Moore’s law continues to provide ever more powerful server processors that provide the ability to support more virtual desktops per server and improve the cost per desktop accordingly. How do these more powerful servers affect the value of PCoIP protocol offload provided by PCoIP Hardware Accelerator?
The table below compares a typical VDI server from 2012 (the first full year that the Hardware Accelerator was available) to typical VDI server in 2016. While the number of processor cores has increased by 75% in the past 5 years allowing 75% more VDI VMs per server, the clock speed of each core is actually down 11%. Thanks to microarchitecture improvements, the cores provide 9% higher system level performance as measured by SPECint_rate. Any way you measure it, the average 2 vCPU VM performs nearly the same as it did 5 years ago.
| Typical Server: 2012 | Typical Server: 2016 | Difference | |
| Sockets | 2 | 2 | N/A |
| Processor | Xeon E5-2680 | Xeon E5-2680v4 | N/A |
| Architecture | Sandybridge | Broadwell | N/A |
| Cores/server | 16 | 28 | 75% ↑ |
| VDI VMs/server | 64 | 112 | 75% ↑ |
| Clock speed (GHz) | 2.7 | 2.4 | -11% ↓ |
| SPECint_rate/core | 36.25 | 39.64 | 9% ↑ |
Moore’s law is primarily driving increased core counts, not individual VM performance. This means that the performance benefits of the PCoIP Hardware Accelerator for each individual VM remains essentially the same as it was when the product was launched.
Any way you measure it, the average 2 vCPU VM performs nearly the same as it did 5 years ago.
The graphs below show the relative CPU load and frame rates for an individual VM with a typical, 24 frames per second 1080p video being played in a Google Chrome browser for both a 1 vCPU and 2 vCPU VM. The PCoIP Hardware Accelerator frees up >35% of the CPU within the VM allowing the Chrome browser to have more cycles to decode the videos and deliver the native frame rate of the video and provide a better experience for the desktop user.
This demonstrates that PCoIP Hardware Accelerator continues to provide improved user experience and higher consolidation ratios (more VMs per server) even though the underlying servers continue to become more powerful.
[1] Intel® Xeon® CPU E5-2680 v4 @ 2.4 GHz; Hyper Threading Enabled, ESXi 6, Resources 2400MHz, Windows 7 SP1 64-bit, VMware Horizon 7.0.2 using PCoIP protocol connected to a PCoIP Tera 2321 Zero Client