NEWS         TESTS         INTERVIEWS         DOSSIERS         GALLERY 3D         3D SITES
 Q&A

AMD GPU Computing,
Gary Silcott,
September - 2009

 

“ In the CPU you ask a computer a lot of different questions rapidly one after the other vs. asking the GPU the same question many, many times for a different set of data. Both types of workloads are found in software today, therefore we need to have both of CPU and GPU processing together to deliver the optimal user experience.”

   
Q1 Are ATI prodcuts ready to take advantage of GPU power for general purpose computing?
A1

Absolutely, and ATI has been a vocal proponent of GPGPU using open standards rather than proprietary solutions, so the recent developments around OpenCL and Direct X 11 reinforce that strategy.

   
Q2 In the case of GPU computing, what is the role of the CPU? Does the CPU affect the performances?
A2

This is exactly the right question to be asking because the CPU remains key to AMD’s focus on platform performance rather than the individual components! The CPU and GPU are like co-workers who get along fabulously and get a lot more done working together, but they use completely different styles to achieve their goals. In the CPU you ask a computer a lot of different questions rapidly one after the other vs. asking the GPU the same question many, many times for a different set of data. Both types of workloads are found in software today, therefore we need to have both of CPU and GPU processing together to deliver the optimal user experience.

   
Q3 AMD was one of the first company to take advantage of Apple OpenCL.What are the advantage of OpenCL Library vs Cuda and Stream SDK?
A3

(note: OpenCL is a Khronos run standard that Apple, AMD and others support – it’s not “Apple OpenCL”.)

Another very important question, because OpenCL (along with the DirectCompute functions in DirectX 11) really do change the game in GPGPU. OpenCL allows programmers to target multi-core and heterogeneous chip architectures more easily and only write their code once to access a variety of hardware. CUDA is proprietary to Nvidia hardware, so it isn’t particularly us eful in a world where ATI holds significant GPU market share. OpenCL is also supported on multi-core CPUs, so AMD’s position as the only supplier of both GPUs and CPUs on our PC platforms make OpenCL a great way for developers to access both.

   
Q4 Windows 7 will introduce DirectCompute : are ATI cards ready to take advantage of this new library?
A4

Absolutely and as the only supplier of DX11 hardware that is supported by Windows 7 AMD has a significant advantage over the competition.

   
Q5 It seems that nVidia create proprietary solutions under CUDA : PhysX, 3D renders (Interactive RayTracing, V-Ray RT...). Do you think that users will prefer to choose nVidia cards to be more compatible with GPU Computing?
A5 Absolutely not, developers don’t want to have to chose one hardware over another, they want to run on as many systems as possible, and that’s why AMD has consistently worked to have ATI Stream support open standards such as OpenCL for CPU and GPU, as well as DirectCompute under Windows 7.
   
Q6 Tim Sweeney (Epic Games) predict the end of GPU in the next Decade, saying that CPU will definitively surpass GPU for creating next generation realtime 3D engines (Ray Tracing). What do you think of this affirmation?
A6 Obviously we disagree and in fact, just the opposite is true. The GPU is becoming more critical to computing not less, and is going to play a crucial role in delivering the best experience to the PC user.
   
   
   
 
  Search on 3d-test with Google
      A PROPOS DE 3D-TEST            INFORMATIONS LEGALES               LINKS              TRANSLATE WITH GOOGLE
 
© 2001-2007 3d-test, Panorama of web 3D technologies, 3d-test is edited by Westimages Realtime 3D Solution Provider