Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

dalupus

macrumors regular
Original poster
Jul 19, 2011
132
0
I do a lot of work in CUDA and the design of the nMac Pro is absolutely perfect for CUDA development since 1 card is dedicated to processing. Unfortunately none of the cards offered for the system have CUDA support.

Any chance Apple will offer an nVidia card in the future? I find it odd that they went down the AMD route as they use nVidia as the discrete cards in their rMBP products.

Yes I know I could use OpenCL but I find it much more difficult to use and debug.
 
I do a lot of work in CUDA and the design of the nMac Pro is absolutely perfect for CUDA development since 1 card is dedicated to processing. Unfortunately none of the cards offered for the system have CUDA support.

Any chance Apple will offer an nVidia card in the future? I find it odd that they went down the AMD route as they use nVidia as the discrete cards in their rMBP products.

Yes I know I could use OpenCL but I find it much more difficult to use and debug.

I'm guessing that for Apple is not a matter of nVidia vs AMD (as they already use GPUs from both, as you stated), but more of a CUDA vs OpenCL. They want to push the latter (and since that makes sense only on a workstation machine), hence the AMD-only GPUs in nMP.

On the other hand, I consider the nMP as a huge bet from Apple's side, including a vast amount of risk; they might win or lose it. So it is very early to predict how the nMP GPU options will evolve, at that point.
 
Remember when iPad didn't play flash and everyone lost their ****? :)

Indeed. And Macbook Air got a lot of negative criticism when it was initially released. Both iPad and MBA are now two of Apple's most successful products. On the other hand, though, there's Xserve... :p
 
I think you got it all wrong.

The question should be targeted towards whoever makes the software you use and asking them when they will support OpenCL ;)

But Apple have been alternating between nVIDIA and ATi/AMD since the early days, so next years model might have it.
 
Depends really on the deal ATI/AMD have with Apple regarding how many models past the 6,1 they will offer the GPU's for. As the 7,1 may have a lot of changes with the Haswell EP Xeon's there's probably a good chance of the GPU boards being different in the next model so probably unlikely Nvidia will make any cards for the 6,1.

If it stays with AMD for the next gen and Adobe pushing in the OpenCL direction then Nvidia has to improve it's OpenCL support. Though probably it's very much in Nvidia's interests to take the hit, get the 7,1 GPU deal and swing the pendulum back to equilibrium.
 
I doubt it will be anytime soon for a few reasons:

1. Nvidia won't drop their GPU price to AMD level, they are not bothered. Nvidia basically gave the whole next gen console market to AMD cause the customer (Sony, Microsoft, Nintendo) demand is deemed unrealistic. And since Apple offer dual GPUs in all MacPro, using Nvidia card would lead to a price increase, they could offer it as an upgrade option but I doubt it as Apple custom design the cards in nMP.

2. Apple is pushing OpenCL to be adopted cause it's an open format vs Nvidia proprietary CUDA and going for CUDA would mean giving Nvidia more monopoly which Apple as a company wouldn't want (look at Foxconn-Pegatron, Samsung-TSMC).

3. AMD OpenCL performs much better then Nvidia, Nvidia looks really pathetic when it comes to OpenCL (probably they focus on CUDA only), and this makes AMD the better option for Apple emphasis in OpenCL applications.

4. The current FCPX improved performance is mainly cause of OpenCL, if Apple was to use Nvidia card, you'll not see the performance boost that FCPX is enjoying now. Of course on the other side, Premiere would see a performance boost with Nvidia then AMD but then, Premiere is FCPX competitor and obviously Apple wouldn't want to show that it performs faster then FCPX. On the bright side, Adobe promises to improve OpenCL support in the coming updates, so it's a win win situation for everyone (okay, maybe not for Nvidia :p)
 
I think you got it all wrong.

The question should be targeted towards whoever makes the software you use and asking them when they will support OpenCL ;)

But Apple have been alternating between nVIDIA and ATi/AMD since the early days, so next years model might have it.

The original poster made it clear that they develop their own software and find CUDA much easier to use than OpenCL.

For commercial development going the OpenCL route might make more sense, but for software that is part of a bigger research programme say, and might not be run very much before evolving, what ever is easier for the writer makes more sense.
 
