Monday, February 28th 2011
New CUDA 4.0 Release Makes Parallel Programming Easier
NVIDIA today announced the latest version of the NVIDIA CUDA Toolkit for developing parallel applications using NVIDIA GPUs. The NVIDIA CUDA 4.0 Toolkit was designed to make parallel programming easier, and enable more developers to port their applications to GPUs. This has resulted in three main features:
"Having access to GPU computing through the standard template interface greatly increases productivity for a wide range of tasks, from simple cashflow generation to complex computations with Libor market models, variable annuities or CVA adjustments," said Peter Decrem, director of Rates Products at Quantifi. "The Thrust C++ library has lowered the barrier of entry significantly by taking care of low-level functionality like memory access and allocation, allowing the financial engineer to focus on algorithm development in a GPU-enhanced environment."
The CUDA 4.0 architecture release includes a number of other key features and capabilities, including:
For more information on the features and capabilities of the CUDA Toolkit and on GPGPU applications, please visit: http://www.nvidia.com/cuda
- NVIDIA GPUDirect 2.0 Technology -- Offers support for peer-to-peer communication among GPUs within a single server or workstation. This enables easier and faster multi-GPU programming and application performance.
- Unified Virtual Addressing (UVA) -- Provides a single merged-memory address space for the main system memory and the GPU memories, enabling quicker and easier parallel programming.
- Thrust C++ Template Performance Primitives Libraries -- Provides a collection of powerful open source C++ parallel algorithms and data structures that ease programming for C++ developers. With Thrust, routines such as parallel sorting are 5X to 100X faster than with Standard Template Library (STL) and Threading Building Blocks (TBB).
"Having access to GPU computing through the standard template interface greatly increases productivity for a wide range of tasks, from simple cashflow generation to complex computations with Libor market models, variable annuities or CVA adjustments," said Peter Decrem, director of Rates Products at Quantifi. "The Thrust C++ library has lowered the barrier of entry significantly by taking care of low-level functionality like memory access and allocation, allowing the financial engineer to focus on algorithm development in a GPU-enhanced environment."
The CUDA 4.0 architecture release includes a number of other key features and capabilities, including:
- MPI Integration with CUDA Applications -- Modified MPI implementations automatically move data from and to the GPU memory over Infiniband when an application does an MPI send or receive call.
- Multi-thread Sharing of GPUs -- Multiple CPU host threads can share contexts on a single GPU, making it easier to share a single GPU by multi-threaded applications.
- Multi-GPU Sharing by Single CPU Thread -- A single CPU host thread can access all GPUs in a system. Developers can easily coordinate work across multiple GPUs for tasks such as "halo" exchange in applications.
- New NPP Image and Computer Vision Library -- A rich set of image transformation operations that enable rapid development of imaging and computer vision applications.
o New and Improved Capabilities
o Auto performance analysis in the Visual Profiler
o New features in cuda-gdb and added support for MacOS
o Added support for C++ features like new/delete and virtual functions
o New GPU binary disassembler
For more information on the features and capabilities of the CUDA Toolkit and on GPGPU applications, please visit: http://www.nvidia.com/cuda
77 Comments on New CUDA 4.0 Release Makes Parallel Programming Easier
I still do not get why Nvidia gets soo much shit for making such a good product, while Intel's (and a bunch of other companies') USB gets no flak.
CUDA is royalty Free and you don't need to pay anything
But it's not the Case about PhysX , its not free and if you want to use
it on your Hardwares you must pay for the Licenses fee .
free to code a program to use cuda, is not free to enable your own video cards to run CUDA. no one but nvidia is allowed to accelerate cuda on their hardware. this is why we say its not open source. (whether the wording is correct or not, THAT is what we mean)
if it was truly open like openCL, everyone could implement it free.
a CUDA Driver for AMD Cards
well it's the attitude when it come to this matterit was from The Inquirer
So until OpenCL can offer exactly the same, CUDA must exist because that's what is best for developers and that's what it's best for us, because without it developers could not create the amazing things they are doing with CUDA right now, they would have to wait 1 or 2 years until OpenCL is in that same state as CUDA is right now. That would not benefit anyone, the fact that you need a Nvidia GPU is completely irrelevant, you can at least have access to those applications which you would not have otherwise.
