GPU question

Hi there,

I have a question connected with the way Fusion deals with the hardware, which seems not to be like Resolve :

Resolve sees one GUI gpu (amd) connected to my 32" screeen and one GPU (nvidia gtx) for processing, whereas Fusion shows it sees both of them (to be checked) BUT seems not to have the gtx work at any time, which is a big problem as you can understand...

BTW, is Fusion OpenCL only ? What if the gtx was connected to the screen ?

Sorry if the info stands in the manual, I admit I didn't read it entirely so far

Can't check both. Only one.

Ok Chad, thx for your answer...
Then I should probably try connecting the gtx to the screen to test it, but I still wonder wether Fusion takes CUDA into account or not (=OpenCL only)

Fusion doesn't use CUDA, but Fusion plugins may. Because of OSX support, it's unlikely base package will ever use CUDA.

Ok, I appreciate this specific answer, thank you.
It's a pity it doesn't use CUDA as Resolve does, because OSX works very well with CUDA (and PC GPUs), since 4-5 years now... (therefore I didn't understand your "Because of OSX support")

I think the real problem is the difference between Fusion and Resolve, especially since the last NAB where the roundtrip between Resolve and Fusion was announced ; for ex. if you have - in Resolve - a cheap or middle range AMD GPU for GUI only and a strong GTX 980 ti or TitanX to compute and you decide to take advantage of Fusion Connect to roundtrip on a good project, then you come from 3000 CUDA cores strength (in playback, computing nodes, caching others, etc) to... nothing... except a cheap GPU working alone...

You can admit your workflow is seriously "imbalanced" (if I can say so), and it's not good for it. What d'you think ?

Ok, I appreciate this specific answer, thank you.
It's a pity it doesn't use CUDA as Resolve does, because OSX works very well with CUDA (and PC GPUs), since 4-5 years now... (therefore I didn't understand your "Because of OSX support")

It's been a few years since Apple sold a computer that could run CUDA. Why would you add support in a future product for something that's years out of manufacture? OSX is a decent OS, but Apple makes horrible computers. Supporting them necessarily involves significant compromises.

I think the real problem is the difference between Fusion and Resolve, especially since the last NAB where the roundtrip between Resolve and Fusion was announced ; for ex. if you have - in Resolve - a cheap or middle range AMD GPU for GUI only and a strong GTX 980 ti or TitanX to compute and you decide to take advantage of Fusion Connect to roundtrip on a good project, then you come from 3000 CUDA cores strength (in playback, computing nodes, caching others, etc) to... nothing... except a cheap GPU working alone...

You can admit your workflow is seriously "imbalanced" (if I can say so), and it's not good for it. What d'you think ?

Why would you expect an advanced OpenGL application like Fusion go run well on a cheap GPU? Just because Resolve doesn't support advanced OpenGL features doesn't mean that Fusion wouldn't.

Hey Chad,

I don't care about Apple and its computers, its politics and strategy...
I'm indeed very satisfied of OSX, which comes with support of nVidia GPUs by default since X.7 Lion (the worst osx ever btw), and up to now...
Recent iMacs (gtx780) can run CUDA, macpro legacies also do, hackintoshes do, even thrashcan macpro will - if someone support the form factor - because OSX does and nvidia updates its drivers and cuda for mac...
This said, what's at stake is not Apple official computers, for users have been stuffing PC graphic cards in them for a while now, but basically that there's no specific mac support to add : just work with CUDA too and that's all

Why would you expect an advanced OpenGL application like Fusion go run well on a cheap GPU

I don't expect anything like that, I said : Resolve is ok for OpenCL and CUDA > now roundtrip is supported between Resolve and Fusion > Fusion may be ok for both of them too.
Basic logic at the end of the day...
The "cheap GPU" example was only to recall that Resolve understands GUI GPU + Compute GPU, Fusion doesn't but should now, especially because BMD is the same puppet master for both of them.

Just because Resolve doesn't support advanced OpenGL features doesn't mean that Fusion wouldn't.

No. Just because Fusion supports advanced OpenGL features doesn't mean it couldn't also work with CUDA.

But you can choose which GPU you want fusion to use for OpenCL stuff. Just go to the preferences

Fusion doesn't support CUDA because Apple doesn't support CUDA.

Fusion supports OpenCL. So does Resolve.