I doubt it will be anytime soon for a few reasons:

1. Nvidia won't drop their GPU price to AMD level, they are not bothered. Nvidia basically gave the whole next gen console market to AMD cause the customer (Sony, Microsoft, Nintendo) demand is deemed unrealistic. And since Apple offer dual GPUs in all MacPro, using Nvidia card would lead to a price increase, they could offer it as an upgrade option but I doubt it as Apple custom design the cards in nMP.

2. Apple is pushing OpenCL to be adopted cause it's an open format vs Nvidia proprietary CUDA and going for CUDA would mean giving Nvidia more monopoly which Apple as a company wouldn't want (look at Foxconn-Pegatron, Samsung-TSMC).

3. AMD OpenCL performs much better then Nvidia, Nvidia looks really pathetic when it comes to OpenCL (probably they focus on CUDA only), and this makes AMD the better option for Apple emphasis in OpenCL applications.

4. The current FCPX improved performance is mainly cause of OpenCL, if Apple was to use Nvidia card, you'll not see the performance boost that FCPX is enjoying now. Of course on the other side, Premiere would see a performance boost with Nvidia then AMD but then, Premiere is FCPX competitor and obviously Apple wouldn't want to show that it performs faster then FCPX. On the bright side, Adobe promises to improve OpenCL support in the coming updates, so it's a win win situation for everyone (okay, maybe not for Nvidia :p)

This makes a lot of sense. Hard not to agree with this, even if you don't like it.
 
I doubt it will be anytime soon for a few reasons:

1. Nvidia won't drop their GPU price to AMD level, they are not bothered. Nvidia basically gave the whole next gen console market to AMD cause the customer (Sony, Microsoft, Nintendo) demand is deemed unrealistic. And since Apple offer dual GPUs in all MacPro, using Nvidia card would lead to a price increase, they could offer it as an upgrade option but I doubt it as Apple custom design the cards in nMP.

2. Apple is pushing OpenCL to be adopted cause it's an open format vs Nvidia proprietary CUDA and going for CUDA would mean giving Nvidia more monopoly which Apple as a company wouldn't want (look at Foxconn-Pegatron, Samsung-TSMC).

3. AMD OpenCL performs much better then Nvidia, Nvidia looks really pathetic when it comes to OpenCL (probably they focus on CUDA only), and this makes AMD the better option for Apple emphasis in OpenCL applications.

4. The current FCPX improved performance is mainly cause of OpenCL, if Apple was to use Nvidia card, you'll not see the performance boost that FCPX is enjoying now. Of course on the other side, Premiere would see a performance boost with Nvidia then AMD but then, Premiere is FCPX competitor and obviously Apple wouldn't want to show that it performs faster then FCPX. On the bright side, Adobe promises to improve OpenCL support in the coming updates, so it's a win win situation for everyone (okay, maybe not for Nvidia :p)

Well said!
 
I doubt it will be anytime soon for a few reasons:

1. Nvidia won't drop their GPU price to AMD level, they are not bothered. Nvidia basically gave the whole next gen console market to AMD cause the customer (Sony, Microsoft, Nintendo) demand is deemed unrealistic. And since Apple offer dual GPUs in all MacPro, using Nvidia card would lead to a price increase, they could offer it as an upgrade option but I doubt it as Apple custom design the cards in nMP.

2. Apple is pushing OpenCL to be adopted cause it's an open format vs Nvidia proprietary CUDA and going for CUDA would mean giving Nvidia more monopoly which Apple as a company wouldn't want (look at Foxconn-Pegatron, Samsung-TSMC).

3. AMD OpenCL performs much better then Nvidia, Nvidia looks really pathetic when it comes to OpenCL (probably they focus on CUDA only), and this makes AMD the better option for Apple emphasis in OpenCL applications.

4. The current FCPX improved performance is mainly cause of OpenCL, if Apple was to use Nvidia card, you'll not see the performance boost that FCPX is enjoying now. Of course on the other side, Premiere would see a performance boost with Nvidia then AMD but then, Premiere is FCPX competitor and obviously Apple wouldn't want to show that it performs faster then FCPX. On the bright side, Adobe promises to improve OpenCL support in the coming updates, so it's a win win situation for everyone (okay, maybe not for Nvidia :p)

Don't AMD got their "Mantle" thing too? Which is not open format..
 
Don't AMD got their "Mantle" thing too? Which is not open format..

