Yeah, you're right. At my university some groups tried to use CUDA but encountered many problems. Perhaps the situation is better for Windows but OpenCL simply works on nvidia AND ati cards AND with Linux and most likely with Windows, too ;-).
OpenCL is managed by the Khronos Group which is open to submissions from any company, so Nvidia are free to enhance OpenCL with whatever CUDA-type features they like. It does not matter what nifty features CUDA has, nobody is going to make a serious commitment to it unless it is cross-platform and Open.
It is a perfect excuse. MS used the same tactics to distinguish themselves. Why use "open" crap like Java when you can use much nicer proprietary .Net? Same with Apple, same logic - comfort wins. Why use crap "open" PC when more comfortable MacBooks, iPads and iPhones win although they are 100% proprietary (crap like darwin/webkit/etc doesn't help, don't even try to pull that BS here)
I mean nice sweet comfortable quality products like DirectX/Mac/Windows/.Net always pwn open crap like Linux/Unix/Java/OpenGL, simply because they are way more user friendly. People don't care about open crap of religious freaks like RMS, they want nice stuff to work on, who cares if crap is free and nice stuff is not? This is how it's supposed to be in a market economy.
Same here with CUDA. nVidia wants CUDA to be nice sweet proprietary product. All the religious freaks keep sucking RMS and use "open" CL and "open" everything. Normal people who want the task done as quickly and swiftly as possible just pay up for proprietary nVidia/Windows/etc shit and do the job.
If you ever hear so called "scientists" complain about oh their sweet "open" Linux not working with CUDA - ignore then. Real scientists just use CUDA on Windows and get the job done. Fake ones keep complaining about Linux shminux etc, them losers just need another excuse to slack off.
Yeah and don't forget 3DFX, oh wait... Also before you bring up the Bullshit about Opengl vs Direct X, Direct X is able to be implemented by anyone. It just so happens though that it is only usable on Windows.
I think that even NVIDIA knows that OpenCL is the future, but why not try to make as much money as possible with proprietary CUDA while they can? It's business, not charity. AFAIK, NVIDIA's cards also support OpenCL functions, so they can always fall back on it if (when) CUDA dies.
Assuming that Microsoft doesn't take over with an API of their own at some point...
I think that OpenCL is the future, but based on nVidia's past I don't think for one second *they* (more specifically, Jen-Hsun Huang) believe that the market desires a less-expensive and competitive open platform.
CUDA is literally the ability to run C code on an nvidia GPU (and fortran)... I see no reason for it to die. There is absolutely no reason for nvidia to do things via openCL, and there is nothing wrong with nvidia doing things via CUDA. don't like it, don't run your scientific processing on CUDA...
Just because they're adding to CUDA doesn't mean they've forgotten OpenCL. Most of what OpenCL is, is Cuda being toned down so it can work on ATI cards.
You know as soon as the Khronos group gets around to updating OpenCL NVidia will be more than ready as their cards will already support it, and they'll have 90% of the driver code/utilities already written. They've supported OpenCL since its inception and have had drivers for it as soon as was possible.
Its great to see CUDA being updated. Because progress on CUDA will eventually mean more progress on OpenCL as the 'must have' features trickle down.
CUDA as far superior, more capable, and has better tools than OpenCL. Theres things it can do right NOW that OpenCL simply cant. Are these people that use it meant to give up on what they do because they're not using something thats "open"? WTF kind of logic is that? Are Microsoft meant to stop developing Windows and DirectX because they are not as open as Linux and OpenGL? Are people not meant to have a choice and choose which is best for them?
All these people like you commenting here are so stupid i actually feel sorry for you.
And open does not always = better. It has advantages, but many open standards take far too long to be updated because of all the parties involved. A good example is Flash vs most other web tech/languages. Flash gets updates every year with new features and will always be way ahead of HTML because of this, the last time HTML was updated before version 5.0 was in the late 90's. I'm sure it wont be this bad with OpenCL but it wont catch up to CUDA, or atleast for many years.
