you need Nvidia cards for it to work. OpenCL implements interfaces from both CUDA and DirectCompute, it can be used by both ATI and Nvidia cards.

Using a gpgpu language has a different usage if you compare it with a shading language. Think for example if you want to have a huge particle system with lots of members, to implement that on a shading language can be very tricky.

It would be nice to have a gpgpu implementation in jitter for sure.

CUDA is hardware dependent. The only app that i know that implements cuda directly is touchdesigner. They are very specific that you need Nvidia cards for it to work. OpenCL implements interfaces from both CUDA and DirectCompute, it can be used by both ATI and Nvidia cards.
Using a gpgpu language has a different usage if you compare it with a shading language. Think for example if you want to have a huge particle system with lots of members, to implement that on a shading language can be very tricky.
It would be nice to have a gpgpu implementation in jitter for sure.


CUDA usage vital for heavy Jitter GL work? Choosing new GFX card for Mac Pro