Greg Leuenberger
Forum Replies Created
-
Thanks for the honest update. I’ll just be upgrading the drives in my current system…we’ll see how NAB 2011 looks.
-G
Greg Leuenberger
CEO
Sabertooth Productions, Inc.
http://www.sabpro.com -
Hi guys, I’m writing this from my hotel – I’m at Siggraph this week. Take this with a grain of salt but at the NVidia station I noticed a Mac Pro running smoke with a little placard on the table announcing it was a Fermi based system (they’re rolling out Fermi Quadros down here). I asked the Nvidia guy and he said that there wasn’t actually a Fermi board in that Mac Pro (it had a Quadro 4800 in it) and that somebody had just placed the card there. When I pressed him about it he finally said that Fermi cards will be avail. on the Mac Pro before too long (I think 6-9 weeks or something)..I don’t believe the drivers are done yet. This would jive with the story the other day about somebody seeing Fermi drivers in an OS patch.
So I do think Fermi boards are on the way – they will be at least double the Cuda performance of the current OS X compatible Nvidia boards. So I would just wait a little – the Fermi Quadros are expensive but with up to 6GB of RAM onboard and a ton of Cuda cores the performace will be through the roof. The 12 cores will certainly help with rendering (I demo’d on a 12 core HP box at the ATI booth this morning showing some 3D stuff) – the extra cores will be great for multi-threaded apps like smoke and AE…as well as all the major 3D apps. So I think we’re in a good place – the 27″ LCD’s are nice as well for a main display (I use a Dreamcolor for reference)
best,
Greg
Greg Leuenberger
CEO
Sabertooth Productions, Inc.
http://www.sabpro.com -
I wouldn’t get too excited yet. Even if you could connect additional cards – unless there’s OS X drivers for Fermi based GForces or Teslas your Mac isn’t going to see them. The site doesn’t really explain how this works – I don’t think you’re connecting the cards directly to the machine so much as they are acting like a remote render station accessed via ethernet (hard to tell from the website). There’s other solutions for this.
I’m not sure BM will be all too enthused to support something like this either – since it’s similar to how their 100K+ Linux version works (via infiniband).
I’m also not sure why they’re advertising for a $99 rendering product that just got out of alpha phase and is nowhere close to production ready – sounds like they’re desperately searching for a market.
NVidia has rackmount GPU (Tesla based) render stations – they are great for HPC or rendering work where you are dealing with a frame at a time. I’m not sure you you can possible get real-time performance from color grading with a remote renderer unless you have a high-bandwidth connection like infiniband though.
I think we’d all be happy with Fermi support in an OS X update.
-Greg
Greg Leuenberger
CEO
Sabertooth Productions, Inc.
http://www.sabpro.com -
Hi – sincerely doubt DaVinci will be making use of more than 1 core. It gets it’s processing power from the GPU, so you would be much better off spending your money on the best (NVidia only) GPU you can get. We all need to cross our fingers and hope that there’s a Fermi based NVidia GPU on the way from Apple… and that we can put it in older machines. If not then BM better hire some gfx card driver coders so get the newest NVidia cards (or Teslas) working on OSX. If that doesn’t happen have fun working on 3rd generation hardware (this release would have been much better on Win. 7….).
I’m pretty sure Smoke is heavily multi-threaded…I haven’t had time to run the demo yet but it’s 64-bit only and from an answer I got on another forum it will take advantage of all your cores. 64-bit Apps and Multi-core rendering is coming – and it will be worth it… it’s just taking a while. The 3D apps (I own a production company but I’m a 3D animator by trade) are way ahead of everything else out there – both in terms of being 64-bit and massively multi-threaded. I can peg all 8 of my CPUs at 100% when rendering in modo or Maya. CS5 should help with this as well (CS4 is kind of a hack…but it works).
I suspect the next version of FCP will move us closer to leveraging our cores and 64-bit OS – if not then Apple has failed miserably to leverage it’s hardware and OSX with it’s most professional applications.
Utilizing CUDA like DaVinci does is very much cutting edge – eventually you’ll be able to stack two or three Teslas in a Mac and get ridiculous performance… just not yet (and like I said, fingers crossed on Fermi… or there’s going to be a lot of confused people…) It’s also possible (**possible**) that you could migrate the CUDA processing over to OpenCL – which is an open standard supported by Apple and will work on Nivida and ATI cards. I’ve heard OpenCL doesn’t have the performance of CUDA yet – and any coding of that nature is definitely non-trivial.
best,
Greg
Greg Leuenberger
CEO
Sabertooth Productions, Inc.
http://www.sabpro.com -
Jeremiah is doing what many of us will try and do – and that’s leverage our current investment in hardware to take advantage of the DaVinci software.