Anyway, did you even know that Nvidia has always been first with their OpenCL conformant drivers? No? Check it: www.khronos.org/adopters/conformant-products/Exactly. Pushing OpenCL compatibility NOW. CUDA exists since 2006. That's a quite a difference. That is what you fail to understand.You are forced to use Windows, just the same way you are forced to use a Nvidia card. Supporting DirectX and not supporting CUDA based on the principle of their "openess" is NONSENSE. Both are propietary, free to use APIs and there's no difference at all. Anyone can use DirectX for free, anyone can use CUDA for free. If you want to be able to use it you need Windows, if you want to be able to use it you need a Nvidia card.
And that is only one of the applications. Thanks to CUDA (GPGPU actually, but we have discussed just enough why for the time being GPGPU == CUDA) they are going to be able to predict hearthquakes or tornados better, they can find out how molecules interact, they can find starts and other worlds, they will find minerals more easily, facial/GPS image recognition can help detect terrosists and terrorist camps. This is real not science fiction. Is having all those things now better than having to wait yet another 2 years? Definately. The fact that they need to use a Nvidia card for doing so is minuscule and completely irrelevant.
And like always, you are missing the point anyway. Open standards are better than propietary ones, and you should know by now that's also my POV, but actual applications are better than none and OpenCL it's been here for only 1 year or so, and it's not yet as good as CUDA. Most of the GPGPU applications in existence are CUDA based because CUDA exists since 2006 and it's in that time period when they started working on them. The world is definately better with those apps (even if just a little bit, it is better) and having CUDA updated so that those apps can get better and better is only a good thing for everyone. Wanting the demise of CUDA only so that it lefts the place for an API that, as it stands now, is inferior is evil and selfish.
Think how many people used Internet Explorer a few years back. Internet Explorer has undoubtedly saved many lives and puppy dogs, but Firefox, an open-source competitor, has managed to eat up a sizeable chunk of the market in a relatively short period of time and has probably saved a number of lives and puppies. Perhaps your estimate of 2 years to make the transition is overly pessimisitic.
Life saving and puppy dogs are not inherent to the technology you are discussing and if open-source alternatives exist that can serve the same purpose, they should be investigated and afforded preferential treatment.
Moreover, citing Windows use as an example of the hypocrisy of those who criticise the proprietary nature of CUDA adds no weight to your argument: two wrongs do not make a right.
This feels like Inception.
Don't read too much into the comparison, my point is that open-source alternatives exist and can rapidly achieve mass acceptance, provided they are given the chance.
But would have been the world (the internet) better without the propietary Navigator? Would have been better without IE, which became better than Navigator because Netscape stopped improving Navigator? It would not. I'm sorry but without Navigator there would be no IE and without either of them Firefox (or any open source code for that matter) would have never been posible, since it was created by people on the net, using the propietary software they were trying to displace.
So in two or 3 years we we'll all be using OpenCL apps and CUDA will be a thing of the past or something that catters to a shrinking niche or something. And it will be a good thing. But for the time being CUDA is nothing but a good thing. Simple as that.
Nobody is arguing that CUDA is not helpful as things stand, the argument is that it could and should be replaced by a non-proprietary alternative and how long that would take. But as you seem to be saying just that, we appear to be agreeing, although I will not easily forget that you have tried to present Jen-Hsun Huang as Mother Teresa of Calcutta in a CUDA thread.
