Andrew Richards
Forum Replies Created
-
[Lance Bachelder] “just would be nice to have some options when it comes time to BTO a new MacPro”
It would be nice, but I doubt Apple would have a BTO option for a card that would make its own software run poorly. I could see them selling a third party card in their store, but getting it built-in might be a non-starter for them if the OpenCL performance is indeed that poor on the Keplers.
Best,
Andy -
I saw a review of that new GTX 680 that made me worry about its immediate future as an Apple OEM card. Dig the last graph, the one that tests OpenCL performance. The 680 sucks at OpenGL compared to the new AMD 7000 series. Apple is an OpenCL partisan, so it seems unlikely to me they would ship a Mac Pro with a card that sucks at their favored technology.
That doesn’t stop NVIDIA from putting forth the effort to make a couple Kepler cards Mac-friendly, but if the status quo of Apple having to OEM cards for them to work on a Mac carries forward, it does not bode well for modern CUDA on the Mac.
Best,
Andy -
[Bobby Mosca] “Well, that Red Rocket demo that came up a little while ago seemed to work pretty well”
Hmm, that is really impressive. Makes me wonder about the Red Rocket’s specs. Why do they call for minimum PCIe 2.0 8x if it can be happy with half as much bandwidth via Thunderbolt?
GPUs, particularly the high-end ones that would be desirable for NLE co-processing, often saturate 8-16 lanes of PCIe 2.0 (at least when they are driving big displays along with doing traditional GPU-intensive tasks). Thunderbolt 1.0 is essentially bandwidth-equivalent to 4 lanes of PCIe 2.0. Maybe if they are not driving monitors and are only being used as GPGPUs it would be plenty of pipe, but it would probably depend on the card and the application.
Intel has said that Thunderbolt’s next rev will live atop PCIe 3.0 and could double the bandwidth. Then we might have something.
Best,
Andy -
[Bobby Mosca] “Anyway, if Mountain Lion will include external GPU support via Thunderbolt, we can attach whatever we want, upgrade at will, and that would be great!”
Thunderbolt wil need to get a lot faster for an outboard GPU to offer up all its available power. And by faster, I mean at least double or quadruple the speeds of Thunderbolt 1.0.
Thunderbolt is a great step up from FW800 for MacBooks and iMacs, but a replacement for PCIe slots, it ain’t. By a long shot.
Best,
Andy -
[Steve Connor] “The car crash that was the FCPX launch happened 9 months ago, why are you still spending your valuable time here and what are you getting from it?”
This board is the window on the Post world for me. Since the death of Final Cut Server (my bread and butter till 6/21), I’ve drifted away from Post in my work (shared storage) because without FCP legacy, shared storage for post is a very different beast. I’m still very interested in where this is all going, and I’ve still got a glimmer of hope that there is a shared/server capability waiting in the wings for FCPX. Though I’m skeptical that the large facility market will ever embrace it enough to make it viable to be in the FCPX SAN business the way it certainly was with legacy FCP.
Nine months ago I didn’t want to snap to judgement that Apple was bailing on broadcast, and I still think that from a technical perspective FCPX is today only one or two minor updates away from viability in that space (depending on the nature of the work and its flow, of course). But what is clear to me now is that the way FCPX works simply kills it for a huge swath of the folks I used to mingle with in my former niche. I’m sad that at this time last year we had an industry coalescing around a common NLE and now that same space is more balkanized than ever.
Best,
Andy -
[Steve Connor] “Photoshop CS6 Beta released, with magnetic timeline!”
You beat me to it!
But seriously, this is interesting stuff:
1. PS is now underpinned by Mercury Engine and therefore runs on CUDA juice. This is very disruptive in terms of Mac GPUs, way more than PPro is. There are a hell of a lot more PS users than there are PPro users, I’d wager. Particularly on the Mac. Did Apple know this was coming? Could this be a factor in the rumors of a return to NVIDIA for new Macs? This summer holds a lot of promise for new Macs with Ivy Bridge in MacBooks and iMacs and the new Xeons for a new Mac Pro. If we get exclusively AMD GPUs and no Mac Pro this year it will be bad mojo for Mac-centric creative pros well beyond the video realm.
