Apple to Support Nvidia's CUDA at WWDC?

[matthutch]

I just read this on CNet (might be old news to most by now but hey why not :) ) and thought it was pretty cool.

http://news.cnet.com/8301-13579_3-9962117-37.html?tag=nefd.top

This would be awesome if they could leverage GPU power for Quicktime encodes/transcodes. I wonder if Tyler from Visual Hub would be able to use it to speed up Visual Hub's encoding even more ;)

EDIT: Sorry if this is in the wrong section as well, Mods please move if I fubar'd it :)

[Cromulent]

I just read this on CNet (might be old news to most by now but hey why not :) ) and thought it was pretty cool.

http://news.cnet.com/8301-13579_3-9962117-37.html?tag=nefd.top

This would be awesome if they could leverage GPU power for Quicktime encodes/transcodes. I wonder if Tyler from Visual Hub would be able to use it to speed up Visual Hub's encoding even more ;)

EDIT: Sorry if this is in the wrong section as well, Mods please move if I fubar'd it :)

Yes, it is available in Mac OS X. Not sure what the point of the post is though?

CUDA is a pretty impressive technology if you are doing lots of maths heavy programming but because it is pretty limited (only Nvidia hardware for a start with more than 128MBs of graphics RAM) it's use is limited to a small section of the Mac community.

[matthutch]

Yes, it is available in Mac OS X. Not sure what the point of the post is though?


Was more to see what other people thought about it. From what you have said it seems like it is already implemented? Do you know any applications that are actively using it?

CUDA is a pretty impressive technology if you are doing lots of maths heavy programming but because it is pretty limited (only Nvidia hardware for a start with more than 128MBs of graphics RAM) it's use is limited to a small section of the Mac community.

This is more where I was wanting to get into, given the increasing performance of hardware more people would be opened up to these sort of areas so it would be interesting to see how current developers envisage their ability to leverage the technology today. Seeing that even the X3100 in the current MacBook's can utilize 144Mb of RAM quite a few machines would be able to use it, if only a little bit (I'm not to clued in on the details of the memory requirements beyond what you've mentioned I admit), for a little bump in encoding speed.

Apologies if this doesn't make much sense either, working early on a Saturday morning has its draw backs :)

[MacRumors]

http://www.macrumors.com/images/macrumorsthreadlogo.gif (http://www.macrumors.com)

CNet interviewed (http://news.cnet.com/8301-13579_3-9962117-37.html) Nvidia CEO Jen-Hsun Huang and found that Apple may have an interest in Nvidia's CUDA (http://www.nvidia.com/object/cuda_home.html) technology: CUDA is a programming technology that allows software developers to take advantage of the unique parallel processing characteristics of graphics processors such as Nvidia's GeForce 8600M, found in the MacBook Pro. According to Huang, "Apple knows a lot about CUDA" and may announce support for the technology at next week's' Worldwide Developers Conference (WWDC).

During a demo for CNet, Nvidia engineers demonstrated how a CUDA-enabled version of a program could dramatically speed up converting video from one format to another. Transcoding video can be useful to convert existing video to be played on another device (such as the iPhone).


Article Link (http://www.macrumors.com/2008/06/06/apple-to-support-nvidias-cuda-at-wwdc/)

[ubestbsteppin]

So is this something that could be included in a leopard update?

[eXan]

Only 8600?

[arn]

Only 8600?

Here's a full list of supported GPUs.

http://www.nvidia.com/object/cuda_learn_products.html

arn

[Peace]

Ah.So there could be a firmware update for the MacPro 8800 video card.
Man. Cuda combined with the 8-cores would make video a breeze.

[twoodcc]

hey, i'll take a speed increase on converting my videos

[mahonmeister]

So then will Apple start offering more powerful video cards, at least as a bto? They're a little better now, but there is tons of room for improvement.

Perhaps they could even end the (mostly true) stereotype that Macs suck at gaming.:p

[iceman1234]

So then will Apple start offering more powerful video cards, at least as a bto? They're a little better now, but there is tons of room for improvement.

Perhaps they could even end the (mostly true) stereotype that Macs suck at gaming.:p

I know really, why can't apple use any decent graphic cards anymore:(

[mrgreen4242]

I don't see Apple integrating a vendor specific hardware function into the OS. Look at Quartz Extreme, for example... it works on any manufacturers hardware, as long as it meets certain requirements (certain level of support for OpenGL, I believe).

I could see them adding this as a feature to Final Cut or something, but to roll it into to OS would only benefit a very small number of their users and just would seem to a very Apple thing to do.

[DavidCar]

Does anyone know how CUDA compares with the EyeTV Turbo.264? Could a CUDA graphics processor do the same thing as the Turbo?