Fusion supports separate GPU's for OpenGL and OpenCL. So does Resolve.

So why does CUDA matter?

Recent iMacs (gtx780) can run CUDA, macpro legacies also do, hackintoshes do, even thrashcan macpro will - if someone support the form factor

Basically affirming that Apple doesn't support CUDA. Right now OpenCL is the only option for cross platform development unless you're willing to nix Apple. Considering that Apple hasn't announced support for Vulkan, I'm all for that, but some customers would rather stick with OpenCL and (old) OpenGL to maintain Apple support.

That doesn't affirm that CUDA isn't supported, in fact the opposite, he's saying even the latest Mac Pro will support them if you could get them in the form factor (or if you plug them in via external TB boxes)

You can put in a modern GTX 900 series into a an older Mac Pro with the latest OSX and it will still run perfectly fine as long as you update the NVidia drivers to the latest just like you would with anything else. OSX does ship with drivers, but there are newer and better ones that have been released since the last update to OSX. This doesn't mean it's not supported. People are also using them in external boxes connected via thunderbolt on newer systems to boost rendering in Resolve. If CUDA wasn't supported then you wouldn't have CUDA acceleration in Resolve 12 or in other programs such as Premiere CC etc. All it means is that at the moment Apple are using AMD and the built in Intel GPU's, but it doesn't mean Nvidia cards and CUDA aren't supported. I wouldn't be surprised if it at some point it swings back to Nvidia cards in the future

Fusion doesn't support CUDA because Apple doesn't support CUDA.

Chad, I "came in peace" and answered , but please READ...
Apple supports CUDA, can you understand that ?
And it's been supporting it for years now, followed bu nvidia which regularly updates drivers + CUDA, which result into 2 separate Preference System panel aso... Everything is really official and very well supported. Even if you don't use latest nvidia drivers, you could use Apple ones because OSX ships with'em. I would'nt be astonished if Apple get back to nvidia btw
So excuse me but your first words are super false

Fusion supports OpenCL. So does Resolve.

Correction : "Fusion supports OpenCL. So does Resolve, which also supports CUDA", or "Fusion supports OpenCL. Resolve supports both"

Fusion supports separate GPU's for OpenGL and OpenCL. So does Resolve.

Correction : Fusion supports separate GPU's for OpenGL and OpenCL. So does Resolve, in addition with CUDA ones if necessary, whatever the card is used for (gui or compute).

So why does CUDA matter?

It matters because : perf are better on benchmark tests so far + users who have been used to CUDA wouldn't change anything + some other pieces of software require CUDA (it's not my fault if nvidia leads the market) + larger support is always better and avoid bothering satisfied users + BMD itself recommends CUDA on Resolve, then may also recommend it for its fellow software.

Basically affirming that Apple doesn't support CUDA.

Yep, and basically wrong.

It's a shame that fusion doesn't support CUDA. Apparently it was put in in version 6 before BM took it over, but then taken back out in Version 7.

Oups, didn't read Adam before rushing into response... It looks like he's confirming my words btw.

Edit : About your second post above, I admit I didn't follow the technical evolution of Fusion up to now, Adam. Anything I wrote here comes from the recent NAB fact : the Connect Link aka the roundtrip between Resolve & Fusion is extremely important, and for me increases the need to match the minimum requirements and config needed to run one and the other.
Plus, I'm sure that Resolve + Fusion is a huge FCPX + Motion killer, and I'm pretty sure BMD knows it too. This may count.

It's a shame that fusion doesn't support CUDA. Apparently it was put in in version 6 before BM took it over, but then taken back out in Version 7.

That's incorrect.

Basically affirming that Apple doesn't support CUDA.

Yep, and basically wrong.

I'm sorry, you can say whatever you want about Apple having LEGACY support, but the fact is you cannot walk into an Apple Store and buy a new Mac with CUDA support. You're basically asking BMD to spend development resources to add support for something that moving forward fewer and fewer OSX users will have access to. Let's say it takes 1 year to add CUDA support. By that time every Mac that does support CUDA will be even 1 more year obsolete. More and more users will be moving to non-CUDA Mac's (or they'll be ditching OSX entirely).

OpenCL is supported by Apple for both legacy AND current machines. It's even supported on machines with Intel GPU's.

