This article is more than 1 year old
Intel is over GPUs and CPUs – it's all about 'XPUs' now that OneAPI code-abstraction tool is golden
And why not have a server-grade GPU dedicated to game-streaming to test the new abstraction tool
Intel would prefer developers stop talking about CPUs and GPUs and instead target "XPUs".
The word about Chipzilla's new preferences came as it formally launched the OneAPI abstraction layer that lets applications tap whatever available resources are most appropriate to its needs, be that a CPU, a GPU, or an FPGA, from one interface as all of these are now just "XPUs" in Intel's world.
The silicon slinger first floated OneAPI back in December 2018, and issued a 1.0 spec for the project in September 2020, as our sibling site DevClass reported at the time. Now Intel has said it will make a gold release of OneAPI toolkits available in December.
Once that happens, Intel's Parallel Studio XE and System Studio tool suites will transition to become oneAPI products.
Nvidia's CUDA has done the same job as OneAPI for a number of years, so Intel appears not to have leapfrogged its GPU-centric rival with this announcement.
Also announced today was Chipzilla's new server GPU dedicated to servers that stream Android games and media.
The new device packs four Intel GPUs into a single card made by H3C (HPE’s server-making partner in China). Each GPU uses Intel’s Xe-LP microarchitecture and offers a 128-bit pipeline, plus 8GB of DDR4 memory.
Intel said the card is designed for Android streaming and promised that two-card servers that also pack a Xeon Scalable CPU will be able to handle over 100 simultaneous game sessions at high resolutions and frame rates.
Those numbers offer an interesting peek into gaming economics as they suggest the kind of rigs that big gaming concerns need to run in order to satisfy smartphone-wielding gamers. That Intel sees such operators as worthy of a device dedicated to their needs may also be more interesting than either OneAPI or the devices themselves.
Nvidia has tried to funnel server users to its Tesla range of GPUs and suggests servers pre-configured to make them hum, but doesn't offer a direct competitor to Intel’s new offering. ®