Pr CS5 - List of supported CUDA Cards

[Curt Wrigley]

Adobe is working on a playback and rendering engine for Adobe Premiere Pro called the Mercury Playback Engine. This new engine is NVIDIA® GPU-accelerated, 64-bit native, and architected for the future. Native 64-bit support enables you to work more fluidly on HD and higher resolution projects, and GPU acceleration speeds effects processing and rendering.

The Mercury Playback Engine offers these benefits:

• Open projects faster, refine effects-rich HD and higher resolution sequences in real time, enjoy smooth scrubbing, and play back complex projects without rendering.
• See results instantly when applying multiple color corrections and effects across many video layers.
• Work in real time on complex timelines and long-form projects with thousands of clips — whether your project is SD, HD, 2K, 4K, or beyond.

Ensure your system is ready to take advantage of the Mercury Playback Engine in a future version of Adobe Premiere Pro. The Mercury Playback Engine works hand-in-hand with NVIDIA® CUDA™ technology to give you amazingly fluid, real-time performance. See it in action

* PR CS5 supports the following list of CUDA cards:

	GeForce GTX 285	Windows and MAC
	Quadro FX 3800	Windows
	Quadro FX 4800	Windows and MAC
	Quadro FX 5800	Windows
	Quadro CX	Windows

More hardware details:

http://www.adobe.com/products/premiere/systemreqs/

[cts51911]

I'm returning the GTX 260 I purchased.  Not even going to buy a 285 or wish for a quadro.

Can't wait for those reviews to come rolling in.

[Guest]

Jeff Bellune wrote:

nothing (including this thread) has answered to anyone's satisfaction the question of whether all CUDA-compatible nVidia GPU model numbers, OTHER THAN those few listed, have been "flicked off" from Mercury's playback engine to tap in for acceleration

The question has been answered.  The reality is that you don't like the answer.  Current cards that support CUDA acceleration in Premiere Pro CS5 are:

--Begin List--

• GeForce GTX 285 (Windows and Mac OS)
• Quadro FX 3800 (Windows)
• Quadro FX 4800 (Windows and Mac OS)
• Quadro FX 5800 (Windows)
• Quadro CX

--End of list--

What a typically arrogant answer.  Sort of like answering the question, "Will my Ferrari run on standard unleaded gasoline," and answering, "Ferraris are supported to run on premium unleaded gasoline," while consequently mocking the question.  Thanks for nothing.

Anyone else, without an M.B.A.-style aversion tactic?  The specific question is whether GPU model numbers, even those that, e.g., exceed or approach the GTX 285 by a margin of near-insignificance, are functionally ignored for GPU acceleration by the Mercury playback engine, even though the operative word is CUDA which is a scalable architecture.  Again, the examples:  TMPGEnc and vReveal engage any nVidia card with the CUDA architecture to use GPU processing for acceleration of rendering.

If the answer is yes, Adobe has a minor scandal on its hands.  And instead of spitting back arrogant answers, be prepared for the "blogosphere" to erupt with courtroom-style impeachment:  "What were you thinking?"

[cts51911]

""What a typically arrogant answer.  Sort of like answering the question,  "Will my Ferrari run on standard unleaded gasoline," and answering,  "Ferraris are supported to run on premium unleaded gasoline," while  consequently mocking the question.  Thanks for nothing.""

Well put.

I've gotten the same smug answers here before.  If you lack the credentials here (<2,000 posts), there isn't much room for discussion.  We are crashing their exclusive club.

Fact, CS5 is going to give wannabe fimakers (me) some amazing tools.

First HD DSLR...Now real-time NLE...Amazing!

Imagine the crap you'll see in HD!  HAHA

[Stan Jones]

I know I missed something; I'm not going to pay  much attention to this whole thinkg till the reviews are in.  My CS3 is working well enough...

The specific question is whether GPU model numbers, even those that, e.g., exceed or approach the GTX 285 by a margin of near-insignificance, are functionally ignored for GPU acceleration by the Mercury playback engine, even though the operative word is CUDA which is a scalable architecture.

When Dennis (or Wil?) was talking about this early on, I did not get the idea that they were turning the functionality on or off, but that they would only certify cards that they had tested.  Can someone point me to the place where they said that cards that functionally would work won't be allowed to?

Also, isn't part of the issue that it is not just CUDA, but additional functionality that is being provided?