[RTee]

I don't see Apple integrating a vendor specific hardware function into the OS. Look at Quartz Extreme, for example... it works on any manufacturers hardware, as long as it meets certain requirements (certain level of support for OpenGL, I believe).

I could see them adding this as a feature to Final Cut or something, but to roll it into to OS would only benefit a very small number of their users and just would seem to a very Apple thing to do.

That's true, if it's available for those that want to use it then that's great.

[geerlingguy]

I'd be quite happy if this happened, but I don't expect it. Half of my computer's waking hours are spent converting video (transcoding my DVD collection, and also some of my HD-DVDs... although they take much longer), so this would make my life a lot easier.

[exabytes18]

I believe CUDA is now bundled with all 8 and 9 series drivers for Windows though it seems like most implementations of CUDA are done through Linux.

CUDA is capable of some incredible feats, but programming for it looks brutal. I believe there is also talk of nVidia adding a feature to the CUDA compiler to allow for compiling the code for multi-core CPUs in addition to the GPU. Currently, CPUs can emulate the GPU code, but it's inefficient. The idea is that the programmer writes the code once and then the compiler takes care of the rest. If there's a CUDA enabled GPU, the program will utilize the GPU, else the program will execute entirely within the CPU (utilizing any cores available.) I read this awhile back so it could either be.... a) scraped, b) still in the works, c) implemented in CUDA 2.0 or something, d.) I misunderstood and it never happened.

Anyway, CUDA-like technologies are amazing. It would be awesome to see it included in OS X... or at least for the 8-series cards.

[matthutch]

I believe there is also talk of nVidia adding a feature to the CUDA compiler to allow for compiling the code for multi-core CPUs in addition to the GPU. Currently, CPUs can emulate the GPU code, but it's inefficient. The idea is that the programmer writes the code once and then the compiler takes care of the rest. If there's a CUDA enabled GPU, the program will utilize the GPU, else the program will execute entirely within the CPU (utilizing any cores available.) I read this awhile back so it could either be.... a) scraped, b) still in the works, c) implemented in CUDA 2.0 or something, d.) I misunderstood and it never happened.


That sounds pretty cool, if it is being implemented :)

On a side note: First page news, awesome :) I sure didn't think that would happen. I suppose there are benefits about working early on a Saturday :)

[winterspan]

Yes, it is available in Mac OS X. Not sure what the point of the post is though?

CUDA is a pretty impressive technology if you are doing lots of maths heavy programming but because it is pretty limited (only Nvidia hardware for a start with more than 128MBs of graphics RAM) it's use is limited to a small section of the Mac community.

So what! It's good that ANY existing Macs have the technology for it. All middle and higher end Nvidia 8-series and higher cards can support CUDA. It's really going to be an incredible technology.


Seeing that even the X3100 in the current MacBook's can utilize 144Mb of RAM quite a few machines would be able to use it, if only a little bit (I'm not to clued in on the details of the memory requirements beyond what you've mentioned I admit), for a little bump in encoding speed.
Apologies if this doesn't make much sense either, working early on a Saturday morning has its draw backs :)

Intel's integrated graphics are a joke compared to even low-end Nvidia graphics cards. Only the middle-range/upper-range Nvidia Geforce 8-series and later cards can run CUDA.

I know really, why can't apple use any decent graphic cards anymore:(
This would be GREAT if implementing CUDA spurs apple to take graphics cards seriously!

I don't see Apple integrating a vendor specific hardware function into the OS. Look at Quartz Extreme, for example... it works on any manufacturers hardware, as long as it meets certain requirements (certain level of support for OpenGL, I believe). I could see them adding this as a feature to Final Cut or something, but to roll it into to OS would only benefit a very small number of their users and just would seem to a very Apple thing to do.

CUDA is capable of some incredible feats, but programming for it looks brutal. I believe there is also talk of nVidia adding a feature to the CUDA compiler to allow for compiling the code for multi-core CPUs in addition to the GPU. Currently, CPUs can emulate the GPU code, but it's inefficient. The idea is that the programmer writes the code once and then the compiler takes care of the rest. If there's a CUDA enabled GPU, the program will utilize the GPU, else the program will execute entirely within the CPU (utilizing any cores available.) ...

@mrgeen & exbytes

I bet Apple is going to either buy or work with one of those startup companies that have been developing code and compiler technology to speed up applications on multi-core CPUs and GPUS. This way they could implement a common abstraction layer for both CUDA and ATI's Close-To-Metal, not to mention Intel's Larabee in the future, and potentially even just multi-core processors. Then just like some of the frameworks in OSX, it could actively run the application's code on whatever particular hardware you have on the machine it available. That would be the best way to keep their suppliers diversified, while being able to take advantage of all these new technologies...

[Analog Kid]

You know, the last time a CEO made a public statement about Apple adopting their technology was Sun saying ZFS would be "the" file system for Leopard.