If you really want to see CUDA support added, you'll have to come up with a business case other than "It will help you sell software to people with 4 year old workstations who are interested in the slight performance increase from using CUDA over OpenCL but are not interested in the enormous performance increase from using a machine that's newer than 4 years old."

There simply isn't enough reason to move from OpenCL to CUDA even if customers could use it because the performance difference is small. There's plenty of other places where performance could be improved that wasn't vendor specific. Plugin developers are free to use CUDA, but few are, for the same reasons.

Your current configuration of having one GPU for OpenGL and one for OpenCL is perfectly valid right now. It doesn't afford OpenCL-OpenGL interop, but that's not a huge deal and it doesn't even work currently. CUDA support won't change that one way or the other. Would it be nice? Sure. Would it drain resources from other things, like getting OpenCL working better? Sure.

Which part? Maybe it wasn't 6 but rather 6.4, but there were a lot of press release in 2012 to say that the rendering side was now running on CUDA

The eyeon Software Fusion 3D renderer is now running on CUDA,

All current MACS have current CUDA support, not just legacy, if it was just legacy support then they wouldn't have added in support for the GTX 900 series in the latest builds of OSX as none of the Macs that came with Nvidia card came with 900 series cards. Legacy support implies only supporting older cards which is not true.

If you really want to see CUDA support added, you'll have to come up with a business case other than "It will help you sell software to people with 4 year old workstations who are interested in the slight performance increase from using CUDA over OpenCL but are not interested in the enormous performance increase from using a machine that's newer than 4 years old."

Not sure where the 'enormous performance increase' is from. On the Mac side people have been upgrading their older Macs with dual X5675 CPU's which when put together make the CPU power equal and in some cases faster than the single CPU in the current Mac Pro with the ability to upgrade to more reliable and faster GPU's than the ones in the current Mac.

On the Windows PC side you can have much more powerful CPU's, but not on the Mac until they release a newer model with something that isn't 3 years out of date

Chad, please, I begged for you to read before, I'm trying again now, but it's beginning to be weird
Or maybe you think I'm telling you ******** ? Come on, what's the point of that, we like Fusion and that's all.

It's not a problem (at all !) that you don't fell comfortable with Apple computers, especially because I'm no fanboy, I don't care about whatever connected to praising what one doesn't manage, I'm not talking about Apple anyway, and I think I know Apple, which I don't rely on.
I just admit I deeply know OSX, ok ok, and I appreciate it, ok ok... You read it ? The OS ! Not the company, not the hardware, not the iProducts or whatever : the OS(X)
So now, I will write as I would tell my 13 years old daughter, JUST because you don't wanna understand (on purpose), just because you don't READ what's written above.
Well, let's only start with your own 1st assumption, because as before, your misguided approach led you to a false conclusion, so that the following is incorrect too :

" you can say whatever you want about Apple having LEGACY support, but the fact is you cannot walk into an Apple Store and buy a new Mac with CUDA support."

SWEAR YOU WILL FOCUS ON THESE FOLLOWING LINES THIS TIME : Any Mac, since OSX Lion (end of year 2010), ANY computer running OSX has current support of CUDA. You got that ?
Now watch that :

01.png (172.73 KiB) Viewed 5473 times

Hehehehe, you knew I was french, now see above a french Pref Sys panel of OSX 10.11.4 (current OSX). French or not, you see the CUDA panel ?
This is what you get when you click on it :

cuda.png (89.93 KiB) Viewed 5473 times

And here what you have when you click on the drivers one:

GD01.png (34.71 KiB) Viewed 5473 times

YOU SEE THE SECOND LINE (not checked because I opt for nVidia ones) ? OSX has been officially supporting CUDA for years now, and consequently current one does too Therefore this is what you get when ... "you walk into an Apple store" aso aso... You just go to nvidia website and download https://www.nvidia.com/object/mac-driver-archive.html , regularly updated, and this page can be interesting for those going further
A last click on the same panel to certify you my post of the day : I'm up to date
[Oh no, I'm not allowed to cos' 3 attachments are the max. - we don't care, it was just showing drivers up to date on this end of April 2016 hehehehehe].

Anyway Chad, OSX has been supporting CUDA for years, even if you weren't aware of that ****, and Fusion should support Cuda too, at least because its partner Resolve does. This are my 2 cts on it, and I wish you a good week-end.

Chad, please, I begged for you to read before, I'm trying again now, but it's beginning to be weird
Or maybe you think I'm telling you ******** ? Come on, what's the point of that, we like Fusion and that's all.

