LiquidSky Desktop Hosting Service Adopting AMD’s Upcoming Vega Architecture

AMD announced that LiquidSky, a cloud-based gaming desktop service, committed to adopting its upcoming Vega GPU architecture to power the machines in its 13 global data centers.

LiquidSky is a cloud-hosted service that provides gaming class desktop performance for people who don’t have a gaming PC. LiquidSky’s service is accessible through almost any internet-connected device, such as a low-end Windows PC, a system with Linux installed, or a Mac. LiquidSky even offers an Android app to let people play their PC games on the go.

The service already boasts 1.5 million subscribers across 13 data centers, and it’s still in beta. That’s a lot of GPUs to house in data center servers. And when the LiquidSky service officially launches, the company hopes to increase its subscriber base, so the company is preparing to bring even more systems online soon.

LiquidSky turned to AMD’s Vega architecture to help stifle its staggering cost of growth. Dedicating one GPU per client is simply not a sustainable business model--the company needed a solution that could share the resources of one GPU with multiple virtual machines without introducing latency constraints.

AMD’s Vega GPUs offer exactly the feature that LiquidSky was looking for. AMD said that in a virtualization situation, you could split Vega into as many as 16 VMs per GPU. AMD uses a hard split method to segment the GPU and memory buffer together, which is to say the allocation per VM isn’t dynamic. In most cases, a pool of virtual machines will divvy the available compute resources on the fly, so if 16 VMs are online, each VM will get one 1/16 of the performance. But if four VMs are offline, the rest of the compute power gets shared between the remaining VMs.

That sort of performance variation is fine for virtual machines in an office setting, but LiquidSky demanded guaranteed performance for its clients. The Vega architecture allows LiquidSky to lock a predetermined section of the GPU for each VM. Each client gets no more and no less than they sign up for.

“LiquidSky’s game streaming service delivers the very best visuals, detail, and pure performance, regardless of the device you’re using,” said Ian McLoughlin, LiquidSky’s co-founder, and CEO. “AMD’s Vega-based GPUs will have the perfect blend of bleeding-edge hardware virtualization features and tremendous rendering horsepower. This means consistent framerates and quality of service that’s simply not possible with existing technologies.”

AMD’s Vega GPUs are capable of powering 16 simultaneous virtual machines, but LiquidSky leaves you with more than 1/16 of the GPU resources. The desktop hosting service gives one-quarter of each GPU, which includes 4GB of memory, to each subscriber.

AMD didn’t say when LiquidSky would receive its first shipment of Vega-based graphics cards, but it said that Vega-based GPU products would ship in the first half of 2017 and LiquidSky will have Vega cards on day one.

LiquidSky offers a range of subscription prices, including an ad-supported plan, which gives you access to a high-end gaming PC free of charge.

 Kevin Carbotte is a contributing writer for Tom's Hardware who primarily covers VR and AR hardware. He has been writing for us for more than four years. 

  • Andrewst1021
    Now can if Unraid could support this it would the perfect in home game server.
    Reply
  • bit_user
    The Vega architecture allows LiquidSky to lock a predetermined section of the GPU for each VM. Each client gets no more and no less than they sign up for.
    This is standard practice, for hard-realtime systems. If you want predictable latency and throughput, you can't oversubscribe the resources. This means some cycles might go to waste, but modern power management techniques mean that at least at least they won't have to pay the electricity costs of a maxed out GPU, when not all subscribers' portions are maxed.

    AMD’s Vega-based GPUs will have the perfect blend of bleeding-edge hardware virtualization features and tremendous rendering horsepower.
    Anyone remember when "bleeding-edge" was a pejorative term? At least, to the extent that you wouldn't say it as a compliment about your new partner?

    The desktop hosting service gives one-quarter of each GPU, which includes 4GB of memory, to each subscriber.
    Okay, so we know there'll be a 16 GB version of Vega. Nice.
    Reply