[Guest]

Stan Jones wrote:

I know I missed something; I'm not going to pay  much attention to this whole thinkg till the reviews are in.  My CS3 is working well enough...

The specific question is whether GPU model numbers, even those that, e.g., exceed or approach the GTX 285 by a margin of near-insignificance, are functionally ignored for GPU acceleration by the Mercury playback engine, even though the operative word is CUDA which is a scalable architecture.

When Dennis (or Wil?) was talking about this early on, I did not get the idea that they were turning the functionality on or off, but that they would only certify cards that they had tested.  Can someone point me to the place where they said that cards that functionally would work won't be allowed to?

Also, isn't part of the issue that it is not just CUDA, but additional functionality that is being provided?

Further evidence that Adobe have created a P.R. train wreck; they (and their so-called "evangelists") are emphasizing the rote talking point that the Mercury Playback Engine speeds things up generally with respect to the CPU, in response to specific queries about whether the engine taps into the CUDA acceleration capabilities of the nVidia product line.  So far they are purposefully silent (as well as failing to explain why).  Until someone steps up, this is a joke (or a predatory marketing strategy involving collusion between Adobe and nVidia to stimulate hardware sales).

[Jeff Bellune]

If the answer is yes

The answer is yes.  You won't be able to turn on GPU acceleration for any unsupported card.  The MPE will stay in software rendering mode if the installed card is unsupported.

As for the question, "What were you thinking?", you'll have to ask the engineering team to get the real answer.  There are logical arguments to be made on either side of the issue.

-Jeff

[Guest]

Jeff Bellune wrote:

If the answer is yes

The answer is yes.  You won't be able to turn on GPU acceleration for any unsupported card.  The MPE will stay in software rendering mode if the installed card is unsupported.

As for the question, "What were you thinking?", you'll have to ask the engineering team to get the real answer.  There are logical arguments to be made on either side of the issue.

-Jeff

How clever to stay strategically vague in the face of a bizarre outcome; you want to stay on Adobe's "good side" for some reason.  Yet it is exceedingly strange and nakedly dumbed-down for Adobe to "tick-on" GPU acceleration for the GTX 285, but "tick-off" GPU acceleration for the only slightly better-performing GTX 295 (and notably, these cards and many others share the scalable CUDA architecture in common).  I doubt that you or anyone else could fathom more of a "logical argument to be made on either side of the issue," than the stimulus this will create for a sudden increase in sales of specific nVidia GPUs - even with the absurd result of a GTX 295 owner tossing that GPU to downgrade to a GTX 285, merely to fit within Adobe's bizarre marketing strategy.

Start looking for this scandal on the tech blogs as CS5 launches, and if you do have a direct line to Adobe, it would be a big favor for them if you suggest they line up damage control.

[chupacabracobra]

The reality is that they definitely did just "flick off" support

for other cards. How is the GTX 285 any different from say oh a GTX 275?

If Nvidia is playing tricks with certain CUDA commands then why not use OpenCL?

Anyway I can make it work with unsupported cards. The GPU doesn't seem to get used as much as I'd expect though. So maybe something is still a bit locked out, hard to say. It does help though (some things). Oddly it does absolutely zero for a basic preview of an unedited clip. The CPU can be going at 60% on all cores and the GPU is at 1%, kinda of weird. Other programs you would see the GPU going mad and the CPU at like 3%.

One thing that I find a little odd is that you can use a really low-end $100 card that has full h.264 support and playback a high-bitrate h.264 file in programs that support h.264 decoding hardware and they will barely even blip your CPU but CS5 still will drive multiple cores to each 50-80% usage even for a simple unedited clip playback. That said it does perform a lot better than CS4 once you add effects and transitions or basically do anything at all. Granted an editor may need to process in a different way than a simply playback program but it still seems like they are not really levering all that a card can offer, again even a simple unedited playback in the preview, not edit, window will really hit your CPU and other programs that use full HW acceleration for h.264 will barely cause a blip on the CPU performance views.

Anyway whatever the story, it does work better than CS4 no matter what graphics card you use (althoughit still seems like they may have left a good deal of performance on the table, but it's hard to truly say without knowing all the details of how the engine needs to run).