Mantle doesn't run on OS X. What folks do in other ecosystems is largely immaterial to Apple making decisions to optimize the stack in their own.

When Mantle gets ported to something that isn't Windows perhaps can revisit but I wouldn't hold my breath. It wouldn't be surprising to see it show up on Xbox (windows dervative and it is reportedly partially a clone of anyway). Or to see it used to move games to Playstation. Maybe Steam.

But OS X is likely much further down the list than others despite the rumors and whisperings that Mantle ports showing up any time soon. As long as AMD is:

a. giving max support to OpenGL and OpenCL
b. adding Mantle so it doesn't destabilize what OS X is doing with GPUs. (the two stacks to the shared graphics hardware are going to have to get along. )

Apple won't care.

Nvidia has problems with Apple because they have been drinking as well as selling the kool-aid that OpenCL lost and wasn't good reason to put full effort into it anymore. As long as Nvidia is slacking on OpenCL 1.2 (and above ) support they are at risk of being perpetual loser in Apple design bake-offs.
 
Don't AMD got their "Mantle" thing too? Which is not open format..

Hmm, I just read about AMD Mantle and it does sound similar but Mantle if I'm not mistaken is more of a low level graphics API whereas OpenCL is low level compute API (OpenComputingLanguage), their purposes in application will probably be different. Correct me if I'm wrong :).
 
Any chance Apple will offer an nVidia card in the future? I find it odd that they went down the AMD route as they use nVidia as the discrete cards in their rMBP products.

Nvidia did NOT win the design bake-off about 2 years ago because they had CUDA. At the time they had a three good and one neutral things going for them.

a. better perf/W than AMD
b. better OpenCL support
c. more motivated because had lost the last design bake-off and were frozen out (so probably had better pricing).
d. timing coming out in time to be at volume production for the laptop refresh.

Over the last year or so Nvidia has thrown some of those under the bus.
'.c' they had won so probably feeling cocky (especially with the Intel cash still flowing in )
'b' somehow decided to stop putting serious work into OpenCL. ( Perhaps thought AMD and Intel would continue to stuggle to get their act together on OpenCL... those two are on OpenCL 1.2 and Nvidia is not. )

Whether they still have a large edge on 'a' is up in the air. We'll see when the next generations roll out. It will AMD's second stab at GCN so decent chance they will close the gap since not doing a radical shift in architecture this time. (far more design work going into optimization than creating from scratch.) Same thing happened ~4 years ago when Nvidia shifted architecture and AMD moved in with better match to Apple's design criteria.


For Mac Pro 'd' is really a non issue since toward end of generation for both AMD and Nivida. On new generations Nvidia's larger GPUs need Apple's dates to be far enough away from the TMSC (since both AMD & NVidia using same fab) transition date.

Not fully supporting OpenCL is a even bigger looser for a Mac Pro design bake-off.

Yes I know I could use OpenCL but I find it much more difficult to use and debug.

Sometimes this is a "I like IDE toolset A, better than IDE toolset B" issue. CUDA is a lock-in construct. If find yourself locked in then that is partially because it is doing what it is designed to do.

If there something broken about the debugger than should let Apple/AMD know.
 
Hmm, I just read about AMD Mantle and it does sound similar but Mantle if I'm not mistaken is more of a low level graphics API whereas OpenCL

I don't think that is the point he was trying to make. They are both proprietary so it is notion that AMD is "just as bad" because they are doing proprietary API too. That completely misses the relevant issues.

If AMD gets to the point that in order to push Mantle they start throwing OpenGL under the bus then they'll be in the same "was nice working with you" situation as Nvidia is drifting toward with Apple. So far there are zero signs that Mantle means AMD is backing off fully supporting DirectX or OpenGL.
 
If AMD gets to the point that in order to push Mantle they start throwing OpenGL under the bus then they'll be in the same "was nice working with you" situation as Nvidia is drifting toward with Apple. So far there are zero signs that Mantle means AMD is backing off fully supporting DirectX or OpenGL.

Hmm, I guess it all falls on what will the software developer do, cause choosing Mantle or CUDA would result in their software being locked in to the respective GPU manufacturer, I'm not talking about game development cause from gaming benchmark, you can clearly see some games are optimized for AMD card whereas some Nvidia.

