Transaction Processing Performance Council to deliver AI benchmark
GPU-makers aren't members of the TPC. Will they join?
The Transaction Processing Performance Council has decided the world needs a benchmark for systems running artificial intelligence workloads.
Raghunath Nambiar, chairman of Council's (TPC's) TPC-AI Working Group told The Register that interest in AI means users need the kind of comparison tools in which the organisation specialises. The TPC has therefore convened the new working group and given it six-to-twelve months to get the job done.
That's an ambitious timeline because the TPC has not always been able to move that fast. The organisation has, however, sped things up of late, delivering an IoT gateway benchmark in 15 months and a hyperconverged infrastructure test in just seven.
Nambiar told The Register that he feels the new group won't have trouble finding workloads to use, as the 25-plus TPC members will be able to find or create something useful.
Hardware, however, presents a challenge because AI infrastructure can be rather exotic. It is widely expected that field-programmable gate arrays (FPGAs), often integrated with CPUs, will be widely deployed to bring custom on-silicon code to AI applications. The group will need to determine how to measure such machines.
And then there's GPUs, which are almost standard issue for AI applications. Major GPU-makers aren't currently members of the TPC. Given GPUs' prominence in AI, Nambiar said he's hopeful they'll jump at the chance to work on a benchmark so that users can understand how to put such hardware to work. ®