Jeff Bellune wrote:

nothing (including this thread) has answered to anyone's satisfaction the question of whether all CUDA-compatible nVidia GPU model numbers, OTHER THAN those few listed, have been "flicked off" from Mercury's playback engine to tap in for acceleration

The question has been answered.  The reality is that you don't like the answer.  Current cards that support CUDA acceleration in Premiere Pro CS5 are:

--Begin List--

• GeForce GTX 285 (Windows and Mac OS)
• Quadro FX 3800 (Windows)
• Quadro FX 4800 (Windows and Mac OS)
• Quadro FX 5800 (Windows)
• Quadro CX

--End of list--

[shooternz]

At the end of the day...I just want something that works as advertised and

the cost is not too big a deal.  Like one edit, easily pays for the stoopid card.

Thats the way I am looking at it and  I even think the "wow" factor this will have with my clients will be a business builder for me.

CS5 is a bigger investment than previously (new hardware, software and OS) but this is a breakthrough and it just needs to be done in my circumstance.  I consider it to be well worth while for the money the NLE Suites generate from such  relatively small $$$ compared to yesteryear.

[canuscanus]

I side with those who are a bit put out by the limited support of CUDA cards.  Many models with very similar components and specs to the 285 have been available for a long time, yet aren't supported.  And there is no explanation from Adobe about their rationale, or their plans to support more cards in particular product lines.  I was very excited to hand over my money for the upgrade immediately until this issue of not getting (great?) acceleration from "generic CUDA" cards surfaced.  Sadly - for Adobe and me - I think I have to wait several months before giving them my money.  Perhaps they didn't want to confuse the message during release, but I think they've achieved the exact opposite.

[Harm_Millaard]

Apart from the pressure from nVidia and the time constraints, the overriding concern was and is STABILITY. Every card certified must meet Adobe's stability tests.

If there has not been sufficient time to test other cards extensively to guarantee stability, they are not certified and thus not enabled. This is further complicated by all the driver changes that nVidia performs, which can cause instability with new driver versions.

[Ed Edited It Productions]

To date, we have been a big supporter of Adobe products, and I am looking forward to CS5. I'm planning on investing in new "stable" workstations, and looking forward to the MPE and everything such a solution offers. But, here are my questions:

1) The list of graphics cards that are supported with the inital CS5 release is clear. Thank you. But what is not clear is what additional existing cards (available today) will be supported in the future. If you're telling me "this is it forever until CS6", then I can go buy one of these cards based on what I can afford, and call it a day, and get the stability I require. But if you're suggesting to me that an existing cheaper alternative card available today *might* be supported in Q3 or Q4 after a patch or free software upgrade, then I'm going to feel ripped-off investing in something I might not really need.

2) I understand that it will take some time to release performance data based on supported cards. What I need to know is a timeframe. Like most businesses, I have to budget for hardware and software purchases, and I have limited windows in a fiscal timeframe to do so. I've heard "it's coming", then "it's here", and now it's "wait for the performance data". At some point I either have to make an Adobe purchase, or jump ship to a platform/hardware solution that just works TODAY. Otherwise, I'll continue to loose business as my legacy machines get outdated.

3) Let's assume I buy CS5 with a supported CUDA card. Does that give me some sort of warranty that I wouldn't get without a supported card? I hear about "stability" - are you guarantying this will work without crashing? Every editor wants maximum stability. If Adobe could point to a few machine configurations that could guarantee levels of featureset stability (relatively speaking) based on hardware investment, I would buy into one of them in a hearbeat.  I'm tired of software companies blaming hardware cards, computer configurations, and vise-versa. At the end of the day, it either works or it doesn't. My business can't be sustainable if I invest in something - then I have to wait for version CS5.1 (release date TBD) in order for things to work, ESPECIALLY if I've bought into the NVidida-Adobe 'supported' solution from the onset.  I would appreciate clarifcation on what "support" for "supported cards" really means.

I appreciate Adobe's continuing efforts to share what they can with us through these forums and other avenues. I hope everyone understands that under these economic times though, individuals and business are tight with the purse strings, and are willing to wait and pay for quality and stability, given the results are what is promised. Thanks again!

[Harm_Millaard]

Ed,

Welcome to the forums.

You have to realize that these forums are USER-to-USER FORUMS and Adobe presence here is limited, specially with NAB.

99.9% of what you read here are users posting, not Adobe.

