Intel Server GPU: The XG310 combines 4 Xe GPUs on one PCIe x16 card



[ad_1]

The low-power version of Intel Xe is now also available for the server for the first time. The manufacturer uses the known chips from DG1, combines them with dual memory and packs four of them on a PCB, which fits into a normal PCIe Express slot. The result is a special solution and not a real mass product.

It should be another test ball on the way to the big market in the high-performance, high-performance gaming industry. Because the new Intel Server GPU is quite special, which is already visible during construction.

Put simply, Xe-LP graphics, first used with Tiger Lake, are now not only offered by early manufacturers as a discrete notebook solution for DG1; Intel adapted a similar procedure for the first server solution. But here four chips are installed on one circuit, each combined with 8GB LPDDR4 in the form of four chips, powered by an additional 8-pin power connector to the PCIe slot.

A single graphics processor with its 96 EU and a maximum clock rate of 1.1 GHz of 23 watts, the total power of the passively cooled PCIe solution card is 150 watts. A clearly visible PLX chip above the PCIe interface controls the whole thing with PCIe lanes, as has been the case with multi-GPUs for years or decades.

PCIe H3C XG310 graphics solution

PCIe H3C XG310 graphics solution

PCIe H3C XG310 graphics solution

PCIe H3C XG310 graphics solution

Intel is working with Chinese industry giant H3C. The target group for the cards are their and other servers in the cloud, mainly in Asia. Companies like Tencent cite Android games on smartphones as an example, which ultimately run on servers in the cloud. Four of the new GPUs, each with four chips, can be used per server, and per server cabinet there is a high level of performance at relatively low cost according to Intel, although exact figures are not revealed.

The fact that Intel also uses the same GPU for the third solution has advantages in the development of subsequent products. The software base is identical; with oneAPI, Intel wants to make this even easier in the future and to involve as many partners as possible. Thanks to the wide availability of small GPUs – especially through the notebook – they should be introduced quickly to the topic with several million users, so that optimizations are available more quickly and problems are resolved more quickly. With various toolkits, Intel wants to set the foundation even more broadly in the future.

Intel oneAPI Basis-Toolkit

Intel oneAPI Basis-Toolkit

Intel oneAPI additional toolkits

Intel oneAPI additional toolkits

The calendar: Intel Xe HP, HPG and HPC

Raja Koduri, as head of the divisions, insisted on reporting on the state of development even today. He was proud that LP variants are now being presented and delivered, but the next step will be significantly broader.

In Intel’s labs, there are no more virtual GPUs than the most powerful solutions, but real silicon back from production. The Intel Xe HP will be driven as a two-card version, Intel explained in advance in a press-on-demand briefing, in the market it is expected with one, two, and four cards. The home gaming PC version, HPG, will ultimately be just an offshoot of HP and may even run.

Intel's graphics solutions based on Xe
Intel’s graphics solutions based on Xe

Much more important to Intel is the look of HPC, codenamed Ponte Vecchio. They are still in development there, but have now delivered the first designs to internal and external factories (TSMC) and are awaiting the return of the chips. When the proper ignition has been performed, final tuning can begin.

ComputerBase received information on this article from Intel under the NDA. The only requirement was as soon as possible publication time.

[ad_2]
Source link