We see how that turned out...

I agree this doesn't belong in the core OS. Apple tends to bring generic technologies into the system when they can abstract them across hardware vendors. Core Image and the Accelerate framework do a lot of what people here are asking for.

[fuziwuzi]

here is a post and link for a video of CUDA.

i am highly impressed, and kinda stoked i got the new MBP. i really hope they put this technology to good effect, and hopefully they won't charge us too much for the software

http://forums.vr-zone.com/showthread.php?t=281340

[sgntscrawn]

I for one think this'd be cool, especially since I own a MBP with 8600M... what sort of increase in performance are we talking, here though?

[fuziwuzi]

look at the link above yourpost

[exabytes18]

This is not the sort of thing that's included in the core or the OS (kernel?) Not everyone is going to use it, but its offerings are too good to be ignored.

The observed FLOPS for a card like the 8800GT totally trounces the theoretical peak FLOPS of any Intel processor. If you talk theoretical, the 8800GT can do 336 GFLOPS while a pair of 3.2ghz Harpertowns can do 204.8 GFLOPS. That's the kind of power Apple's interested in. :)

[Cromulent]

Seeing that even the X3100 in the current MacBook's can utilize 144Mb of RAM quite a few machines would be able to use it

Unfortunately the X3100 is not an Nvidia graphics card and thus would not work with CUDA.

I've been thinking about looking into CUDA for a financial program that I need to write at some point which will need to do some analysis on a large amount of data very quickly. As others have mentioned video would also benefit a lot from the technology. But CUDA is a very specialised technology which can only benefit mathematical and scientific applications at the end of the day.

[fuziwuzi]

Unfortunately the X3100 is not an Nvidia graphics card and thus would not work with CUDA.

I've been thinking about looking into CUDA for a financial program that I need to write at some point which will need to do some analysis on a large amount of data very quickly. As others have mentioned video would also benefit a lot from the technology. But CUDA is a very specialised technology which can only benefit mathematical and scientific applications at the end of the day.

here is a medical application of the CUDA. seriously, i am very excited about this technology. i am so shocked that it's taken this long for someone(nvidia) to figure out that people would use video card for more than games.

http://www.youtube.com/watch?v=DnIvodB2RzU

[kornyboy]

Wirelessly posted (iPhone: Mozilla/5.0 (iPhone; U; CPU like Mac OS X; en) AppleWebKit/420.1 (KHTML, like Gecko) Version/3.0 Mobile/4A102 Safari/419.3)

This is really cool. Does anyone know how much it will increase conversion times? Like 10% or more?

[matthutch]

Unfortunately the X3100 is not an Nvidia graphics card and thus would not work with CUDA.


Yeah I didn't realise that CUDA required nVidia cards to work at the time of my post unfortunately :(. I've been reading up on it throughout the course of the day, it seems like it could be really cool if developers can leverage its potential in the right way.

[Pressure]

This is not the sort of thing that's included in the core or the OS (kernel?) Not everyone is going to use it, but its offerings are too good to be ignored.

The observed FLOPS for a card like the 8800GT totally trounces the theoretical peak FLOPS of any Intel processor. If you talk theoretical, the 8800GT can do 336 GFLOPS while a pair of 3.2ghz Harpertowns can do 204.8 GFLOPS. That's the kind of power Apple's interested in. :)

CUDA is basically just marketing branded General-Purpose computation on GPUs.

This is something all DirectX 10 hardware can do easily with their programmable nature.

If you drool about the Geforce 8800 GT theoretical GigaFLOPS, you should really be interested in the GT200 chip that delivers 933GigaFLOPS and the Radeon HD 4800 series that deliver 1 TeraFLOPS of computational power.

[matthutch]

As others have mentioned video would also benefit a lot from the technology. But CUDA is a very specialised technology which can only benefit mathematical and scientific applications at the end of the day.

Surely the applications which could/would benefit from the technology would range in their real world application, it is merely a case of the technologies full potential is not truly realised. Video was one application mentioned directly in the article (see below), and I remember reading [on Macrumors if I'm not mistaken] that Adobe was planning to utilize GPU based processing in an upcoming release of their CS packages [CS4 or CS5]. Found the article (http://www.macrumors.com/2008/05/24/next-version-of-photoshop-to-be-gpu-accelerated/)

From the Article in my original post.
For example, during my visit on Wednesday, Nvidia engineers demonstrated how a CUDA-enabled version of a program similar to QuickTime running on a desktop or laptop could dramatically speed up the processor of transcoding a movie or television show into a format suitable for the iPhone.

If you drool about the Geforce 8800 GT theoretical GigaFLOPS, you should really be interested in the GT200 chip that delivers 933GigaFLOPS and the Radeon HD 4800 series that deliver 1 TeraFLOPS of computational power.

