AMD is developing a completely new type of graphics card, designed specifically for large dataset applications, including VR content creation, computational engineering and high-resolution rendering.
The AMD Radeon Pro SSG will feature a Terrabyte of Radeon Solid State Graphics, giving the powerful ‘Fiji’ GPU fast access to large datasets stored on an embedded NVMe SSD.
Current high-end GPUs are limited to a maximum of 32GB of GDDR5 memory. While this is plenty for many users, for more advanced visualisation and engineering workloads it is not possible to hold large datasets completely in memory.
Instead, slices of data are processed individually and later merged by software. This, says AMD, can result in significant latency as additional data needs to be fetched from system memory, via the CPU.
At the launch this week AMD demonstrated how 8K raw video timeline scrubbing could be accelerated from 17 frames per second to 90+ frames per second, a five-fold increase in performance.
However, readers of DEVELOP3D will be more interested in how the technology could be applied to virtual reality and CAE (simulation).
Limited memory has been a significant barrier to wider adoption of GPUs in simulation software. By having what is essentially a complete computer inside a workstation (processor, memory and fast storage) this could encourage more CAE software developers to explore the technology, with the end goal of significantly cutting solve times.
The AMD Radeon Pro SSG will be available initially as a developer kit for $9,999. Full availability is planned for 2017.