Report from the NVIDIA GPU Technology Conference (GTC)
Greetings from NVIDIA's GTC 10! While the last year in NVIDIA GPU computing was about Tesla Fermi and hardware, this year will prove to be about software. NVIDIA has made a number of exciting announcements:
- MATLAB GPU support is here! Mathworks has been working hard to accelerate MATLAB using CUDA. This offers an exciting performance leap for the millions of MATLAB users out there.
- Enhanced GPU support in AMBER 11. Now multiple GPUs can be used to accelerate an AMBER run.
- GPU acceleration has been announced for the most common engineering design application: ANSYS Mechanical. AutoDesk's AutoCAD and 3ds max products have also added CUDA support. Combine these with NVIDIA's 3D Vision Pro for immersive realistic 3D environments. Cloud rendering in 3ds max with iray ray-tracing was demonstrated during Tuesday's keynote presentation.
- Using Wolfram's Mathematica software? CUDA GPU support is coming there too.
- CUDA-x86 compilers through NVIDIA's collaboration with PGI: this is great news for developers wanting to program in CUDA, but still waiting for GPU hardware. Design and code highly parallel applications in CUDA, compile with PGI, and the application will run on your x86 CPUs today. The same application will accelerate on your GPUs tomorrow. CUDA-x86 should also allow for easier collaboration with colleagues without GPU hardware: they can still participate in coding and debugging CUDA applications. Pricing and availability information to come.
- NVIDIA has announced the roadmap and code names for their next two GPU architectures: Kepler in late 2011 and Maxwell in 2013. Expect 3-4x the performance per watt on the 28nm process for Kepler and better programming features like virtual memory. Maxwell will deliver up to 11x the performance per watt.
View a recording of NVIDIA's keynote presentation
here.
Thank you to all the customers who are stopping by our booth. We're looking forward to seeing more of you throughout the GTC event - visit us at Booth #105, and at Booth #4113 at SC10 in New Orleans.
Also, stop by our booth to see a preview of Microway's OpenCL IDE (CLIDE). CLIDE empowers users to easily develop, test, and deploy OpenCL kernels.
-- from Brett Newman at NVIDIA GTC 10
Latest Microway GPU Offerings
Tesla S2050 GPU Compute Servers
- 4 GPUs in a 1U footprint: total of 2 TFLOPS double- and 4.13 TFLOPS single-precision floating point performance
- Total of 12 GB GDDR5 memory (4x 3GB GPUs) with optional ECC
- Available now - lead time varies, depending on quantity and configuration
Tesla C2070 GPUs - Now Shipping!
- 448 CUDA cores provide 515 GFLOPS double- and 1.03 TFLOPS single-precision floating point performance
- 6 GB GDDR5 memory with optional ECC
- Same power envelope as Tesla C2050 GPU
- Order now!
Microway OctoPuter 4U Server with 8 Tesla GPUs
- Highest GPU/CPU ratio available
- Eight NVIDIA Tesla GPUs - Over 8 TFLOPS single-precision performance
- Two Quad- or Six-Core Intel Xeon CPUs
- Up to 144GB DDR3 memory
- Optional 40Gbps QDR InfiniBand
