Sunnyvale (CA) – AMD unveiled a new stream processor card and announced a new stream software development kit as well as a new Catalyst driver update that promises showcase the floating point horsepower that is currently hidden in the firm’s graphics cards. There is also a new Avivo video converter that will convert videos 17 times faster than before, AMD said.
Stream processing and floating point acceleration is finally making visible steps in consumer application steps. A few weeks ago, Adobe began showcasing GPU acceleration in its Photoshop and Premiere CS4 software packages and it seems that AMD is now ready to also highlight its general purpose GPU (GPGPU) capability not only in scientific and industrial applications, but in actual consumer products.
Catalyst 8.12, due for a December release, will be released in tandem with a new and free Avivo video converter tool that will leverage the GPU to convert videos between different formats. AMD promises a 17x speedup and claims that a HD video that took 3 hours to convert will now be converted in just 12 minutes on a Radeon HD 4850 card.
To highlight a greater focus on stream processing, AMD introduced a new brand “Stream”, which will appear on every product that includes or supports the technology – such as upcoming Catalyst drivers and the firm’s graphics cards. 4000 series graphics cards already include all the components required for AMD’s latest stream processing move; 3000-series cards support the technology, but will require extra software components AMD will offer for download, we were told by the company.
AMD also puts a greater effort behind supporting the development of stream applications by providing a new version of its stream software development kit. Version 1.3 includes AMD’s compute abstraction layer (CAL), support for the C++-like high-level programming language Brook+ and tools, libraries as well as middleware. While Brook+ is to AMD what CUDA is to Nvidia, both companies are also supporting OpenCL as an open, standardized platform for the development of floating point accelerated applications.
And just like CUDA, Brook+ and OpenCL do not quite provide a free lunch for developers. While it is considered to be fairly easy to translate existing applications into versions that take advantage of the GPGPU (at least if a developer has sufficient knowledge of parallelizing applications), the fine tuning of software is difficult. The final 10-20% of performance can only be tapped with extensive programming and GPU design knowledge. AMD confirmed that Brook+ will also reward those developers with GPU design knowledge and put those without this knowledge at a certain disadvantage.
AMD updated its Firestream product line with the new 4870 GPU-based Firestream 9270 card. Priced at $1499, it is much more expensive than a regular 4870, but the card includes 2 GB of GDDR5 memory and more capable cooling. The theoretical maximum single-precision floating performance is about 1.2 TFlops, but it is the double-precision horsepower that makes the card a standout: GPU typically take a significant hit when moving into double-precision applications – for example, Nvidia’s 1 TFlop Tesla card is rated at less than 100 GFlops in double-precision scenarios. AMD, however, promises a double-precision performance of 240 GFlops for its Firestream 9270 (and Radeon HD 4870 GPU).
Stream processing and floating point acceleration is finally making visible steps in consumer application steps. A few weeks ago, Adobe began showcasing GPU acceleration in its Photoshop and Premiere CS4 software packages and it seems that AMD is now ready to also highlight its general purpose GPU (GPGPU) capability not only in scientific and industrial applications, but in actual consumer products.
Catalyst 8.12, due for a December release, will be released in tandem with a new and free Avivo video converter tool that will leverage the GPU to convert videos between different formats. AMD promises a 17x speedup and claims that a HD video that took 3 hours to convert will now be converted in just 12 minutes on a Radeon HD 4850 card.
To highlight a greater focus on stream processing, AMD introduced a new brand “Stream”, which will appear on every product that includes or supports the technology – such as upcoming Catalyst drivers and the firm’s graphics cards. 4000 series graphics cards already include all the components required for AMD’s latest stream processing move; 3000-series cards support the technology, but will require extra software components AMD will offer for download, we were told by the company.
AMD also puts a greater effort behind supporting the development of stream applications by providing a new version of its stream software development kit. Version 1.3 includes AMD’s compute abstraction layer (CAL), support for the C++-like high-level programming language Brook+ and tools, libraries as well as middleware. While Brook+ is to AMD what CUDA is to Nvidia, both companies are also supporting OpenCL as an open, standardized platform for the development of floating point accelerated applications.
And just like CUDA, Brook+ and OpenCL do not quite provide a free lunch for developers. While it is considered to be fairly easy to translate existing applications into versions that take advantage of the GPGPU (at least if a developer has sufficient knowledge of parallelizing applications), the fine tuning of software is difficult. The final 10-20% of performance can only be tapped with extensive programming and GPU design knowledge. AMD confirmed that Brook+ will also reward those developers with GPU design knowledge and put those without this knowledge at a certain disadvantage.
AMD updated its Firestream product line with the new 4870 GPU-based Firestream 9270 card. Priced at $1499, it is much more expensive than a regular 4870, but the card includes 2 GB of GDDR5 memory and more capable cooling. The theoretical maximum single-precision floating performance is about 1.2 TFlops, but it is the double-precision horsepower that makes the card a standout: GPU typically take a significant hit when moving into double-precision applications – for example, Nvidia’s 1 TFlop Tesla card is rated at less than 100 GFlops in double-precision scenarios. AMD, however, promises a double-precision performance of 240 GFlops for its Firestream 9270 (and Radeon HD 4870 GPU).