Your Best Source for CAD, CAM & CAE Articles

| About | Submit Article

CAD, CAM & CAE Articles



More News



  Independent Content    

Industry-Experts Cover
Products & Services You
Need to Know About.
Only on
Find Out More!

Q&A: AMDs' Graphics Business Unit and its New Graphics Boards

By Ralph Grabowski, Sep 24, 2012

Allen Bourgoyne is director of AMD's professional graphics ISV alliances. CADdigest's Ralph Grabowski interviewed him about the history, heterogeneous computing, the clouds, and the company's new line of FirePro graphics boards.

Allen Bourgoyne

Q: ATi began as an independent graphics board (GPU) designer in Canada, and then AMD bought the company. About a year ago, the ATi name was dropped in favor of the AMD name. How do you differentiate between the AMD CPU and ATi GPU teams?
A: They keep doing the same thing as before the acquisition, but we do get those smart people together to figure out how to make CPUs and GPUs work in a single chip to create the accelerated processing unit, APU. The latest APUs, the "Trinity" family of APUs, came out this summer.

Q: Does the GPU division have its own name now?
A: Within AMD, it's the Graphics Business Unit.

Q: GPUs are slowly becoming a more important part of the computing capabilities available to CAD users. What are some of the things AMD is working on in this area?
A: GPUs don't behave like CPUs. This means that GPUs cannot switch between tasks, like CPUs; they are good at performing tasks quickly, but have to be dedicated to one at a time; they don't have the ability to do the kinds of time slicing between tasks like CPUs can.

The latest GPUs from AMD, those available on the W8000 and W9000 FirePro cards, are able to run a graphics task and up to two additional compute tasks simultaneously, so GPUs are getting better at running multiple tasks. At the AMD Fusion Developer Summit (AFDS) back in June, the company described the Heterogeneous System Architecture (HAS). The long term vision for HSA is to have CPUs and GPUs more tightly integrated, where they share the same address space and GPUs support preemption so that they can be full partners in the system, allowing for the system to schedule code to run where it runs best, CPU or GPU.

Q: Will this be transparent to the user? What I mean, does CAD software need to be rewritten to handle HSA?
A: It should be transparent; the user notices only that the system operates faster and more efficiently. There might be controls that allow users to help determine whether to target the CPU or GPU for specific workloads, but it is too early to know exactly how details like this will work for users.

We plan to develop tools for software developers to help them easily access the benefits from HSA. Compilers might generate multiple binaries, or else the operating system might include support for examining CPU vs GPU loads, and then acting accordingly. These are just a couple of examples of this.

Q: One of the problems of using GPUs for accelerating computations is that you and nVidia are taking somewhat different approaches.
A: This is why we push for OpenCL [open computing language], instead of having a proprietary compute technology like nVidia's CUDA. [nVidia also support OpenCL. - Ed.] The HSA Foundation is a consortium consisting of many members interested in building a heterogeneous compute ecosystem. You can read more about the HSA Foundation at

Q: What kind of an effect is the cloud having on AMD?
A: There are lots of GPUs in servers. Graphics are used for online rendering, and there is software running on the servers that allows multiple users to access the server: it passes the jobs to the GPUs to do the rendering, and returns the completed job to the user.

For CAD, the holy-grail is real-time iterative design, running through the design and analysis phases quickly in order to converge on an optimal design in a short period of time.

In the old days, an engineer would do a design and then hand it over to the FEA [finite element analysis] team to run an analysis job on a server, and then might get it back tomorrow - if they weren't too busy. With most users having multi-core CPUs and GPUs in their workstations, analysis software like ANSYS and Abaqus can run on the desktop. So the goal is to work on the model and have analysis running at the same time - a good candidate for the cloud.

The idea is infinite rendering power and infinite computing power enables designers to bring design and analysis closer together than before.

Q: I notice that private clouds are quite cheap, like $350 for the box and software, not including multiple hard drives. Do your graphics boards get used on these?
A: Our boards are also used by private clouds. Private clouds are popular, because it provides greater security than on third-party clouds. Even if the security is great, companies get nervous about keeping their data on someone else's computer. So for some companies public clouds are the right solution, for others, they are choosing private cloud-based solutions.

Q: How big is the cloud apart of your business?
A: From a professional graphics perspective, is still a small part of our business but growing; I think businesses aren't moving as quickly as some might want; they are preparing for it, not stampeding towards it. Online gaming is big; in China, it is growing quickly.

 Q: What's new for your FirePro [FirePro W5000 through W9000] graphics boards?
A: The latest FirePro boards feature our GraphicsNext architecture; you don't want any idle cycles, and so the new architecture does a good job at keeping the GPU busy.

The new paradigm is to minimize $/W/sq mm - that's dollars per Watt [of power consumed] per square millimeter of graphics board - and so we looked at how to build them more efficiently and subsequently cost the customer less.

This summer we refreshed our mid-range models and up. We doubled the geometry throughput; our previous boards processed one triangle per clock click; these new ones process two triangles in the same time. We upped the processing power of the GPU from 2.5 teraflops to just under 4. And we increased the memory on the boards; they now start at 2GB and go up to 6GB.

We are working with ISVs (independent software vendors, like CAD companies) to certify the new FirePro boards as quickly as possible; a large number of third-party applications are already certified. The ISVs do the actual certification of the boards and drivers themselves, and we work with them to complete the tests as soon as they can. You can find the complete list of certified applications and drivers on the AMD website:

Back to Top

Additional Information

Back to Top

About the Author


Ralph Grabowski, TenLinks senior editor, is one of the leading CAD journalists and authors, with over a 100 books and many hundreds of articles. His upFront.eZine may be the industry’s longest running newsletter. Ralph holds a civil engineering degree. More…

About Independent Coverage

TenLinks uses over 125 expert authors for unprecedented coverage of CAD, CAM, CAE products and services. Find out more about Independent Content and apply for coverage.