Octane Render for example is a pure CUDA 3D rendering software, meaning they'll be relying on Nvidia to be the dominant workstation GPU market holder, but if someday AMD somehow manages to beat Nvidia, Octane is pretty much in trouble, and switching their software 3D kernel isn't exactly an easy thing as they'll have to do a converter for old files and bla bla bla.

Inventor for example chose Direct3D which means they'll be locked in the Windows platform for the time being, Siemens NX being an OpenGL software is available on the Windows, Mac and Linux platform.

So I guess it really ends up on what will the developers do, to embrace a proprietary or an open platform.

Now that we're talking about this, I wonder could this be the very reason why most 3D rendering software is still CPU based?
 
Maybe Charlie can shine some light on this once the article is available to the rest of us.

http://semiaccurate.com/2013/12/30/nvidia-lose-apple-macpro-line/

Not sure that is going to be insightful given there is questionable stuff in the part you can read.

As a single product the Mac Pro probably is the best selling 'workstation' system. Most other workstation vendors have multiple workstation products. Individually those products don't match up but collectively they are much larger. So for the overall workstation market Apple's share is about as low (or lower) as there overall PC market share (i.e., sub 8% ). Apple is well outside of the top 4 vendors in the market.

Similar with AMD's 10% jump when they smaller share.

http://www.anandtech.com/show/6137/the-amd-firepro-w9000-w8000-review-part-1/3

Part of AMD's problems has been shooting themselves in the foot by not executing at times. Some of the swing is them just getting their act together. That is hardly going to trigger Nvidia to give away GPUs at firesale prices to Apple (who going to charge prices to customers the supports fully paying for them. )

Winning the Mac Pro isn't going to put AMD into the lead or even close to 50% of the workstation graphics market. It would give them a solid foundation where they aren't fluctuating as much and/or having problems with software vendors wanting to go through support certification process.

Nvidia's share is not imploding. Natrually eroding yes, but that is more so because have a competitive market than huge marketing blunder by Nvidia.


If the later stuff tracks into Nvidia somewhat spotty OpenCL track record of late then it would be more so on track.
 
I see no reason to mention Mantle in this discussion. It is a low-level API to replace atrocious DirectX draw call limitations.

This is part of the reason why you can code "to the metal" on consoles, because they offer low-level access, as opposed to how DirectX works in Windows. Microsoft have become lazy with DirectX and have been using it to force users to upgrade to the newest version of Windows to get access to the newest version of DirectX, hence holding the gaming industry back and creating platform fragmentation.

Mantle is NOT like OpenCL or CUDA.
 
Hmm, yeah I was afraid "small chance" was going to be the answer I was going to get. I hadn't thought of the fact that FCP has been optimized for openCL.

Hopefully PCIe thunderbolt expansion chassis that support video cards come along sooner rather than later.

Also most of my programs are more of the "Scientific" variety and for the most part, everything at my particular institution is written in CUDA.

Thanks for the thought folks.
 
Hmm, yeah I was afraid "small chance" was going to be the answer I was going to get. I hadn't thought of the fact that FCP has been optimized for openCL.

Hopefully PCIe thunderbolt expansion chassis that support video cards come along sooner rather than later.

Also most of my programs are more of the "Scientific" variety and for the most part, everything at my particular institution is written in CUDA.

Thanks for the thought folks.

Thinking about it logically there is little business sense for Apple to tie itself into Nvidia and CUDA.
 
Hopefully PCIe thunderbolt expansion chassis that support video cards come along sooner rather than later.

the chassis are largue opaque to this being enabled. It is far more enabling the GPU device driver and graphics stack that anything the chassis is going to do. Outside of the top tier power sucking cards, there is not much the enclosure has to do. And frankly that particular subclass is an odd mismatch of the 300W card choked down to ~ x3 PCIe v2 data pipeline. That's alot of horsepower behind a small straw. I doubt there is a huge market for that solution.


Also most of my programs are more of the "Scientific" variety and for the most part, everything at my particular institution is written in CUDA.

Apple is busy weaving OpenCL into the OS X foundational libraries where it makes sense. Apple is making multimillion $ investments in OpenCL.

Institutions have gone for vendor lock-in the past. In HPC over a decade ago it could have been custom Cray Fortran extensions. Or a particular compiler or particular hardware. It is a common, repetitive approach that almost always dead-ends after a number of years where they jump from that niche to another.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.