Guys I know you are writing these articles like mad (and we greatly appreciate them), but the last 2 days has been awful. I don't mind mispellings, but when I have to reread a sentence more than once it gets tiring (or in the case of this beauty, " in large part to the GPU’s ability to the GPU’s unified address space, ECC support, and support for C++", not even grasp what you are trying to say).
Please just do a once over if possible to make them a bit more reader-friendly.
Any chance we could eventually use WDM drivers for our regular desktop GPUs? For example GPU-Grid has serious performance problems udner Vista and Win 7 thanks to WDDM drivers. And since soon all CPUs are going to bring some integrated GPU along it would only be logical to use the discrete GPU for crunching (and switching to games if needed) and the smallest GPU in the PC to drive the desktop.
Unfortunately no. I actually asked specifically about that scenario (Folding, GPUGrid, etc). It's a Tesla-only feature for product differentiation reasons.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
22 Comments
Back to Article
Soulkeeper - Tuesday, September 14, 2010 - link
Too bad they didn't focus on opencl instead.cuda needs to die already.
NaN42 - Tuesday, September 14, 2010 - link
Yeah, you're right. At my university some groups tried to use CUDA but encountered many problems. Perhaps the situation is better for Windows but OpenCL simply works on nvidia AND ati cards AND with Linux and most likely with Windows, too ;-).Lonbjerg - Tuesday, September 14, 2010 - link
Because OpenCL are way behind CUDA.Why should NVIDIA handicap them selfes just to "level the playing field"...make no sense what so ever.
raddude9 - Tuesday, September 14, 2010 - link
"Because OpenCL are way behind CUDA."Not an excuse.
OpenCL is managed by the Khronos Group which is open to submissions from any company, so Nvidia are free to enhance OpenCL with whatever CUDA-type features they like.
It does not matter what nifty features CUDA has, nobody is going to make a serious commitment to it unless it is cross-platform and Open.
Pirks - Wednesday, September 15, 2010 - link
It is a perfect excuse. MS used the same tactics to distinguish themselves. Why use "open" crap like Java when you can use much nicer proprietary .Net? Same with Apple, same logic - comfort wins. Why use crap "open" PC when more comfortable MacBooks, iPads and iPhones win although they are 100% proprietary (crap like darwin/webkit/etc doesn't help, don't even try to pull that BS here)I mean nice sweet comfortable quality products like DirectX/Mac/Windows/.Net always pwn open crap like Linux/Unix/Java/OpenGL, simply because they are way more user friendly. People don't care about open crap of religious freaks like RMS, they want nice stuff to work on, who cares if crap is free and nice stuff is not? This is how it's supposed to be in a market economy.
Same here with CUDA. nVidia wants CUDA to be nice sweet proprietary product. All the religious freaks keep sucking RMS and use "open" CL and "open" everything. Normal people who want the task done as quickly and swiftly as possible just pay up for proprietary nVidia/Windows/etc shit and do the job.
If you ever hear so called "scientists" complain about oh their sweet "open" Linux not working with CUDA - ignore then. Real scientists just use CUDA on Windows and get the job done. Fake ones keep complaining about Linux shminux etc, them losers just need another excuse to slack off.
UltraWide - Wednesday, September 15, 2010 - link
I agree 100%, there is the ideal world and the real world.Sahrin - Tuesday, September 14, 2010 - link
Because more open platform = more users = more profits?Lonbjerg - Wednesday, September 15, 2010 - link
That worked soooo goood for OpenGL right?sc3252 - Wednesday, September 15, 2010 - link
Yeah and don't forget 3DFX, oh wait... Also before you bring up the Bullshit about Opengl vs Direct X, Direct X is able to be implemented by anyone. It just so happens though that it is only usable on Windows.therealnickdanger - Tuesday, September 14, 2010 - link
I think that even NVIDIA knows that OpenCL is the future, but why not try to make as much money as possible with proprietary CUDA while they can? It's business, not charity. AFAIK, NVIDIA's cards also support OpenCL functions, so they can always fall back on it if (when) CUDA dies.Assuming that Microsoft doesn't take over with an API of their own at some point...