You know where I'm going? The thing is that the Greeks won, democracy survived, and we live in a mostly democratic world which we know is better than living under dictatorship.I have at no point mentioned Jen-Hsun. I have never said what you claim I said. We could argue forever about if guns are evil or not. I'm of the particular people who thinks they are evil, in the sense of that they are created to kill and are used to kill. But I'm very well aware of the fact that they can and are used for good puposes and in the particular case in which a gun is used to stop someone from killing innocent people, that gun is a good thing. Hence the existence of that gun is good.
I'm just saying the same about CUDA. I don't care who has created CUDA and what benefits they may get from it, it's irrelevant for me and it's irrelevant for humanity. I know how CUDA is being used and that it's good. It has created a market that is in expansion and has allowed for things that would not be posible without it and that's only good thing. I now that if it's constantly updated it's better than if it's not updated. Wanting it to go away, in the way that you guys seem to want it to go is just as "evil" and nonsensical as wanting the gun I mentioned above to dissapear before the shot that saves those innocent lives has been made.
Glide was abandoned when relevant GPU makers appeared, that used non-vendor-specific APIs. It was abandoned when PowerVR, nVidia, ATI, S3, Matrox, Rendition, etc started to take a significant share of the consumer 3D graphics card market, and 3dfx wasn't the stand-alone king of 3d accelerators anymore.
It had nothing to do with technological advance or OpenGL\DirectX catching up in terms of features <- that's what you fail to understand.lolAnd 3DFx was the first with DirectX drivers, despite having Glide. And Creative was the first with OpenAL drivers, despite having EAX.
Point being?
I find it funny you name 3dfx as irrelevant GPU maker. It shows your ignorance on the matter...POint is that without their support they would have never been posible. You are asking Nvidia to support OpenCL and they are doing it, more so and faster than AMD, S3 bla bla bla. So your point of Nvidia not supporting OpenCL is moot.
NViDIA Developed CUDA to sell More cards
NVIDIA did not develop CUDA to Heal sick puppies - that is the Job of the people buying their cards
Developing CUDA for ATI cards would cost money and have no benefit to them
They were also the first to support open CL - proving the first point, they don't care about the amount of people using CUDA, they just want to use GPU processing as an advertising point
No open CL programs are not NVIDIAs fault, it is the programmers not writing programs for them
In the future new programs may use open CL and existing programs may move over to it, this will depend entirely on whether the programmers feel the extra 30% of pc users they will reach) outweighs the effort of recoding or coding in CUDA
store.steampowered.com/hwsurvey
Most programs take about 2-3 years to make
I heard OpenCL was a Mac project, how come it became opensource? Perhaps Nvidia will make CUDA opensource sometime in the future, eh?
It had lower performance, fewer features (16-bit color render only) and higher price than TNT2. Voodoo 3 sales were abysmal, as people preferred TNT2 for performance and Matrox G400 for image quality.
Add that to the fact that 3dfx had just made their graphics card production exclusive to STB and the Voodoo 3 was the beginning of 3dfx's downfall.
As Î said, Glide had to go when several competing (and competitive) GPUs appeared in the market that couldn't support it.
When GPGPU was a thing that only nVidia poured real money into (even though ATI actually started it with R520's close-to-metal), CUDA may have had some sense.
Now we have nVidia with 8xxx onwards, ATI with HD4000 onwards, the whole PowerVR SGX line + future series 6, ARM Mali T604, Vivante GC400-4000, Broadcom Videocore IV, S3 5400E and many others with full OpenCL compatibility.
It's time to let CUDA go and invest into OpenCL.
Bottom line what cuda brought of real value to the world ? in games was physics and in content creation was faster transcoding speed , look at intel's sandy and the speed at wich it transcodes compared to cuda or amd stream , it's not even funny how it decimates them , that parallelization probably will work in physcs too ( intel having havok ) if this will be of any interest for intel.
When the integration of graphics in cpu's was promised i was thinking of cheap tiny laptops that consume little power but it seems more can be gained from this , intel already developed a plugin for adobe for using quick sync , maybe it's not ready as it should be cuda took a lot of time too to mature , i believe intel will mature this tehnology cause they have the money , amd is a lost hope of course.