I saw some people referring to those GT200 chips today during my hunting around on this, they looked pretty insane with what they could do.

Mind you it would be interesting to see GPGPU develop if this (http://news.bbc.co.uk/2/hi/technology/7439406.stm) IBM technology is implemented on GPU's. Multi-Core GPU's in collaboration with Multi-Core CPU's could pump out some pretty impressive results. It would be a fair way off though I would imagine as they are just working on the technology now, and from what I understand multi-core GPU's would be too big and too hot if they tried with current tech (please correct if I am wrong on this. The more I have been looking into this the more intrigued I am :) )

[needsomecoffee]

You've gotta watch this video. CUDA is the next step to transform desktop computing. Intel is working furiously to catch up to nVidia. Microsoft is also working very hard and appears ready with a ground breaking HPC server 2008.

Really folks, watch this video to understand how HPC has the capacity to change lives in ways we've never understood. Every wild-ass student or scientist likely can afford his/her own supercomputer in a few years. That's what its going to take for real breakthrus in some 'in the box' constrained research areas.

BTW: This video is (intentionally I think) one of the most borderline hilarious nerdy, geeky things ever. But for those whose lives are saved by their tomo process finding a cancer etc. it is also over the line miraculous.

http://www.youtube.com/watch?v=DnIvodB2RzU

[fuziwuzi]

You've gotta watch this video. CUDA is the next step to transform desktop computing. Intel is working furiously to catch up to nVidia. Microsoft is also working very hard and appears ready with a ground breaking HPC server 2008.

Really folks, watch this video to understand how HPC has the capacity to change lives in ways we've never understood. Every wild-ass student or scientist likely can afford his/her own supercomputer in a few years. That's what its going to take for real breakthrus in some 'in the box' constrained research areas.

BTW: This video is (intentionally I think) one of the most borderline hilarious nerdy, geeky things ever. But for those whose lives are saved by their tomo process finding a cancer etc. it is also over the line miraculous.

http://www.youtube.com/watch?v=DnIvodB2RzU
i already posted it, last post, first page

[airfang]

Correct me if I am wrong... but doesn't CUDA already have Mac version toolkit/SDK?

What's the point of "to support CUDA at WWDC"?

http://www.nvidia.com/object/cuda_get.html#macos

[fuziwuzi]

Correct me if I am wrong... but doesn't CUDA already have Mac version toolkit/SDK?

What's the point of "to support CUDA at WWDC"?

http://www.nvidia.com/object/cuda_get.html#macos

yeah, but i think you'll find that this rumours suggests a implementation of cuda within OS X. so, either the release, or suggestion, that apple will have products that take advantage of cuda.

the fact u can develop CUDA on OS X is unimportant to most.

that is what is meant in the front page by "support"

[chickenninja]

i hope cuda is a cute name like google or yahoo, and not an acronym like msn. accronyms are too wide spread, they mean totally different things to different people and they create barriers in coversation. "TLPD" tension line propelling device, better named "rope thrower" or "roper". its not just accronyms its the fact we overcomplicate our titles and names to the point that they need abbreviation. and begging letters are not the answer because they mean different things in different fields. Ex: SC=star-craft, safety-commisioner, social-consensus, spinal-calapse. we need to ween ourselfs off of accronyms and get back to inventing new words when they are warrented. ex: cell-phone, much better than MCD (mobile-communication-device) or is that one for mini-compact-disk.
some may say that an accronym like spd can be used frequently and quickly in a conversation that is about spd. to this i say use the word "it" for what its for.

[TonyHoyle]

here is a medical application of the CUDA. seriously, i am very excited about this technology. i am so shocked that it's taken this long for someone(nvidia) to figure out that people would use video card for more than games.


Folding@Home have been doing this on ATI cores since late 2006. They got a 30x speed increase out of it.

[Cromulent]

Surely the applications which could/would benefit from the technology would range in their real world application, it is merely a case of the technologies full potential is not truly realised. Video was one application mentioned directly in the article (see below), and I remember reading [on Macrumors if I'm not mistaken] that Adobe was planning to utilize GPU based processing in an upcoming release of their CS packages [CS4 or CS5]. Found the article (http://www.macrumors.com/2008/05/24/next-version-of-photoshop-to-be-gpu-accelerated/)

Well Photoshop is a graphical program and thus is very maths based. CUDA will not help one jot with speeding up say a word processor or a web browser. Graphics card are good for one thing and one thing only. Executing mathematical functions, therefore anything that requires the use of the maths will benefit to an extent.

[ert3]

Maybe this is a feature of Snow Leopard

[SPUY767]

Hopefully, Apple will get in good with nVidia and get early, low-yield parts like they do from Intel. That would be sweet. Apple makes Intel look good, and they could probably make nVidia look good with a resonable partnership.