[ad_1]
Following the official launch of Intel’s first discrete GPU for over a generation, the DG1, Intel is launching this morning the server counterpart of this chip, the very clearly named “Intel Server GPU”. Previously known as SG1, the Intel Server GPU is based on the same Xe-LP architecture design as the DG1, but is aimed at the server market. And like the mainstream DG1, Intel plans to take an interesting and somewhat conservative approach with its new silicon, seeking specific markets well suited to Xe-LP’s large investment in video encoding hardware.
One of those markets that Intel has decided to pursue with its new silicon is the Android game streaming market. The company sees both gaming and video as growing markets, and game streaming is the perfect intersection of the two. So, with the hope that Android cloud gaming will grow significantly in the coming years – especially in China – the company is positioning the GPU, hardware, and developing a suitable software stack to use the server GPU to host games. Android in the cloud.
As for the GPU of the server itself, it’s practically identical to the DG1 which was unveiled just two weeks ago. That is, it is essentially a discrete version of Tiger Lake’s integrated GPU, offering 96 EU and 24 ROP, and linked to a 128-bit LPDDR4X memory bus. The only noticeable difference here is that while DG1 laptop implementations all use 4GB of VRAM (so far), Intel expects server GPU installations to get 8GB per GPU.
Either way, the mobile versions of this silicon get what Intel estimates to be around NVIDIA MX350 performance levels, which is certainly nothing out of the ordinary. The server’s GPU should do better than that thanks to its higher clock speeds and TDPs, but not extensively. As a result, the best use cases for Intel’s first discrete GPU aren’t big game bottlenecks or other rendering tasks, but rather tasks that take advantage of the computational capabilities of the dGPU and / or extensive multimedia encoding hardware.
This brings us back to Android cloud gaming, which by its very nature relies heavily on video encoding. The name of the game, as Intel sees it, is to enable high-density server installations for game streaming at a lower total cost of ownership (read: cheap), with Intel taking a particular pot-shot over rival NVIDIA on its software. license fees. Ultimately, with their server GPU, they think they can meet these specific demands.
And while the server GPU doesn’t set any performance records, it should still be way ahead of most mobile SoCs – and especially the type of lower performing SoCs that come into most Android phones. So even at its modest performance levels, Intel estimates that a single server GPU can power 10-20 game instances, and that number increases dramatically when you start talking about multiple GPUs.
To this end, Intel has even worked on developing the Linux drivers and modules needed to support Android game streaming workloads, in order to jumpstart the market. Which for Intel means not only developing the basic kernel drivers, but also the tools needed for gaming-focused virtual machines, and even some internal projects to improve game streaming performance. Of particular note is what Intel calls Project Flip Fast, which is Intel’s optimized client game streaming stack. Among other things, the company has enabled copyless transfers between host and guest to prevent performance from being affected by copying data between the two tiers.
But Intel can only take its server GPU so far by itself. Much like their processor market, Intel is not interested in selling entire systems – or in this case video cards. The company is therefore working with third parties to market the server GPU in commercial products.
Along with today’s launch of the Server GPU, Intel is also announcing its first client and the first product using the GPU: the new XG310 game streaming card from H3C. The Pathfinder product for the server GPU, the XG310 is a quad-GPU card specifically intended for the Android game streaming market. The card offers 8 GB of LPDDR4X per GPU, for a total of 32 GB of onboard VRAM. Meanwhile, connectivity is provided through a PCIe 3.0 x16 connection, which, according to the card’s appearance, is fed into a PCIe PLX switch before being distributed to individual GPUs.
The XG310 is intended for sale to server OEMs and game service providers, and Intel has already registered Tencent to show the card. According to the company, a single XG310 card is capable of running 60 streams of Arena of Valor at 720p30, and doubling the number of cards doubles the number of streams, which goes back to Intel’s overall TCO argument.
Ultimately, as it stands, the Intel H3C card is being validated with servers now. And, according to Intel, it’s already shipped worldwide with orders already arrived. The first customer will likely be Tencent, which will introduce its own cloud gaming service. Otherwise, although Intel is not talking about any other customers at this time, it is clear that they are eager to offer more Android game streaming providers in the future. In addition, these types of multi-GPU cards would also be suitable for Intel’s more traditional high-density media encoding customers, who previously purchased products like Intel’s Visual Compute Accelerator cards.
[ad_2]
Source link