NVIDIA Dramatically Simplifies Parallel Programming With CUDA 6

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
NVIDIA today announced NVIDIA® CUDA® 6, the latest version of the world's most pervasive parallel computing platform and programming model. The CUDA 6 platform makes parallel programming easier than ever, enabling software developers to dramatically decrease the time and effort required to accelerate their scientific, engineering, enterprise and other applications with GPUs. It offers new performance enhancements that enable developers to instantly accelerate applications up to 8X by simply replacing existing CPU-based libraries. Key features of CUDA 6 include:

  • Unified Memory -- Simplifies programming by enabling applications to access CPU and GPU memory without the need to manually copy data from one to the other, and makes it easier to add support for GPU acceleration in a wide range of programming languages.
  • Drop-in Libraries -- Automatically accelerates applications' BLAS and FFTW calculations by up to 8X by simply replacing the existing CPU libraries with the GPU-accelerated equivalents.
  • Multi-GPU Scaling -- Re-designed BLAS and FFT GPU libraries automatically scale performance across up to eight GPUs in a single node, delivering over nine teraflops of double precision performance per node, and supporting larger workloads than ever before (up to 512GB). Multi-GPU scaling can also be used with the new BLAS drop-in library.
 
So wait, is this an API like Mantle? I feel like this describes exactly what I heard in the AMD keynote.
 
So wait, is this an API like Mantle? I feel like this describes exactly what I heard in the AMD keynote.
No, it's a feature like hUMA like in AMD GPUs. It allows a programming model where local device memory (on a discrete GPU) and system memory are treated the same without explicit system memory -> GPU and GPU -> system memory code.

Mantle is a low level graphics API, made to work more efficiently than the D3D and also exposes more hardware features directly.
 
I love both NVIDIA and AMD cards, but would choose an AMD card. However, because industrial applications are CUDA-heavy such as Adobe After Effects and Premiere Pro, my next card will be an NVIDIA. :(
 
long story short even though amd opened the door to nvidia using the mantle api nvidia has slammed it shut
 
long story short even though amd opened the door to nvidia using the mantle api nvidia has slammed it shut
lol, the level of complete ignorance of the topic at hand is astounding. Congrats!
 
long story short even though amd opened the door to nvidia using the mantle api nvidia has slammed it shut
lol, the level of complete ignorance of the topic at hand expressed in that post is astounding. Congrats!
 
Lolnope try again
I have an AMD card, and Adobe After Effects renders at less than 1.0 frames per second.

On top of that, I have an i7-3930K clocked at 4GHz, 32GB DDR3-1600 RAM, running the program, cache, and video files purely off of SSD. Windows 7 Professional x64.
 
I still don't completely understand why we can't use GPUs as CPUs on virtual hosts.
Software incompatibility is a huge barrier. ;) The closest to using GPU-type cores that way is Xeon Phi, which can run an x86 OS on each of its dozens of x86 cores.

While GPUs have been adding capabilities to what type of code it can execute (branches and reconvergence are biggies), those are a lot less capable than a desktop CPU. GPUs are great for graphics, or using its arrays of ALUs to work in parallel on certain types of problems. You may be able to write a simple type of OS that runs on GPUs, but it's kind of pointless since you'd need to run applications to do anything useful. And you could write useful applications that directly target the GPU using CUDA, OpenCL, DirectCompute or whatever.
 
Back
Top