One of my systems is based around the Decklink Extreme HD – I recently purchased an HP Dreamcolor (for another project) that I would like to use in conjunction with the HDLink Displayport to monitor color (originally in After FX….now I will be one of the thousands of people buying DaVinci software). It seems to me that the ideal situation would be to either use one of the new Fermi based NVidia cards (assuming it will be coming out for OS X….not that Apple has ever had their sh*t together regarding graphics cards…but we can hope) or use your existing card (ATI 4870’s in my case) in conjunction with a single slot Tesla card for the CUDA calculations (again…here we are dealing with Apple who is woefully behind the times…but we can always hope…I think the 16xxx gen Tesla is single slot). Quite frankly this would have been a better release on Win7..but I won’t get into that.
Barring a Fermi based card or adding in a single slot Tesla then we’re limited to a two-slot last gen card….(thanks Apple..)
So…a GTX 470/480 + RAID Card + Decklink would work….a single slot Tesla card (if it exists) + the gfx card you already have + RAID card + Decklink may be a better solution. There’s a number of 3D rendering apps that take advantage of CUDA processing…when they’re cranking away the UI is basically unresponsive since all the card’s processing is being used by CUDA. I’m not sure how that translates to DaVinci though – so 2 cards may not necessarily be better.
I’m guessing we will be seeing some new and/or updated control surfaces rather quickly – I’m guessing people won’t want to pay a lot more that 2-3K for a control surface in a boutique shop.
Looking forward to seeing the software.
best,
Greg
Greg Leuenberger
CEO
Sabertooth Productions, Inc.
http://www.sabpro.com -
Dropping the need for a switch is a pretty big cost savings.
-Greg
Greg Leuenberger
CEO
Sabertooth Productions, Inc.
http://www.sabpro.com -
Great! Glad I held out on storage during 09′. Looking forward to what you guys put together, maybe I’ll swing by MacWorld.
-Greg
Greg Leuenberger
CEO
Sabertooth Productions, Inc.
http://www.sabpro.com -
I’ve harped on this before and it still makes ZERO sense to me. There’s nothing about rendering from After FX, Final Cut Studio or a number of 3D render stations that should require anywhere near the network bandwidth or drive array speed that capturing footage or playing back footage in real-time needs…even for modest codecs.
Rendering in After FX will only write a few frames a second at most (and most likely it will take a few seconds per frame instead)…that’s only a few megs per second even at 1080 uncompressed. Similarly a 3D station will save out one frame (a few megs) every few minutes. I simply have not seen a satisfactory answer why you can conceivably have 3 or 4 people reading and writing to an array+network capable of handling 200-400+ megs/second without issues and suddenly somebody needs to render out an After FX animation of a couple megs a second and everything is out of wack. Even rendering out of FCP you are still not rendering/writing in real-time…why can’t you render to the SAN?
Is it that the RAID can’t handle writing a small file like a single HD frame (or small bits of a larger quicktime) intermittently while also serving a constant flow of 34-40 megs/sec to two other places simultaneously?
There are people here saying they have no problem with it – and why should they? I don’t understand which systems can’t handle it and which can – what’s the defining factor? I get the feeling nobody here knows.
The fact is that people want to read and write to the same SAN weather it’s FCP, AE, a 3D render or an audio mix….copying over a PDF script…whatever! People want to organize ongoing productions in ONE place for simplicity, organizational and archiving purposes.
What is it about writing out a small amount of data per second or intermittently (single HD frames or small bits of a larger quicktime) that interrupts an otherwise capable system’s throughput?
-Greg
Greg Leuenberger
CEO
Sabertooth Productions, Inc.
http://www.sabpro.com -
Oh man….if it’s not too late send it back. I get the unsettling feeling that you’re in for a world of hurt..even if you manage to ‘get it working’. If your company just spent a bunch of dough on an HP system and the Art Director (no offense, this shouldn’t be your job) is over at Creative Cow trying to find answers instead of having HP personal in your office then you’re in trouble. People here aren’t even going to be familiar with this system…it’s’ not for Macs and it’s not for Production. Send it back while you still have a prayer to do so. That voice you hear in your head is your common sense saying ‘senditbacksenditbacksenditback’.
-Greg
Greg Leuenberger
CEO
Sabertooth Productions, Inc.
http://www.sabpro.com -
That makes sense, I read the CAT6 phys uses quite a bit more power..
I’d kind of like to see a pricing structure where you can unlock ports on the switch as you need them (like the Fiber switch vendors do). The vendors are striving for a lot of density but for boutique’s like mine 24 10GB ports are overkill…I could make due with half..and at $500/port it would be nice to have that option.
-Greg
Greg Leuenberger
CEO
Sabertooth Productions, Inc.
http://www.sabpro.com