The points you raise are valid, but the only one that can give you hard, tangible answers is Adobe. I suggest you contact Customer Service and try to get an answer from them. But realize that when developing programs or support for hardware, there are many intangibles in the development cycle, that may disrupt the intended RTM (release-to-market) date. Just consider the nVidia Fermi cards, the GTX-480/470 cards, which were intended to be released in November 2009, but only appeared in April 2010. Had nVidia met their deadline as announced last November, chances are that it would be included now in CS5 as a certified card, but nVidia missed their RTM date by 5 months. You can hardly blame Adobe for that delay, but of course it has repercussions on their own delivery of supporting the Fermi cards.

[Animation Guy]

I wonder what the least expensive card costs that would support Mercury in CS5.  Anyone have a rough guess ?

seems like it is going to cost a bundle.  Ouch.  I'm guessing $300 isn't going to somebody a certified card.

[Harm_Millaard]

David,

You have to make a distinction between software and hardware support of MPE. Only certified cards support hardware rendering. All other cards use MPE as well but only in software mode and do not use the MRQ and maximum bit-depth settings that certified cards do.

The least costly one is the GTX-285.

[cts51911]

I saw a workflow video on  the lowest end 13" Mac laptop/Core 2 and the the performance increase in "software only" is significant.  The preview looked much, much smoother than what I see in CS4 with an i7 PC.

Looks like we'll all benefit from 64-bit and Mercury.

[Guest]

Harm Millaard wrote:

Ed,

Welcome to the forums.

You have to realize that these forums are USER-to-USER FORUMS and Adobe presence here is limited, specially with NAB.

99.9% of what you read here are users posting, not Adobe.

The points you raise are valid, but the only one that can give you hard, tangible answers is Adobe. I suggest you contact Customer Service and try to get an answer from them. But realize that when developing programs or support for hardware, there are many intangibles in the development cycle, that may disrupt the intended RTM (release-to-market) date. Just consider the nVidia Fermi cards, the GTX-480/470 cards, which were intended to be released in November 2009, but only appeared in April 2010. Had nVidia met their deadline as announced last November, chances are that it would be included now in CS5 as a certified card, but nVidia missed their RTM date by 5 months. You can hardly blame Adobe for that delay, but of course it has repercussions on their own delivery of supporting the Fermi cards.

Harm, surely you do more harm than good by following in the steps of Adobe's "evangelists" who deflect the tough questions by playing into their wild cards, saying that the lack of time, resources or whatever to "test" the wide nVidia CUDA-compatible product line justifies disabling Mercury GPU acceleration completely despite the fact that CUDA is a standard, and scalable, and adjacent models in the product line such as the 285 and 295 are only marginally different.

You also tout the fact that the information here are from "mere users."  What exactly are you, other than a very very active participant with over 10,000 posts?

cts51911 wrote:

I saw a workflow video on  the lowest end 13" Mac laptop/Core 2 and the the performance increase in "software only" is significant.  The preview looked much, much smoother than what I see in CS4 with an i7 PC.

Looks like we'll all benefit from 64-bit and Mercury.

This logic means nothing to the issue.  It's like saying, I'm paralyzed and some researchers found a way to get one of my legs working!

[cts51911]

Dude...Chill-Out

Scratch that...Keep going!  Your posts are pure comedy.

....I do agree with your point though.

[Jeff Bellune]

zenviolence,

I got to talk some of the Premiere Pro team out in Las Vegas.  I was told that the logic behind limiting GPU-accelerated playback and rendering to a very specific set of cards was to guarantee a certain level of performance for anyone using Pr CS5 and one of the cards.  And yes, they seemed very aware that there will be an outcry from potential CS5 customers about why their favorite CUDA-capable card was excluded from the list.  Also consider the fact that only 2 cards are supported on the Mac.  So there's a whole group of Mac folks who will be even more unhappy.

FWIW, I'm disappointed that my GTX280 isn't supported.  I don't really want to pony up for a new video card.  Nobody does.

I tried to make my previous answers direct and to-the-point.  If you interpret that as arrogance, then I'm sorry.  The reality of the situation is (warning: direct and to-the-point answer coming): CS5's GPU-accelerated playback and rendering will *initially* only be available for certain nVidia graphics adapters.  That will undoubtedly change as time goes by.  Real Adobe Evangelists have said it's just a matter of time before new cards are supported.  It is your choice as to whether or not you upgrade to CS5, now or ever.  If non-accelerated playback and render performance in CS5 is a deal-breaker for you, then I recommend you hold on to your money until the list of supported cards fits your budget and your system.

