Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Buying Mac Pro today

Shaun Cammack said:
Well, if these guys have anything to do with it...

http://www.ebay.com/itm/Nvidia-Tesla-C2050-for-2008-later-Mac-Pro-Resolve-Smoke-AutoCad-3-GB-4000-CUDA-/230734561860?pt=PCC_Video_TV_Cards&hash=item35b8da0244#ht_1235wt_1310
Click to expand...

If you were into unofficial workarounds, GTX 580 would be the best option. I don't know how reliable these are though. In any case, buying a Tesla card when you have only 2 full bandwidth slots available is quite meaningless. Very useful when you need a bunch of CUDA GPUs just for processing.
 
Jon Jones said:
Actually the Tesla card and one RED ROCKET will go into a cubix expansion bay.
Click to expand...

As I mentioned above, you are splitting valuable bandwidth. Luckily, Rocket is not very bandwidth hungry. But the Tesla most certainly is. In any case, since you are dabbling into unofficial and unsupported products, why not just go for GTX 580? It will be faster than any single GPU Tesla.
 
GTX 580 huh? IS there any support for that card at all? Third party or anything?

Subhadip Sen said:
As I mentioned above, you are splitting valuable bandwidth. Luckily, Rocket is not very bandwidth hungry. But the Tesla most certainly is. In any case, since you are dabbling into unofficial and unsupported products, why not just go for GTX 580? It will be faster than any single GPU Tesla.
Click to expand...
 
GTX580 is a waste of time in a Mac. The power and double slot space requirements necessitate it be in a PCIe expander like the Cubix. But even with that, the nVidia Mac drivers won't allow you to make full use of all the CUDA cores, even with the best EFI mods and config changes. Just a limitation of the drivers themselves. If you want to get full use out of a GTX580, you'll have to boot into Windows.

Kinda feel the same way about Tesla. One [expensive] card for Tesla support doesn't provide huge gains in most situations, even less so on the Mac as the support is not as complete or far reaching in terms of apps that can use it. There are (IMO), much better ways to spend that money. Or better yet, don't spend it at all.

Pains me to say it, but the Quadro 4000 is still about the best choice we have in OSX. Some cards work alright after an EFI flash and tweaking some config files for Resolve -- the GTX460 is a good one for this, great as a CUDA GPU for resolve. Worthless as a primary GPU in OSX. ...If you don't mind the double slot width of the card again.


Rumor has it that we will be getting a more open graphics interface in OSX Mountain Lion with increased support or ability to use more off the shelf cards. There are already hooks in some developer builds, and have been for months, showing direct mention of upcoming ATI and nVIDA GPUs. Hopefully this is the case, it would be great if we could buy a GTX 670 or 680 and drop it in the system at some point. I'm not getting my hopes up, though... These sorts of rumors have reared their head before, only to not happen.
 
Yeah maybe I'll just wait to get the upgrade cards, or maybe instead of more GPU's, I can just get 3 more RED ROCKETS, drop them into the Cubix. I mean I only edit R3d's so I mean, that should work for super fast transcoding.

Jeff Kilgroe said:
GTX580 is a waste of time in a Mac. The power and double slot space requirements necessitate it be in a PCIe expander like the Cubix. But even with that, the nVidia Mac drivers won't allow you to make full use of all the CUDA cores, even with the best EFI mods and config changes. Just a limitation of the drivers themselves. If you want to get full use out of a GTX580, you'll have to boot into Windows.

Kinda feel the same way about Tesla. One [expensive] card for Tesla support doesn't provide huge gains in most situations, even less so on the Mac as the support is not as complete or far reaching in terms of apps that can use it. There are (IMO), much better ways to spend that money. Or better yet, don't spend it at all.

Pains me to say it, but the Quadro 4000 is still about the best choice we have in OSX. Some cards work alright after an EFI flash and tweaking some config files for Resolve -- the GTX460 is a good one for this, great as a CUDA GPU for resolve. Worthless as a primary GPU in OSX. ...If you don't mind the double slot width of the card again.


Rumor has it that we will be getting a more open graphics interface in OSX Mountain Lion with increased support or ability to use more off the shelf cards. There are already hooks in some developer builds, and have been for months, showing direct mention of upcoming ATI and nVIDA GPUs. Hopefully this is the case, it would be great if we could buy a GTX 670 or 680 and drop it in the system at some point. I'm not getting my hopes up, though... These sorts of rumors have reared their head before, only to not happen.
Click to expand...
 
