Event Coverage

NVIDIA's Accelerated CUDA Momentum

By Vijay Anand - 22 Sep 2010

The Power of GPU Computing

The Power of GPU Computing

It is only in recent times when the savvy PC users are beginning to realize the usefulness of the GPU outside of gaming. Think about hardware assisted HD video playback or hardware assisted video transcoding - now those are some really useful aspects that take a big load off the CPU and effectively complete the tasks in far less time. Increasingly, more content creation applications like Adobe Photoshop CS5 and Premiere Pro CS5 are riding on the power of the GPU to also speedup on the effects processing while leaving the CPU more headroom for other tasks.

As far as the consumer space is concerned, there's still much more room for software development to harness the GPU. The average Joe on the street is still more than content with integrated graphics solutions, while most common software and usage needs are served fine by the CPU alone. However, this is about to change as even web browsers and web content are gearing up for the next experience level with the power of the GPU. We'll soon touch on that aspect, but for many of us, we often forget to think about the medical, scientific, education, R&D sectors where the power of the GPU is immensely helping progress our society. This is where NVIDIA's CUDA architecture present in their GPUs for the last few years has played a significant role in this arena, along with the software team at NVIDIA who engage and help developers from all levels to better integrate GPU computing into their workflows. This year's GPU Technology Conference started out highlighting the CUDA momentum and adoption and here's the opening video clip that further reinforces this with NVIDIA CEO Jen-Hsun Huang kick starting the event in proper. Just take note that it was shown in 3D, hence the video might lack in clarity, but its main purpose is evident:-

Jen-Hsun showed off this GTC statistics slide comparing last year to this year's interest in GPU Computing and just how many people are getting on the bandwagon to unlock the GPU to accelerate and innovate.

With the first two pillars of science being theory and experimentation, computing is fast becoming the third pillar of science where the GPU is making big inroads to finding solutions for problems.

At the GTC 2010 event, NVIDIA along with its partners made a few big announcements in the CUDA/GPU Computing space. The first is the soon to be launched CUDA-x86 via PGI CUDA C, an x86 compiler for CUDA C. The Portland Group today announced that it's developing this to target the industry standard 32-bit and 64-bit x86 platforms.

PGI's CUDA C compiler for x86 platforms will be available commercially in a couple of months and this is really big news.

So while previously developers can only rely on computing clusters with NVIDIA GPUs, they can now do so effortlessly on any system, regardless of the availability of a GPU. Hence when run a system without a GPU, PGI CUDA C will use multiple cores and the SIMD capabilities of Intel and AMD processors for parallel execution. While the performance is not expected to be anywhere like that of a system with a GPU, it will at least raise yet another barrier and promote even wider use and development of CUDA capable solutions for high performance computing environments. Primarily, with CUDA for x86, The Portland Group is addressing developers who want to use a single parallel programming model to target many-core GPUs and multi-core CPU processors. This will be made available sometime in November.

MATLAB, often fondly referred to as the Excel equivalent for Engineers and Scientists, has now announced the availability of a CUDA-Accelerated Parallel Computing Toolbox. While there are already several other toolbox and toolkits to exend MATLAB's functionality, including a parallel computing toolkit, the CUDA accelerated version will now allow any engineer to tap in to the power of the GPU to speed up technical engineering aspects without explicitly coding for it. More cool announcements below:-

GPU computing for anyone dealing with math is now unleashed with MATLAB's availability of a CUDA-accelerated parallel computing toolbox.

Amber 11, a molecular simulations package, now supports multi-GPUs and is of course CUDA enhanced as well.

Using the JAC benchmark (a molecular modeling benchmark) in conjunction with Amber, NVIDIA showed off that an IBM iDataplex cluster with just 8 Fermi-class GPUs was able to more than rival a 192-node partition of quad-core processors off the Kraken supercomputer.

The Kraken is actually a monster machine cluster with a total of 99072 AMD Opteron-based CPU cores, but even so, tasking 192 of its quad-core processors and yet losing out to 8 GPUs is quite telling of the power of GPU computing.

ANSYS is the foremost simulation driven computational developmental tool and one of the tope simulation packages for engineering. Used by most of the large industrial companies of the world, today, they've announced a CUDA-accelerated ANSYS Mechanical R13 package.

Dr. Subbiah, the VP for Global Business Development for ANSYS, shared with the audience how they saw a 2x improvement in performance speedup in their trials. Now 2x might not sounds a lot, but when considering a job takes several days to weeks typically, a 50% reduction is a great deal of time savings.

Yet another cool CUDA-related announcement is what 3ds Max users will soon get to experience. 3ds Max is of course the leading choice of simulation and rendering software for architects, game studios, animation studios and more. As part of the 3ds Max 2011 Subscription Advantage pack made available to its subscribers, it will incorporate the GPU-accelerated iray renderer from mental images. This is certainly exciting news as we've seen in last year's GTC event just how capable iray is and it being incorporated in 3ds Max only means that a lot of users are going to benefit greatly from its fast photorealistic rendering solution with the aid of an NVIDIA GPU. We've a video walkthrough of the iray renderer at work as opposed to CPU-based ray tracing. Additionally, the folks at 3ds Max even managed to demo GPU cloud computing to manipulate complex works instantly on the browser. You've definitely got to see this in action:-

With these new announcements of CUDA accelerated applications from its major partners, NVIDIA now proudly proclaims that they've put the power of CUDA within reach of every engineer, designer and researcher in the world. With the kind of titles and companies involved, we can't deny them of those claims.

Join HWZ's Telegram channel here and catch all the latest tech news!
Our articles may contain affiliate links. If you buy through these links, we may earn a small commission.