BTW, I wish I was an "Evangelist".  They get paid!

-Jeff

[Harm_Millaard]

I don't know how long you have worked with Premiere Pro from what version, but up to now with every release, whether it was 2.0, CS3 or CS4, the forums were flooded with complaints about bugs (whether these were indeed bugs or OE is immaterial), that gave a negative impression of the product. Adobe has now chosen for a different approach, no more tons of new features, but stability and speed and as bug-free as possible.

With every choice you make, there are some who complain about it and there are others to applaud the choice. We are obviously opposites, I applaud the choice made, while you complain about the choice. However, you forget a number of things. It is not simply about CUDA. It is about the Mercury Playback Engine on a CUDA enabled card and not all cards are the same, even though they may have CUDA.

Your reasoning is comparable to saying: Here are ten cars that all run on gasoline. Install a turbo on all of them and they will all run faster. In theory that statement is correct, but you neglect the different gear boxes installed, the different suspension and brake systems, to name a few. Speed is of course important but for these cars, safety and durability is even more important. In this example engineers may decide only three out of ten cars are can be fitted with a turbo safely and the rest not. Well, you have to acccept that choice, whether you like it or not.

The same applies to PR CS5. The engineers deemed it unsafe to install a Turbo on all ten cards and detrimental to reliability, hence no hardware MPE support.

That is about the story.

[shooternz]

This debate is interesting at many levels but heres a slant on it

.

It is not compulsory to use the MPE  Technology.  It is a switcheable function and CS5 will run fine with out it.  Those that want to wait ..or have to wait...can simply wait it out for whatever reason.

CS5 will either offer something else (like 64bit) that will make or break the decision to upgrade for these users or they wont bother. Its a choice thing not only based on the card.

Those that need it will jump in first.

I am in and will have a brand new Quadro FX3800 Card in a box on my desk this afternoon.

[Harm_Millaard]

I'm leaning towards the GTX-480 which will give me a couple of months for the price to come down a bit and Adobe to come out with the first DOT release.

[cts51911]

Harm,

I with you.  The specs on the Fermi cards are impressive vs. the gt200 cards.  Also, considering what the comments have been from Adobe (although vague), they look to be 2x the performance.

I may just pull the trigger now and see what happens.  I've got other GPU enabled software and I hope to see a huge improvement in speed.

[joshtownsend]

I understand everyones POV on this Adobe's either made a deal with Nvidia to help them sell there most expensive cards by turning off MPE or there acting like Apple and assuming end users need some hand holding so they get the speed up they promised.

Cs4 didn't officially support windows xp 64 right? Because it was buggy on that OS....but they didn't prevent you from installing on it.

And honestly if they were worried about people not being able to replicate the speed they wouldn't use a 8 core 32 gig ram beat with a $4,000 dollar video card in all their demonstrations.

What would suck is if in two months from now they open support for some cheaper cards. That make these forums toxic for a bit.

But, then again if they weren't just trying to help Nvidia the most expensive cards then wouldn't they include the newest Fermi and a few more of the still expensive Quadro cards?

Maybe it isn't Adobe hype, maybe some cards the software isn't scalable like most other CUDA software. Maybe has something to do with the Adobe render engine not Cuda.

Or maybe they wanna sell more cards right now and won't announce until Fermi is out?

I know I'd like to try my nvidia 260 and see what happens because right now if I was able to turn on MPE. Just to see what would happen.

[shooternz]

I doubt this (MPE) will ever open up  to really cheap cards because I suspect that it needs the technology and specifics (eg DDR3 ram)  to make it work.

Cheaper cards ...definitely will evolve.

I am amused by how many (wannabees) consider this some kind of "business conspiracy" between Adobe and NVidia!

BTW: Easy for me to say as I now have choice of two cards that are supported waiting for install in a new system. (GTX285 & FX3800)

I am delighted that Adobe gave us the this heads up on hardware requiremnents by advance releasing the information.  Others can knock them but from me...CONGRATULATIONS on this and I thought it has been an inspired launch which will prove to be a serious entry to the Professional market.

I have always held the belief from my own experience with PPRO and FCP that it has been a serious  (and overlooked) contender since CS2 . My clients agree.