But how fast do you really need to transcode? And to what end or purpose? On a hopped-up 12-core system with fast RAID storage, you can transcode to 4K ProRes 24fps at real-time with one rocket. Going faster than that is possible to some extent, but the bottleneck is actually not the rocket card, it's the codec and how it scales [or rather, how it doesn't] across all those CPU cores. Most all transcodes on fast systems these days with Rocket cards are "codec-locked" -- you are at the mercy of the slowest point in the chain and it's stuff that doesn't use multiple processors well (or at all) and/or stuff that resides off of the Rocket card. ProRes is just one example. On the other end of that argument are formats like DPX sequences. DPX is simple and fast to write, so if you're cranking out DPX sequences, then you will scream along a lot faster with multiple rocket cards. Depends on what you need.

My suggestion would be to start simple and practical. One RED Rocket. One good GUI GPU and if you run Resolve, a secondary compatible GPU for that as well. Research and upgrade from there. You may not need to load up on all this hardware like you think you want to. Some of it you definitely won't want to -- the Tesla card being one such thing. Unless you specifically need it. There is nothing in our industry that currently needs it. In fact, I only know of one app that kinda uses it and it won't accelerate R3D anything. That App, Premiere, may not even provide any Tesla support on Mac. I've only seen it demoed and talked about on Windows PCs.

You can get away with 4 Rockets in a Cubix 4-slot GPU XPander box. There's adequate bandwidth there for all of them... I'm just not sure you will get the return out of such a configuration to justify the steep price tag.

Best thing to do is to start simple like I said above. Identify bottlenecks in your workflow(s) and then see what can be done to address them.
 
One more thought on multiple rockets... I would much rather equip 3 separate systems with a rocket rather than dumping them all into one system. Network rendering/transcoding FTW.
 
Whatever floats your boat... I've run systems with two rockets and have gone back to a single rocket per system for now. Like I said above, the rocket is not usually the bottleneck adding more rockets doesn't necessarily make things go faster. Two rockets are handy within Adobe Premiere as it helps chew through more streams of R3D. However for general work and editing this isn't a big advantage due to how the wavelets work. The rocket isn't as efficient as decoding on the CPU on the fly for lower resolutions. Rocket always does a full decompression and debayer . So at 5K, it's going to max out at about 30fps... However, a good 6-core i7 can play back and edit a few streams of 5K at 1/4 or 1/8 resolution just fine without a rocket. There are a lot of factors to influence how all things work and a lot depends on the amount of footage you're dealing with, your workflow, editing preferences, transcoding needs, etc.. So, like I said, best to start simple and see where the bottlenecks are, then address them. Throwing a bunch of Rockets into a system or two might let you decompress/debayer full resolution R3Ds as fast (or faster) than you want or need, but then what? The rocket doesn't handle transcodes, it just makes a full resolution and full quality image from your R3D file. It spits out full resolution frames or will scale to a lower resolution for output from the full image. After that, the rest of the system takes it from there.
 
That's cool, so I guess with 64GB of RAM, 32TB of RAID, and two quadro 4000's, the only bottleneck that I'm going to have is in the quadro's since their only 2 gig each. But really we've already discussed that their the best option for me as far as graphics goes on the MAC platform. So multiple rockets would be the next thing that would pump up the volume on decode speeds, don't you think? I also run Adobe Premiere, so in that aspect it would be awesome.
 
does adobe even take advantage of multiple rockets? it would be a huge boost for resolve users as well, and any kind of stereo workflow if you want something close to real time playback.


http://www.reduser.net/forum/showthread.php?71788-6-ROCKETS-and-REDCINE-X-PRO-v9


but this shows that 2 rockets on the highest end system can get you double the fps. the rocket is all about decompressing the r3d's. so the faster that can take place, the more time and resources your system will have making it into another compressed codec. 2x faster than real time for 24p is a pretty big difference. I think in a case of a single camera, on set dailies situation it may be overkill, but only having one system for a 2 camera situation, having an additional rocket in they system may be clutch. there are a lot of good reason's to go more than one rocket. just an expensive decision :)
 
You must log in or register to reply here.
Back
Top