AMD to integrate Stream GPGPU features into Catalyst 8.12

Posted at: 10:21am 13th November 2008

December driver update includes free GPGPU accelerated video transcoding app, and a speed up in Folding@home performance

ATI was originally ahead of the game in the land of GPGPU computing, being the first company to support a GPU Folding@home client long before Nvidia’s CUDA was released, but it’s recently fallen behind in the performance stakes. However, AMD has just announced plans to massively overhaul the GPGPU features of its Radeon HD 3000 and 4000 GPUs with a driver update in December.

Arriving on 10 December, Catalyst 8.12 will see the first integration of AMD’s Stream technology for mainstream GPUs. The main addition here is the integration of AMD’s Compute Abstraction Layer (CAL), which sits between the hardware and an enhanced version of Stanford’s Brook language, called Brook+. The launch will also coincide with the release of AMD’s Stream 1.3 SDK, which will enable software developers to use the features.

Speaking to Custom PC, AMD’s director of Stream Computing, Patti Harrell, explained that ‘the SDK 1.3 adds some performance enhancements, specifically around Brook+, and we have in fact completely rearchitected the Brook+ runtime to greatly improve the performance of the code that developers write. So, we have performance enhancements in Brook+, we have support for new hardware in that SDK and then the SDK comes out at the same time as the CAL integration in Catalyst.’

AMD will complement the driver release with a new Stream-accelerated version of its Avivo Video Converter app, which will be available as a free download. AMD is making some bold claims about the performance of the app, particularly in comparison with the CUDA-based Badaboom. AMD’s software product manager, Terry Makedon, said that ‘everybody knows about this Badaboom, Nvidia Elemental stuff. Nvidia obviously doesn’t support an application like we do – we’ve had this for many years doing it on the CPU – now we’re transferring it all to do on the GPU with very, very fast performance.’

‘Our initial results show that it’s faster than the Nvidia-sponsored solution of Badaboom,’ claimed Makedon, and added that ‘it’s also free,’ whereas Badaboom currently costs £19.70. AMD says that in its own tests, the new app took 12 minutes to convert an hour of 1080p video to a 320 x 240 H.264 file; a job that apparently took three hours, 23 minutes on a 3GHz Core 2 QX9650 CPU with the WinQuickTimeMPEG2 pack.

So, on to the question that many CPC readers will want answered – what will this mean for your Folding@home production rate? Speaking to Custom PC, Terry Makedon said that ‘with the new version of CAL there is going to be a little bit of a speed up [when folding].’ However, he added that ‘what we’re waiting for is an updated version of the actual folding client from Stanford.’ This, says Makedon ‘will have significant improvements.’ Basically, the new driver should speed up the production rate a bit on ATI GPUs, but a client that supports all the new Stream features should dramatically bump up the ppd.