Maybe there's a misunderstanding, yes. I'm not saying OSX doesn't support CUDA. It does. I'm saying Mac hardware doesn't support it. Just because the OS supports it for legacy reasons doesn't mean the hardware can do anything. The Mac Pro has proprietary graphics that don't support CUDA. Same with the MBP, MB, MBA, iMac, Mac Mini, etc.. You could add an external GPU via Thunderbolt, but Mac's don't support Thunderbolt 3. So it's only in the most extreme corner cases that it becomes practical to consider CUDA for contemporary Macs.

Yes, Fusion COULD support CUDA, but why? Fusion COULD support FPGA's and DSP's. The issue is that of a business case. Adding support for something that only a subset of your users can benefit from which is redundant with features that ALL of your users already can benefit from seems like a waste of resources. If Fusion was otherwise perfect software and a huge cash cow? Sure. But there's just so many things that need attention before adding redundant features.

Which part? Maybe it wasn't 6 but rather 6.4, but there were a lot of press release in 2012 to say that the rendering side was now running on CUDA

The eyeon Software Fusion 3D renderer is now running on CUDA,

Never was. Marketing-speak, nothing more.

Nvidia was, for a while, cross branding EVERYTHING as CUDA. Compute shaders? That's CUDA for DirectX. OpenCL? That's CUDA for Khronos. They even say their GPU's contain CUDA cores. Which taken to the extreme means your UEFI UI was running CUDA.

Not sure where the 'enormous performance increase' is from. On the Mac side people have been upgrading their older Macs with dual X5675 CPU's which when put together make the CPU power equal and in some cases faster than the single CPU in the current Mac Pro with the ability to upgrade to more reliable and faster GPU's than the ones in the current Mac.

On the Windows PC side you can have much more powerful CPU's, but not on the Mac until they release a newer model with something that isn't 3 years out of date

True. It's speaks more to how poor the Mac Pro is that it's trounced by an EOL CPU. The fact that a $2000 PC can beat any Mac new or old in every benchmark is also an indication of what's wrong with Apple's professional workstation lines. It's really a shame how far things have gone since they dropped "Computer" from their name. Moore's Law only applies if you take advantage of innovation.

Chad, please, I begged for you to read before, I'm trying again now, but it's beginning to be weird
Or maybe you think I'm telling you ******** ? Come on, what's the point of that, we like Fusion and that's all.

Maybe there's a misunderstanding, yes. I'm not saying OSX doesn't support CUDA. It does. I'm saying Mac hardware doesn't support it. Just because the OS supports it for legacy reasons doesn't mean the hardware can do anything. The Mac Pro has proprietary graphics that don't support CUDA. Same with the MBP, MB, MBA, iMac, Mac Mini, etc.. You could add an external GPU via Thunderbolt, but Mac's don't support Thunderbolt 3. So it's only in the most extreme corner cases that it becomes practical to consider CUDA for contemporary Macs.

A mac with an nvidia GPU "can do anything" that CUDA requires it to do. That's all. I don't understand the reason why you insist.
The MacPro thrashcan has D700 AMD GPUs, of course it doesn't support CUDA.
I talked about any mac in which you can put a nvidia card : 2008-2012 MacPro "caddie", iMacs, MacMini. You put an nvidia card (a PC one if you like), you install the 2 pref panels to work with last updates of nvidia and go...
You talk about the 2013 MacPro which has default AMD Firepro GPUs installed, I don't.

Yes, Fusion COULD support CUDA, but why? Fusion COULD support FPGA's and DSP's.

Because any pro software can nowadays AND because Resolve - that I now consider as the official bro' software of Fusion, especially since the official roundtrip support - can too.
There's no point in being stuck in a particular point of view that keeps Fusion aside the natural evolution.
BMD owns Resolve and Fusion, the first takes anything into account, the second should too : this way, this technical part is behind and doesn't deserve any more talks, BMD can focus on what's more interesting. That's all, simple and efficient.

BTW, I already told you but why not saying it again : nobody gives a **** about what you think of Apple, what your specific analysis is or the idea you feed of potential benchmarks with a 2K$ PC aso (which is false btw)... More than being false, it's pointless and off-topic.

