I would appreciate open feedback on the following sets of results. They are run on a single host server, running VMware vSphere 6.0 and Horizon 6.1, with Win7 64-bit guests in scales of 1, 8, and 16 using the K220Q vGPU profile (2x NVIDIA GRID K2 cards), then one more test where we only adjusted the vGPU profile to K240Q. The last column is another K220Q test run on a different host, with a single guest VDI, running on a Haswell but at a slower clock speed and higher core count CPU to see that impact.
The tests were kicked off simultaneously to create a peak load, various of the desktops had eyes on to ensure the tests were proceeding as expected. Keep in mind this is PEAK, meaning all users doing the same work at the same time, highly unlikely occurrence. Resolution was set to 1920x1080 (1080P).
The challenge we are looking to resolve is how to find the balance between acceptable performance and scalability, a very different measurement than the single traditional physical workstation max performance benchmarks. Thoughts?
Working on more Haswell results next…