NVIDIA CUDA
8 Nov 2006 22:59:39 UTC
Topic 192046
(moderation:
NVIDIA tries to make GPU programing much more easy.
Does the new NVIDIA G80 GPU (GeForce 8800) have enough mathematical features for Einstein@home?
http://developer.nvidia.com/object/cuda.html
Technology Features
• Unified hardware and software solution for thread computing on CUDA-enabled NVIDIA GPUs
• CUDA-enabled GPUs support the Parallel Data Cache and Thread Execution Manager for high performance computing
• Standard C programming language enabled on a GPU
• Standard numerical libraries for FFT and BLAS
• Dedicated CUDA driver for computing
• Optimized upload and download path from the CPU to CUDA-enabled GPU
• CUDA driver interoperates with graphics drivers
• Supports Linux and Windows XP operating systems
• Scales from high performance professional graphics solutions to mobile and embedded GPUs
• Native multi-GPU support for high density computing
• Supports hardware debugging and profiler for program development and optimization
NVIDIA CUDA
)
Thanks for the hint.
CUDA sounds interesting. Not for the current code, however, but for the next generation "hierarchical" code we're currently working on. I was thinking about using BrookGPU, which is a bit more general.
I'm not sure about the features of the G80, and for us it's a bit too early anyway to say anything about specific GPUs. I probably won't come to make use of them long before a year or so from now.
BM
BM