Lucid reaches for the clouds with GPU virtualization

By now, you're likely familiar with the basic technology behind Lucid's various products. Whether it's the original Hydra GPU load balancer or the Virtu software package that allows IGPs and discrete GPUs better to coexist, the core technology is the same: a software abstraction layer that intercepts calls to graphics APIs like Direct3D and then parcels out the work to the available GPUs as needed.  By standing in between the programming interface and the graphics hardware—that is, by virtualizing the GPU—Lucid is able to make interesting things happen, whether it's allowing users of discrete graphics cards to access the QuickSync hardware in the Sandy Bridge IGP or enabling smoother frame delivery via the very clever HyperFormance scheme.

This year at CES, Lucid was offering an early peek at the next application of its GPU virtualization software, and this one looks to be more conventional. The company has created a software framework that will allow a single GPU on a "server" system to serve many clients simultaneously, and each of those clients can have its own, independent GPU workload.

Lucid Founder and President Offir Remez gave us a peek at this software in action, though he was careful to label it only a technology demo. In the demonstration, we saw a desktop system equipped with a GeForce GTX 500-series graphics card play the role of the server, while a couple of laptops with relatively weak integrated graphics acted as clients. All three systems were connected to one another via a local Wi-Fi network. A Call of Duty game was started up via a shortcut on the first laptop, and immediately a window opened on the desktop system, showing the game. Right afterward, a window opened on the laptop's screen showing the same content, only slightly behind the desktop system. After a few moments, the laptop began to catch up with the desktop, so that the two were almost exactly in sync. The Call of Duty game was playable on the laptop, with reasonable responsiveness and image quality. The input lag wasn't really noticeable by my seat-of-the-pants evaluation, but I only tinkered with the system briefly.

Next, while the CoD game was left running, Lucid fired up another game, the lovely Madagascar 2, on a second client laptop. Again, the game opened in a window on the desktop system, and its contents were streamed to the second laptop. The two games ran concurrently and were both playable, and as far as I could tell, the quality of the experience wasn't degraded substantially on the first laptop when the second game was running.

Remez told us this software relies on H.264 video compression for streaming the content to client systems, and he said the video streams were compressed in real-time via a CUDA-based encoding tool—a third workload running on the virtualized GeForce GPU in the host desktop system. Even so, this demo was just a small slice of the technology's potential. Remez said Lucid has had as many as seven games running at 30 FPS on a single GPU. The host system is not limited to just one graphics chip, either. The server could have a cluster of GPUs installed. Lucid is agnostic about the vendor type, so it could use a mix a GeForces and Radeons.

What Lucid has here appears to be a very capable technology with loads of potential. What they don't yet have, it seems, it a clear sense of what business model to mate with it. The list of possible customers includes everything from cloud gaming service providers to end users running multiple virtual desktops in their home. We'll have to wait and see where it goes next.

On another front, we're still waiting impatiently to get our hands on a version of Lucid's Virtu MVP with HyperFormance. It now appears we'll get that chance, as will end users, once Intel's Ivy Bridge processors hit the market. A great many of today's Intel Z68 motherboards ship with Virtu, and Remez believes Lucid will repeat that success with Ivy Bridge boards and Virtu MVP. Judging by our conversation with a couple of key Gigabyte execs later that same evening, he is probably correct. We're curious to see how effectively HyperFormance works in the wild and whether enthusiast gamers will embrace its drawbacks (that is, losing direct control over the GPU via its control panel and drivers) for the sake of smoother, more responsive gameplay.

HyperFormance uses two GPUs in order to work its magic, but Lucid has a version of the technology, dubbed XLR8, that employs only a single graphics chip. XLR8 does the same thing as HyperFormance, but it uses further virtualization magic to make a single IGP act as two, one in sync with the display and one out of sync. With that split, Lucid can then target frame delivery to match up with the display refresh sync, turning a borderline-painful 30 FPS gaming session into a playable, positive experience. We saw an early demo of this tech in action at IDF last fall, but it's not yet part of a shipping product. We tend to think Intel, an early investor in Lucid, should adopt XLR8 wholesale, absorbing it into its IGP drivers, but that doesn't seem to be in the works at present.

Undaunted, Lucid has moved ahead with XLR8 development not just on laptops, but on Android-based tablets with OpenGL ES. Remez showed us a demo of Modern Combat running on an Asus Transformer tablet without and then with Virtu and XLR8. True to form, XLR8 didn't raise frame rates, but it made a noticeable difference in the fluidity of frame delivery and animation. Remez made sure to point out that XLR8 doesn't cause additional battery drain and, in effect, raises the power efficiency of the whole solution.

Again here, Lucid is still pondering the business model for this technology, which could range from a packaging deal like they have for Virtu on the desktop to a simple app download in the Android Market. We're intrigued to see which chip and solutions providers might bite—and whether firms like Nvidia and Imagination Tech will attempt to cook up their own versions of HyperFormance, once its benefits are more widely understood.

Tip: You can use the A/Z keys to walk threads.
View options

This discussion is now closed.