Sahrin - Tuesday, September 14, 2010 - link
I think that OpenCL is the future, but based on nVidia's past I don't think for one second *they* (more specifically, Jen-Hsun Huang) believe that the market desires a less-expensive and competitive open platform.taltamir - Tuesday, September 14, 2010 - link
CUDA is literally the ability to run C code on an nvidia GPU (and fortran)... I see no reason for it to die.There is absolutely no reason for nvidia to do things via openCL, and there is nothing wrong with nvidia doing things via CUDA.
don't like it, don't run your scientific processing on CUDA...
Crimson001 - Tuesday, September 14, 2010 - link
Just because they're adding to CUDA doesn't mean they've forgotten OpenCL. Most of what OpenCL is, is Cuda being toned down so it can work on ATI cards.You know as soon as the Khronos group gets around to updating OpenCL NVidia will be more than ready as their cards will already support it, and they'll have 90% of the driver code/utilities already written. They've supported OpenCL since its inception and have had drivers for it as soon as was possible.
Its great to see CUDA being updated. Because progress on CUDA will eventually mean more progress on OpenCL as the 'must have' features trickle down.
mindbomb - Tuesday, September 14, 2010 - link
they included directcompute.opencl is basically dead though. It's too difficult to use.
B3an - Wednesday, September 15, 2010 - link
Kids like you need to die already.CUDA as far superior, more capable, and has better tools than OpenCL.
Theres things it can do right NOW that OpenCL simply cant. Are these people that use it meant to give up on what they do because they're not using something thats "open"? WTF kind of logic is that? Are Microsoft meant to stop developing Windows and DirectX because they are not as open as Linux and OpenGL? Are people not meant to have a choice and choose which is best for them?
All these people like you commenting here are so stupid i actually feel sorry for you.
And open does not always = better. It has advantages, but many open standards take far too long to be updated because of all the parties involved. A good example is Flash vs most other web tech/languages. Flash gets updates every year with new features and will always be way ahead of HTML because of this, the last time HTML was updated before version 5.0 was in the late 90's. I'm sure it wont be this bad with OpenCL but it wont catch up to CUDA, or atleast for many years.
7Enigma - Tuesday, September 14, 2010 - link
Guys I know you are writing these articles like mad (and we greatly appreciate them), but the last 2 days has been awful. I don't mind mispellings, but when I have to reread a sentence more than once it gets tiring (or in the case of this beauty, " in large part to the GPU’s ability to the GPU’s unified address space, ECC support, and support for C++", not even grasp what you are trying to say).Please just do a once over if possible to make them a bit more reader-friendly.
catavalon21 - Tuesday, September 14, 2010 - link
" I don't mind mispellings..."You're sure you really want to go there?
dingetje - Wednesday, September 15, 2010 - link
opencl should have (at least) been mentioned in this articleaaron92 - Wednesday, September 15, 2010 - link
"Not to outdone..."Not too outdone or Not to be outdone?
MrSpadge - Wednesday, September 15, 2010 - link
Any chance we could eventually use WDM drivers for our regular desktop GPUs? For example GPU-Grid has serious performance problems udner Vista and Win 7 thanks to WDDM drivers. And since soon all CPUs are going to bring some integrated GPU along it would only be logical to use the discrete GPU for crunching (and switching to games if needed) and the smallest GPU in the PC to drive the desktop.MrS
Ryan Smith - Thursday, September 16, 2010 - link
Unfortunately no. I actually asked specifically about that scenario (Folding, GPUGrid, etc). It's a Tesla-only feature for product differentiation reasons.Richlet - Thursday, September 16, 2010 - link
Just a goof.. when I first *glanced* at the headline "parallel nsight" I honestly thought it read "parasight".. doh..