A 5-year vision for workstation graphics performance

by Bob Cramblitt

Views of SPEC/GWPG members.

 

Every day in their engineering jobs with leading OEMs and graphics card manufacturers, the representatives of the SPEC Graphics and Workstation Performance Group (SPEC/GWPG) assess the latest technologies designed to make their products run professional applications faster and more efficiently.

In their role with SPEC/GWPG, they take that knowledge and develop benchmarks that test real-world performance in ways that are fair and repeatable across different computing platforms.

All this experience makes them uniquely qualified to discuss what factors will influence workstation performance over the next five years.

If correct, their projections will likely have big implications in the years ahead, enabling new processes for applications such as machine learning, generative design, 3D printing, and AR/VR. The developments they foresee will also spur greater productivity in existing applications such as design, engineering, 3D animation, visualization, and special effects.

Let’s take a look at what the people who help define workstation performance see in the half-decade ahead.

Rendering advances will spur greater productivity in areas such as product development, 3D animation, and visualization.

Rendering leaps ahead

Allen Jensen, vice-chair of SPEC/GWPG and SPECapc, sees realtime rendering taking off in a big way over the next five years.

“Ray tracing in realtime is exciting,” says Jensen. “Just like today’s engineers turn their noses up at the idea of designing in wireframe, soon enough people will expect to design with a beautiful ray-traced model that’s generated almost instantly.”

 The real-world impact

Trey Morton, SPECapc chair, sees rendering advances having a major impact on productivity for companies that rely heavily on computer graphics.

“I think as speed continues to increase, productivity can follow, and tangential technologies can really take advantage of this,” he says.

“Given an increase in productivity, a designer can create more iterations of products, and then make physical models via 3D printing and other additive manufacturing processes or virtual product walkthroughs via AR/VR.”

“The end result is getting a solution approved and to market faster than before. The challenge for SPEC is to continue working toward realistic and representative benchmark workloads that characterize these fast-moving advances being implemented into professional applications.”

Balancing the load

The ancient argument pitting CPUs against GPUs is all but dead, but it’s still not easy to develop apps that distribute workloads effectively between these two types of processors. Ideally, CPUs and GPUs should be able to interoperate on the same code better than they do now.

Over the next five years, Anshu Arya, SPECwpc director of compute research, sees CPU and GPU computing breaking through traditional barriers.

“The interplay between CPU and GPU will become more interesting in the upcoming years,” says Arya.

“GPGPUs spurred a lot of developers to offload computation to the GPU, but with 64 or more cores on the CPU, software should aim to balance the workload across the GPU and CPU to tap the full potential of a system.”

Eliminating bottlenecks

Jon Konieczny, SPECgpc vice-chair, agrees with Arya and believes one of the keys will be adopting new APIs for professional applications.

“Most workstation applications tend to trail a bit behind the gaming industry in terms of their usage of the most modern hardware capabilities and programming APIs,” says Konieczny.

“As workstation vendors gradually shift their rendering engines over to these new APIs such as Vulkan, DX12, or more modern usage of OpenGL, the hardware bottlenecks in these applications can be reduced dramatically. This could have a large impact on the end user’s experience, as larger models with better graphical effects can be shown while still maintaining a high degree of interactivity.”

Emerging design approaches such as generative design will provide challenges for benchmarking workstation performance based on professional applications. (Source: Siemens).

Measuring a non-fixed outcome

Allen Jensen sees new applications being enabled by more functionality migrating to graphics cards.

“Graphics cards are big math engines and clever people are figuring out how to do deep learning on graphics cards and bring it down to engineers’ desks. The result will be the ability to harness more compute power for applications such as generative design.”

The challenge for SPEC/GWPG is how to measure performance for an application such as generative design, which enables engineers to create high-performing design alternatives from a single idea. There’s no longer a single solution, but multiple ones that can be tested and tweaked to arrive at an optimal design.

“We’ve been worrying for years about how to compare performance for problems that have two or more different outcomes or those that involve an iterative process for which there is no exact design template,” says Jensen. “How do you benchmark generative design when the end result is different every time”?

The answer, says Jensen, might be developing tools that enable users to extract performance results based on their own unique processes.

“It might require putting tools into engineers’ hands that allow them to insert a model into a benchmark structure, run their typical operations, and get a result that is specific to what they do on a day-to-day basis.”

Welcoming the challenge

Ross Cunniff, SPECgpc chair, acknowledges the challenges new types of application processes bring to performance measurement and welcomes the challenge.

“This is perhaps the fastest-moving time I’ve experienced as a computer graphics engineer,” says Cunniff, who has more than 35 years of technology experience spanning systems software, device drivers, and computer graphics.

“The advent of ray tracing in realtime will have an enormous impact on how computer systems are designed, and on how people interact with them. In addition, machine-learning algorithms are starting to influence graphics work. This cycle is accelerating and will lead to a very different graphics landscape in five years.”

Cunniff thinks that SPEC/GWPG will provide guidance the way it always has: By giving professionals the tools to discover how these types of technological developments impact graphics and workstation performance in the real world.

Bob Cramblitt is communications director for SPEC. He writes frequently about performance issues and digital design, engineering and manufacturing technologies. 

Want to stay informed about graphics and workstation performance? Visit the SPEC/GWPG website, subscribe to the SPEC/GWPG enewsletter, or join the Graphics and Workstation Benchmarking LinkedIn group: https://www.linkedin.com/groups/8534330.

 A reminder: SPEC/GWPG benchmarks are available for free downloading to everyone except vendors of computers and related products and services that are not members of the SPEC Graphics and Workstation Performance Group (SPEC/GWPG).