Q&A with Rys Sommefeldt, Imagination Technologies
Addressing the newest power and performance requirements
Editor’s Note: Imagination Technologies has announced its PowerVR Series8XE GPUs. Rys Sommefeldt, the company’s Business Development Engineering Manager for PowerVR, commented for EECatalog recently about the announcement, Vulkan 1.0, security, and other topics.
EECatalog: What is it important for us to know right off the bat about Imagination’s PowerVR Series8XE GPUs?
Rys Sommefeldt, Imagination Technologies: Series8XE is PowerVR doubling down on improving the PPA—Performance at a given Power and Area—of the GPUs we offer to our cost-focused customers. The driving force behind the XE branch of our Rogue family of GPUs has always been performance per unit of silicon area and performance per unit of power, mixed with just the right feature set that the market needs. That lets us keep area—the primary cost for an XE customer—to a minimum and focus on the best performance at that given size. Our customers want the most for their increasing investment in silicon area. Series8XE represents our best knowledge and expertise about how to make that come true for them.
EECatalog: How do these Series8XE GPUs compare to Imagination’s 7XE series?
Sommefeldt: Building on what I said earlier, to deliver a strong evolution of our guiding principles for the XE range of GPUs, we’ve doubled the pixel and texture fillrates for each configuration, versus the equivalent 7XE configuration. We’ve also adjusted the balance of the ALUs in the core to focus on what’s most important for the primary uses cases. 8XE GPUs also have a focus on delivering very strong 16-bit floating point ALU performance. Having an additional 16-bit ALU pipeline lets us pack more of those pipelines into the equivalent area we might otherwise dedicate to 32-bit hardware, just by virtue of it being a lot smaller—giving us more performance for that area and satisfying one of the key mantras of the XE range. Combined together, 8XE GPUs lets us deliver both a huge increase in fillrate, which is important as the demand for screen resolutions keep going up, and a very strong ALU performance to let us render all of those new pixels at much higher power efficiency.
EECatalog: The Khronos Group has announced the Vulkan 1.0 open standard API specification. Is Imagination in a good position vis-à-vis the leap to Vulkan?
Sommefeldt: We designed 8XE with Vulkan at the forefront of our thinking—so we’ve been in good shape since the specification was first being developed at Khronos. The development of 8XE and the genesis of Vulkan at Khronos have happened in a similar time period, allowing us to react to changes in the specification and make sure 8XE delivers great performance when being driven by the new API.
Vulkan is a great display of what happens when there’s a will to better the state of the art in GPU microarchitecture and design—especially as it enables the trend for GPUs to be tile-based and offers a new take on how the GPU should be programmed by application code.
EECatalog: Why the need for Vulkan?
Sommefeldt: The APIs that came before Vulkan were mired in legacy. They were designed to drive GPUs that thought differently about how to solve the problem of rendering. The old APIs were developed before GPUs were deployed in a more constrained environment like the smartphone, where power, size and memory bandwidth are extremely limited, and where the accompanying CPUs are subject to the same limitations. APIs like OpenGL ES didn’t really consider their CPU cost and the fact that GPU microarchitecture would diverge over time.
Vulkan tries to solve both these problems. The at-runtime CPU cost of calling the Vulkan API to drive the GPU is almost zero, saving power and letting more of the CPU be available for the rest of the applications. Vulkan is also aware of how modern GPUs work, and crucially makes it explicit to the programmer so they’re forced to think about it. This pays dividends in terms of extracting the most efficiency out of the GPU. Vulkan is long overdue in those twin respects!
EECatalog: What problem or problems does the introduction of these GPUs solve?
Sommefeldt: The main problem being solved by the new 8XE range is addressing the newest requirements in terms of performance and power, while keeping costs low for our customers. 8XE will see common deployment in chips driving 4K displays in TV and other consumer device classes.
With the advent of the latest manufacturing technologies, silicon costs are going up; yet end users demand generational improvements along with lower costs. So it’s the job of the 8XE range to let our customers address those downstream demands, while delivering cost benefits.
Our PowerVR GPUs have always solved the problem of bandwidth efficiency because of the unique nature of our tile-based deferred rendering microarchitecture. So our focus in the XE range builds on that capability to deliver great bandwidth-efficient performance at low power.
EECatalog: Security of the GPUs is one of the features you are emphasizing. Has Imagination built a better mousetrap with regard to protecting data?
Sommefeldt: Adding robust and usable security features and secure execution ability to the GPU has actually been one of the biggest engineering challenges we’ve faced in a long time. The class of problems you need to solve there aren’t trivial, and it’s much easier to design a GPU that doesn’t have to care about what’s executing. In contrast, a secure GPU needs to make sure that what’s executing, and crucially the memory that’s read from and written to by the GPU while that happens, is protected and isolated from everything else in the system. It has to do that by working closely with the CPU and the other elements of the system’s memory hierarchy.
The 8XE is able to achieve that level of security by supporting multiple secure contexts simultaneously, all isolated from each other. Our unique take on driving a GPU using our firmware design and driver architecture helps make that possible, and we wrap it up under the marketing umbrella of OmniShield, which is complimentary across our IP families and not just limited to the GPU product line. We learned a great deal about how to do security well in the last generation, and we’ve taken all of those lessons and applied that to the 8XE design. We’re happy to claim it has best-in-class security features as a result.