Because any pro software can nowadays AND because Resolve - that I now consider as the official bro' software of Fusion, especially since the official roundtrip support - can too.
There's no point in being stuck in a particular point of view that keeps Fusion aside the natural evolution.
BMD owns Resolve and Fusion, the first takes anything into account, the second should too : this way, this technical part is behind and doesn't deserve any more talks, BMD can focus on what's more interesting. That's all, simple and efficient.

Why does Resolve matter? You seem to make that argument but ignore the fact that Resolve never added CUDA support on top of OpenCL. They did it the other way around. First supporting CUDA, then adding OpenCL support because of Apple. They had a business case, add OpenCL or lose OSX (and potentially Intel) users. The reverse case, add CUDA or lose users, has never existed for Resolve and it doesn't exist now for Fusion. There is no "natural evolution" towards CUDA.

Why "should" Fusion use CUDA? Not because it "can", tell me why it should. What's the business logic? CUDA support will cost development time and money and increase support load. The only reason Resolve supports CUDA is because it originally supported CUDA. The work was already done on that years ago. Adding CUDA support to Fusion isn't a checkbox in the compiler. To make a business case you need to come up with a way for CUDA to increase sales or you need to come up with a list of features that can be deferred to allow time for CUDA development or you need to come up with a means to increase the selling price of the software because of CUDA. I don't think you can do any of those three.

You can't put the technical part "behind" and focus on more "interesting" things if the technical part eats up resources that could have been used for interesting things. Make a list of the 100 things you would most like Fusion 9 to have in order of importance and see where "CUDA support" ends up. I don't think you would put it in the top 80, right?

BTW, I already told you but why not saying it again : nobody gives a **** about what you think of Apple, what your specific analysis is or the idea you feed of potential benchmarks with a 2K$ PC aso (which is false btw)... More than being false, it's pointless and off-topic.

You're 100% correct. Your claims about the ease and utility of CUDA support spoke for themselves and didn't need to be diminished with discussions about the inferiority of Apple's current hardware lineup. Furthermore, I was incorrect about the pricing, I was comparing a "stock" Mac Pro, which is >3x the price of a significantly faster $1300 PC. A maximally configured custom Mac Pro requires a PC that is nearly half it's price to outpace it across the board. My mistake.

Why "should" Fusion use CUDA? Not because it "can", tell me why it should. What's the business logic?

It "should" support cuda because the "natural evolution" I talk about is to support anything that can be supported, especially in the fields of hardware acceleration, GPGPU and performance.
I don't consider the dvpt of cuda support as a loss of time which may be dedicated to other features : I think this may be a parallel work, so that it doesn't eats up resources needed for Fusion new features.
Of course, the list of your urgent new features suggestion won't take in account what's under the curtain because one would focus on specific actions inside Fusion that are experienced during work, but I don't agree with your point of view about what may have effects on the users motivation or what could be used to increase selling prices. Too much short term point of view.

Resolve matters because in the global strategy, users see that BMD listens to them and provide close tools in a common workflow, and Fusion will more and more belong to this workflow. Therefore if users see consistence in the developments of these 2 softwares, BMD will take a huge advantage of it, including sales, % and statistics of pro (or not) usage of the duo in the post-prod world. This is far more valuable than a narrow minded strategy that doesn't spread good signs (as Apple did with the suicide of numbers of software btw). Resolve allows one GPU to compute in Lite version ? All right, this is enough to increase sales...
There's no point in arguing about OpenCL before or after CUDA, as long as users can feel the same strategy - of supporting both - in Resolve and Fusion.
This could be one of the bigger asset of BMD in the field of grabbing market share : 2 softwares working close and wide spread all over the world is a smarter way of grabbing users than your considerations about what Fusion and Resolve "separate" roadmaps.

I don't consider the dvpt of cuda support as a loss of time which may be dedicated to other features : I think this may be a parallel work, so that it doesn't eats up resources needed for Fusion new features.

You think the developers can add CUDA by just installing keyboards under their feet? It's not as trivial as you make it out to be. Whatever time is spent working on CUDA (or any new feature, really) is time not spent on something else.

I don't agree with your point of view about what may have effects on the users motivation or what could be used to increase selling prices. Too much short term point of view.