2. Those in-line re-sampling selectors make me very happy, I wish I had them all along.
3. I’m curious to see how this new DRM/Licensing thing plays out. I know there was a great gnashing of teeth about the new licensing model back when it was first announced, particularly with regard to how upgrade pricing would be treated (though I haven’t followed it since to know if it has been adjusted).
4. Some of the new features in that video strike me as feature creep. I know why video editing is there, photogs dipping toes in the video waters with their ubiquitous DSLRs, but it looks awkward at first glance. Same goes for the vector and text editing stuff. Isn’t video, vector, and text layout what Premiere Pro, Illustrator, and InDesign are for? I guess this is for the Photoshop user who needs to work with vectors and text and video on the periphery but wouldn’t be doing dedicated video, vector, and text layout work. I dunno. Mixed feelings.
5. 64-bit only! So the Macs that can’t run Lion also can’t run CS6. There are also features on the Windows side that won’t work in XP at all. Progress!
Best,
Andy -
-
From the article:
“A dual-socket Mac Pro based on the new chips with 16 processors could potentially outperform iMacs and store more data.”
…ya think?
Best,
Andy -
[Craig Seeman] “The ATI 1900 supported OpenCL but the ATI 2600 did not? “
The X1900 never supported OpenCL. My guess is in keeping with the sloppy launch of FCPX, that card was omitted from whatever plist tells the App Store what is and isn’t supported hardware. That it was corrected in 10.0.1 underscores that idea.
That’s the only thing that restricts App Store distribution, not the app itself and its dependencies, but a check of the host system prior to install. You can install FCP7 on a G5 if you modify the scripts in the install pkg. You’ll find parts of it not working properly, same as you would with FCPX on hardware that does not support OpenCL.
[Craig Seeman] “Apparently the Intel HD 3000 isn’t OpenCL either but ti works.
OpenCL-capable graphics card or Intel HD Graphics 3000 or later
(bold emphasis added by me).”OpenCL can be used in addressing both GPUs and CPUs. From what little I could find out there, OpenCL on OS X can address the CPU if it supports AVX, which debuted on Sandy Bridge, home of the HD 3000. My guess (again) is that the engineers told the marketing guys “just tell them they need an HD 3000 or Open CL GPU”.
[Craig Seeman] “There is a difference when a piece of hardware can’t run something vs a piece of hardware Apple feels no longer meets their idea of acceptable results and they, block the installation. Apple shouldn’t have to “support” what they don’t want to, but if something works good enough for the end user, they shouldn’t block the install. I know there are are FCPX users who managed to get it on unsupported systems by copying over the install from another system (thanks to self contained apps) and they’re OK with the modest performance they’re getting. “
This is the App Store protecting users from themselves. You’re saying Apple should sell software to customers who don’t have the hardware to properly run he software, and then just shrug and say caveat emptor when those customers deluge them with support calls? The whole point of the App Store is to take the guesswork out of buying apps. It shouldn’t sell you something you aren’t equipped to run.
[Craig Seeman] “Engineer may decide what’s qualified but blocking the user who may be OK with working on an “unqualified” but functional system is coming from another place.”
Damned if they do, damned if they don’t.
Best,
Andy -
I saw the same rumors. Supposedly NVIDIA was on the hook for new MacBook GPUs but has failed to deliver with their new Kepler chips. Ivy Bridge’s built-in GPU is supposed to be much improved, but better than a discrete GPU? Hard to believe. NVIDIA is struggling to get its Kepler line to market (it was due this quarter) and the reports I’ve read point to poor manufacturing yields. AMD has been shipping their Southern Islands GPUs for weeks now.
Best,
Andy