Following the formal launch of Intel’s first discrete GPU in over a technology, the DG1, this morning Intel is launching the server counterpart to that chip, the very plainly named “Intel Server GPU”. Previously known as SG1, the Intel Server GPU relies on the identical Xe-LP structure design because the DG1, however aimed on the server market. And like the patron DG1, Intel is planning on taking an fascinating, considerably conservative tack with their new silicon, chasing after particular markets which are nicely suited to Xe-LP’s vital funding into video encode {hardware}.
One such market that Intel has determined to chase with their new silicon is the Android recreation streaming market. The firm sees each gaming and video as development markets, and recreation streaming is the proper intersection of the 2. So, with an expectation that Android cloud gaming goes to significantly develop within the coming years – particularly in China – the corporate is positioning the GPU the {hardware} and growing an acceptable software program stack to make use of the Server GPU for internet hosting Android video games within the cloud.
As for the Server GPU itself, it’s just about similar to the DG1 that was unveiled barely two weeks in the past. That is to say that it’s primarily a discrete model of Tiger Lake’s built-in GPU, providing 96 EUs and 24 ROPs, and sure to a 128-bit LPDDR4X reminiscence bus. The one notable distinction right here is that whereas DG1 laptop computer implementations are all utilizing 4GB of VRAM (up to now), Intel expects Server GPU installations to get 8GB per GPU.
In any case, the cellular variations of this silicon get what Intel estimates to be round NVIDIA MX350-levels of efficiency, which is admittedly nothing to write down house about. Server GPU ought to do higher than this due to its larger clockspeeds and TDPs, however not extensively so. As a consequence, one of the best use circumstances for Intel’s first discrete GPU are usually not large, shader-bottlenecked video games or different rendering duties, however moderately duties that reap the benefits of the dGPU’s compute capabilities and/or in depth media encode {hardware}.
This brings us again to Android cloud gaming, which by its very nature depends closely on video encoding. The title of the sport, as Intel sees it, is all about enabling high-density server installations for recreation streaming at a decrease complete value of possession (learn: low-cost), with Intel taking a specific pot-shot at rival NVIDIA over their software program licensing prices. Ultimately, with their Server GPU they imagine they will meet these particular calls for.
And whereas the Server GPU isn’t going to be setting any efficiency information, it ought to nonetheless be nicely forward of most cellular SoCs – and notably the form of lower-performance SoCs that go into the majority of Android telephones. So even at its modest efficiency ranges, Intel is estimating {that a} single Server GPU can energy between 10 and 20 recreation cases, and that quantity scales up considerably if you begin speaking about a number of GPUs.
To that finish, Intel has even labored on growing the mandatory Linux drivers and modules to assist Android recreation streaming workloads, so as to bootstrap the market. Which for Intel means not solely growing the bottom kernel drivers, but additionally the mandatory instruments for gaming-focused digital machines, and even some inside tasks to enhance recreation streaming efficiency. Of explicit be aware here’s what Intel is looking Project Flip Fast, which is intel’s optimized consumer recreation streaming stack. Among different issues, the corporate has enabled zero copy transfers between the host and the visitor so as to keep away from the efficiency hit from copying knowledge between the 2 ranges.
But Intel can solely take their Server GPU up to now on their very own. Not in contrast to their CPU market, Intel isn’t eager about promoting complete programs – or on this…