It's not short term, though. As a business, BMD wants to sell as many products as they can for as high of margins as they can. That's something that will be BMD's motivation over the long term. Real choices need to be made about how much investment is made and how that can be turned into revenue. BMD has made some bold choices already, investing two years of development time in order to expand the potential customer base. There's business logic behind that, though. If you want to see the addition of a meaningless feature that will add cost to the product development you have to provide some rationale for that. You can't just say it is "natural" because business doesn't work that way. Businesses don't do what's natural, they do what makes money.

Resolve allows one GPU to compute in Lite version ? All right, this is enough to increase sales...

I'm not sure what you mean by that? You're suggesting that Fusion offer different tiers of GPU support for the lite and paid versions? How would that affect OpenCL vs CUDA?

There's no point in arguing about OpenCL before or after CUDA, as long as users can feel the same strategy - of supporting both - in Resolve and Fusion.
This could be one of the bigger asset of BMD in the field of grabbing market share : 2 softwares working close and wide spread all over the world is a smarter way of grabbing users than your considerations about what Fusion and Resolve "separate" roadmaps.

What kind of "feel" are the users supposed to get? Does the cooling fan whine at a different frequency when running CUDA vs OpenCL? How would users get any sensation at all about whether one language is supported over another? If they users can't tell the difference while the computer is running, then what are they getting the feeling from? Marketing? BMD could say with a straight face that Fusion supports CUDA right now, because technically it does. Technically every frame interactively rendered on a workstation with an Nvidia card running the GUI is done with CUDA. It's just marketing speak, of course, but users would feel better about how Fusion works? I guess that's a question for BMD's marketing people to decide if they want to make that claim or not.

If BMD want users to feel like Resolve and Fusion have the same roadmap, then they should do something that would benefit users in appreciable ways. Add fuscript to Resolve. Extend the Fusion SDK to include Resolve. Provide working LUT interchange. Allow settings to move between the applications' common tools. Offer context sensitive help in Resolve. There's a ton of things that would represent taking the most useful parts of each application and using that to make a better, more productive experience. CUDA doesn't fit that type at all, much less in a way that demands higher priority.

sorry to interupt you two, but there is no point in this discussion.

we heard now that some users think cuda will be good, other dont.

we know now that apple sells currently no computer which supports cuda, only 5-6 year old ones does

i believe openCL is fast enough for the task you have in Fusion.
for me it would be greater if we see more tools using the GPU, rather then adding a third code base to the software

and as a total offtopic: i think in 1-2 generations everything apples sells will be ARM based and they make a big jump to iOS everywhere. look at the iPad Pro...

Just for context, my company published the first CUDA plugin for Fusion back in... 2008? This was before Fusion had OpenCL. At the time OpenCL was introduced to Fusion it was not as capable as CUDA, but it had the benefit of being hardware agnostic, not just GPU agnostic, but could be supported by CPU's and APU's and FPGA's and all sorts of neat things, even ARM SoC's. Not all of that proved to be practical, most users just wanted GPU's, but it's a cool idea. OpenCL got better, so Fusion's support of it got a bit easier too, as there was less jumping through hoops. At one point you could cache the output of the tools in the GPU memory, so you could view them in OpenGL without a download and upload. That was a neat performance boost that would be really nice to get going again if more OpenCL tools existed and supported it. With OpenCL you can also run kernels over multiple devices, though Fusion doesn't support this yet. That would also be really cool as it would allow you to build some really powerful workstations.

But the biggest thing is that there has to be more tools that actually use this. While my main desire is for better support for OpenCL, at some point the actual tools need to be made, especially tools that are already existing in Fusion which are unlikely to be replaced by 3rd parties. Fusion is a tool box, not a box of tools, but it's just not practical to expect the existing tools to update independently.

Totally relevant, I admit

Ok, I schedulded to stop anyway, M. Obvious killed me (my motivation at least).
Actually I'm a fan of OpenCL and have ever been (yes yes), even if you didn't have the pleasure to read it here, and anyway very often (always ?) fond of what's not closed, especially as a developer for different applications...
These few lines as a potential cuda fanboy gave me fresh air , no matter if a guy thought I'm stupid. I'm still deeply convinced of what I tried to share with you, but there's indeed no point in arguing for nothing...
I like Fusion, its tools, its nodal flow, and the quicker way it allows you to achieve your ideas, and that's the real point.

Oh so what if i have only one gpu say a gtz 870 or quadro m1000m and i swithc from rsolve to fusion, doethe card automaticaly change from cuda to cl?

Richard