It has been revealed via AMD’s press deck that the hardware manufacturer’s flagship GPU i.e. the Radeon R9 290X, features as many Asynchronous Compute Engines as the PS4. Furthermore, the number of available queues for Compute commands in each ACE pipeline are also the same i.e. 8.
This means that a maximum of 64 Compute commands can be queued for execution by the Compute Units in the R9 290X. This much expanded ACE pipeline was initially touted by Sony’s Mark Cerny as the primary hardware modification made to the PS4’s GPU. It’s interesting to see that AMD has rolled the enhancement into its updated GCN 2.0 architecture. It’s well-known that GCN in itself is an in-order architecture. With such an expanded ACE pipeline, the idea is to enable it to function like an out-of-order processor.
Whether this will give AMD an edge over Nvidia in terms of potential GPGPU driven middleware, such as Havok or other kinds of physics simulation, remains to be seen. Developers taking advantage of the PS4’s expanded ACEs for scheduling and multi-tasking of Compute tasks could very well do the same for the PC domain, especially with the aid of AMD’s own Mantle API.
On numerous occasions, Cerny has stressed upon Compute-driven game world simulation as being the key to a truly next-gen experience. It remains to be seen if this does indeed turn out to be the case, although the idea of a fully-realized GPGPU driven game world does sound intriguing. Crytek has hinted at taking advantage of such hardware capabilities with its real-time weather system, which is entirely GPGPU driven.
ACE may also play a vital role outside the realm of gaming. With certain kinds of processes (such as hash data), it’s likely that the R9 290X would be popular among data miners for its independent scheduling and work item dispatch functions.
Could GPGPU driven real-time world simulation really the future of gaming? Let us know what you think in the comments below.