Since the MacBidet 970 production thread has started talking about blades, it seems appropriate to give this thread a shot of reanimator fluid. Or worcestershire sauce.
I'm totally uniformed about the virtues of the different proposed interconnect standards, but I'll stick with the theory that Apple is producing blade-like mobos for 970 desktops that will scale via RapidIO interconnects.
I'm not overly familiar with the idea of what a blade or cluster really is. most of the stuff I've red seems to assume you know quite a bit about it already. I have 20 G4s and we run premiere, after effects, poser etc on them (we'd also like to run maya soon). the biggest problem we have is rendering. many of our students come up with great ideas and are fully capable of realising them but have to cut back because of the time it takes to render. in a two hour lesson for example its ridiculous when you add some colour filter to 20mins of video and find that to see it in premiere takes over an hour to render. similarly, if we move to maya (we use MAX at the mo). I'm afraid that slow rendering speeds will kill off any real work.
what I'd like to be able to do is:
say a student is sat at a machine and is using premiere. if he needs to render something then he hits a button and it gets sent straight to a blade/cluster/server? which renders it out and sends it back to him (ideally, it would render for him right then and there). then if he jumps to maya/lightwave etc and needs to render a still to take a look at some caustics for example he hits render and instead of taking 6 minutes just to render the still it takes seconds via a cluster/blade/server?.
does anyone know if this is at all possible? realistic? or have I got the wrong end of the stick altogether?
I'm totally uniformed about the virtues of the different proposed interconnect standards, but I'll stick with the theory that Apple is producing blade-like mobos for 970 desktops that will scale via RapidIO interconnects.
Thanks for the bump boy_analog.
I think the basic thought I had when I started this topic is that a future Powermac by whatever name (BladeRunner?) would be a chassis that accepts blades in the same way that blade servers do, in fact the same module to be used in servers with a 3 or 4 Unit rack chassis holding a dozen or more blades.
say a student is sat at a machine and is using premiere. if he needs to render something then he hits a button and it gets sent straight to a blade/cluster/server? which renders it out and sends it back to him (ideally, it would render for him right then and there). then if he jumps to maya/lightwave etc and needs to render a still to take a look at some caustics for example he hits render and instead of taking 6 minutes just to render the still it takes seconds via a cluster/blade/server?.
does anyone know if this is at all possible? realistic? or have I got the wrong end of the stick altogether?
It's very, very realistic -- it's Pixar's operation in so many words. They've got to have the power to render an entire movie, not just effects. I'm sure you could get that set up for your students, but the obvious object would be money. One blade server alone is terrifically expensive (but awe-inspiringly powerful).
Good luck -- I understand your and your students' frustration. Hopefully this summer's crop of Power Macs will alleviate some of that.
Pixar's move to Intel for it's renderfarms was determined to meet it's current needs. You can bet that Steve wants to eat his own dogfood in this space.
The only thing that would dislodge all those Intel based blades would be IBM 970 based blades. If they come from Apple, it would be a great showcase for Steve's "other" company.
Comments
Since the MacBidet 970 production thread
I'm totally uniformed about the virtues of the different proposed interconnect standards, but I'll stick with the theory that Apple is producing blade-like mobos for 970 desktops that will scale via RapidIO interconnects.
Tell me I'm wrong!
I'm not overly familiar with the idea of what a blade or cluster really is. most of the stuff I've red seems to assume you know quite a bit about it already. I have 20 G4s and we run premiere, after effects, poser etc on them (we'd also like to run maya soon). the biggest problem we have is rendering. many of our students come up with great ideas and are fully capable of realising them but have to cut back because of the time it takes to render. in a two hour lesson for example its ridiculous when you add some colour filter to 20mins of video and find that to see it in premiere takes over an hour to render. similarly, if we move to maya (we use MAX at the mo). I'm afraid that slow rendering speeds will kill off any real work.
what I'd like to be able to do is:
say a student is sat at a machine and is using premiere. if he needs to render something then he hits a button and it gets sent straight to a blade/cluster/server? which renders it out and sends it back to him (ideally, it would render for him right then and there). then if he jumps to maya/lightwave etc and needs to render a still to take a look at some caustics for example he hits render and instead of taking 6 minutes just to render the still it takes seconds via a cluster/blade/server?.
does anyone know if this is at all possible? realistic? or have I got the wrong end of the stick altogether?
Originally posted by boy_analog
Bump!
I'm totally uniformed about the virtues of the different proposed interconnect standards, but I'll stick with the theory that Apple is producing blade-like mobos for 970 desktops that will scale via RapidIO interconnects.
Thanks for the bump boy_analog.
I think the basic thought I had when I started this topic is that a future Powermac by whatever name (BladeRunner?) would be a chassis that accepts blades in the same way that blade servers do, in fact the same module to be used in servers with a 3 or 4 Unit rack chassis holding a dozen or more blades.
...
Originally posted by spooky
what I'd like to be able to do is:
say a student is sat at a machine and is using premiere. if he needs to render something then he hits a button and it gets sent straight to a blade/cluster/server? which renders it out and sends it back to him (ideally, it would render for him right then and there). then if he jumps to maya/lightwave etc and needs to render a still to take a look at some caustics for example he hits render and instead of taking 6 minutes just to render the still it takes seconds via a cluster/blade/server?.
does anyone know if this is at all possible? realistic? or have I got the wrong end of the stick altogether?
It's very, very realistic -- it's Pixar's operation in so many words. They've got to have the power to render an entire movie, not just effects. I'm sure you could get that set up for your students, but the obvious object would be money. One blade server alone is terrifically expensive (but awe-inspiringly powerful).
Good luck -- I understand your and your students' frustration. Hopefully this summer's crop of Power Macs will alleviate some of that.
Originally posted by fred_lj
-- it's Pixar's operation in so many words. --
Pixar's move to Intel for it's renderfarms was determined to meet it's current needs. You can bet that Steve wants to eat his own dogfood in this space.
The only thing that would dislodge all those Intel based blades would be IBM 970 based blades. If they come from Apple, it would be a great showcase for Steve's "other" company.
...