Paperspace is the first hosted-desktop provider to come standard with a GPU. Why does this matter? Two primary reasons:
Fluid OS experience
Applications today are built to leverage GPUs. The GPU, short for Graphical Processing Unit, is what enables media-rich content to be created and displayed within computers. Traditionally however, GPUs have not been available within virtual desktops[1]. This results in a severely degraded VM experience and prohibited entire industries like architecture, engineering, graphic design, animation etc. from benefiting from this otherwise incredible technology. That is until now: Paperspace is the first provider of hosted-desktops that includes a GPU in every machine. For the first time, it is possible to achieve near native performance in a remote machine.
Streaming performance
Without a GPU available, desktop streaming protocols have traditionally used the CPU to encode video frames. CPUs are really good for serial tasks (one large calculation after another) but bad at parallel tasks (lots of tiny calculations simultaneously). Spoiler alert: Encoding frames is of the latter computational type. GPUs are basically made for encoding video and can cut down on latency, the holy grail of real-time streaming applications. Desktop streaming needs to happen in less than ~60 milliseconds for the human eye not to perceive lag. In the last few years, the internet has become really fast and GPUs are almost always available on the client side to decode the incoming video frames. This leaves the encode side as the only significant area for optimization and is the reason why we put GPUs in all our machines.
Check out the difference between a CPU based versus a GPU based desktop stream:
GPU Passthrough has been an option but is not comparable to other virtualized hardware like CPUs, storage, RAM, network cards etc. ↩︎
