AMD is the first to a fully virtualized GPU with their new Firepro S7100X module. If you have been looking for real SR-IOV compliant graphics, AMD’s new MxGPU tech is what you have been waiting for.
There have been lots of things on the market that claim GPU virtualization but the S7100X is the first to do full SR-IOV in hardware. In an ironic twist the new SX7100x module uses a Tonga ASIC which is hardly new so the hardware has been around for a while. You might recognize it as the thing under the FirePro sticker on the module below.
Note the elegant red PCB and brilliant copper connectors
That said the chip now has the hardware virtualization functions exposed to the user and will support up to 16 simultaneous VMs. In hardware. Once again it is really in hardware, and yes they are the first to get there despite some floating rhetoric. And yes it is actually that important, enough so that we will repeat the Multiuser GPU (MxGPU) term without wincing like we usually do for such things. If you use VDI or have a virtualization infrastructure in place, you get why this is important. The overhead brought on by virtualizing a GPU is pretty large, you can pay the penalty there or pay it with a 1:1 VM to GPU mapping. Either way the result isn’t all that great, overhead or cost and power. Now you can do it all in a 100W module made for HP blade servers.
So in the end we have an old ASIC with newly exposed features that really does move the state of the art forward. AMD now has a fully hardware virtualized GPU module for blade servers that will vastly increase the efficiency of VDI deployments. SR-IOV compliant GPUs should have been out long ago but at least they are here now. This is once again a really big deal.S|A
Latest posts by Charlie Demerjian (see all)
- AMD talks about Vega at a high level - Jan 17, 2017
- Intel unleashes more Kaby Lake SKUs on the yearning public - Jan 4, 2017
- Qualcomm opens up a bit more on the 10nm Snapdragon 835 SoC - Jan 3, 2017
- AMD’s Freesync 2 changes the display game - Jan 3, 2017
- Coffee Lake points to issues with Intel’s 10nm process - Dec 28, 2016