Intel, Samsung and different chipmakers again new accelerator programming initiative

The Linux Basis has launched a brand new trade group, the UXL Basis, to ease the duty of writing functions that may run nicely on a number of kinds of accelerator chips.

The initiative was introduced on Tuesday at an trade occasion in Spain. The UXL Basis is launching with backing from a number of of the largest gamers within the semiconductor market. The group’s members embrace Intel Corp., Arm Holdings plc, Qualcomm Inc., Samsung Electronics Co. Ltd. and U.Okay.-based chip firm Creativeness Applied sciences Ltd.

The initiative is extensively seen as a foil to Nvidia Corp.’s rising dominance in accelerated computing, which has exploded thank to the fast rise of generative synthetic intelligence functions corresponding to OpenAI LP’s ChatGPT. “We have now an open supply competitor to Nvidia’s CUDA platform for leveraging accelerators corresponding to GPUs,” former Google LLC engineer Kelsey Hightower posted Tuesday on X, previously Twitter.

Google’s cloud enterprise and Fujitsu Ltd. are additionally backing the UXL Basis. Google designs customized accelerator chips for its cloud platform. Fujitsu, in flip, has a enterprise unit that makes specialised semiconductor merchandise corresponding to FeRAM, a kind of high-speed reminiscence just like DRAM.

The objective of the UXL Basis is to make it simpler for builders to put in writing software program that may run on accelerator chips. These are processors which might be optimized to carry out one set of computing duties with a excessive diploma of effectivity. Due to their specialised design, accelerators can often run the workloads for which they’re optimized sooner than a central processing unit.

Lots of the accelerators in the marketplace at present are designed to hurry up synthetic intelligence workloads. There are additionally merchandise that concentrate on different duties. Intel Corp.’s IPU chipsfor instance, are constructed to hurry up the computations concerned in managing a knowledge middle’s storage and community tools.

An software constructed for one accelerator chip can’t essentially run nicely on others. That’s true even when these different chips are optimized for a similar sort of workloads. To port their software program from one accelerator to a different, builders usually must make intensive code modifications, which requires a major quantity of effort and time.

The newly launched UXL Basis will work to simplify the creation of functions that may run nicely on a number of accelerator chips. In line with the group, it plans to attain that goal by creating an “open commonplace accelerator programming mannequin.”

On the coronary heart of the trouble is an present expertise known as oneAPI that was initially developed by Intel. The chipmaker has contributed the expertise to the UXL Basis. Utilizing oneAPI, builders can port software program from one accelerator chip to a different with a fraction of the work traditionally required for the duty.

Intel based mostly oneAPI on an earlier expertise known as SYCL. The latter software program permits builders to make use of C++, a well-liked programming language, to put in writing functions that may run on a number of kinds of chips. SYCL-powered software program can run not solely accelerators corresponding to AI-optimized processors but additionally CPUs and FPGAs, that are reconfigurable chips that lend themselves to a wide range of use instances.

The openAPI expertise that Intel has contributed to the UXL Basis expands upon SYCL’s function set. It provides efficiency optimizations and capabilities that ease software program debugging duties. Moreover, Intel has included fsoftwa instrument that may mechanically adapt AI software program written for Nvidia Corp. graphics playing cards to run on openAPI-compatible chips.

The UXL Basis will concentrate on “driving innovation and implementing the oneAPI specification throughout the trade,” the group acknowledged. To that finish, it plans to collaborate with each chipmakers and software program builders. The UXL Basis additionally intends to crew up with trade organizations corresponding to Khronos Group Inc., the nonprofit behind the SYCL expertise that underpins openAPI.

“The muse members aspire to construct the most important open ecosystem for accelerated computing,” mentioned Rod Burns, chair of the UXL Basis’s steering committee. “Our final purpose is to foster a multi-architecture and multi-vendor programming platform for all accelerators.”

The UXL Basis plans to share extra details about its targets and improvement roadmap later this 12 months.

Picture: Unsplash