Virtualizing the GPU the AMD Way

AMD says their new multiuser AIBs (add-in boards) are equipped with special multiuser technology embedded into the GPU, and through its use it can deliver consistent and predictable performance. When appropriately configured (to the needs of an organization), end users get the same access to the GPU no matter their workload. Each user, says AMD, is provided with the virtualized performance to design, create and execute their workflows without any one user tying up the entire GPU.

Based on SR-IOV (Single Root I/O Virtualization) technology, AMD says their multiuser GPU continues AMD’s embracement of non-proprietary open standards. SR-IOV is a specification developed by the PCI SIG, and provides a standardized way for devices to expose hardware virtualization.

The SR-IOV has been out for a while, but MAD is the first to embrace it for virtualizaton (Source AMD)

AMD thinks their multiuser GPU addresses limitations of current virtualized GPU solutions that may not provide predictable performance for CAD/CAE, Media and Entertainment, and general enterprise GPU needs.

Claiming to be created for GPU-accelerated workflows such as GPU compute and OpenCL, AMD says their approach is designed to overcome the limitations of software-based virtualization such as reduced end-user performance. The users, says AMD, have access to native AMD display drivers for OpenGL, DirectX and OpenCL acceleration, enabling work without restrictions.

AMD claims their deterministic GPU allocation is the industry first hardware virtualization for GPUs utilizing SR-IOV Technology (source AMD)

The multiuser GPU is designed to work in environments using VMware vSphere/ESXi 5.5 and up, with support for remote protocols such as Horizon View. Citrix Xen Desktop, Teradici Workstation Host Software and others. It uses VMware’s VIB driver with VMware certification, and AMD says no changes from pass-through are needed to run on vSphere/ESX 5.5 or vSphere/ESX 6.0. Other hypervisors with SR-IOV support can be enabled in the future.

AMD says the AIBs for multiuser virtualization will be available to OEMS in Q3 2015.

AMD’s approach to specific GPU deterministic partitioning, rather than a dynamic load-balancing approach like Nvidia offers is bound to have adherents because AMD said they did it based on feedback they got from users. AMD claims their direct hardware solution offers better throughput, and tighter communications with the drivers and hypervisor, and points out one user cannot affect another user’s GPU performance.

I asked AMD if they planned to build a rack-mount unit as Nvidia has done with their Grid product. AMD said no, they did not want to compete with their customers and would support their customer’s in the development of such chassis. So there won’t be any specific AMD Virtualize boxes in the immediate future. However, we can envision AMD developing a generic box for white-box, tier three server suppliers, and VARs who want to offer a turn-key solution with virtualization. 

AMD will be showing this technology at the JPR Virtualize conference in San Francisco on 29 October.