CDNA Based AMD GPU to rival NVIDIA’s Ampere in late 2020

AMD has revealed that its first GPU designed particularly for computing and based on CDNA structure will grow to be obtainable later this year. Being developed primarily for synthetic intelligence and high-performance computing (HPC) functions, the brand new Radeon Intuition will compete towards Nvidia’s just lately launched A100 GPU primarily based on the Ampere structure.

Having realized early sufficient that GPUs for AI and HPC requires help for options and directions that aren’t essentially wanted by customers, Nvidia has launched a number of datacenter-specific GPUs which were adopted by a whole bunch AI and HPC-oriented machines. To that end, the multi-billion greenback GPU computing market is essentially dominated by Nvidia at the moment.

AMD launched its first information center GPU — the GCN-based Vega II — again in late 2018 and whereas the unit has not been as profitable as components from Nvidia, it’s at the moment utilized by a number of cloud service suppliers and HPC prospects.

AMD’s next attempt on the server GPU front embraces its new CDNA structure that’s tailor-made primarily for AI in addition to HPC and due to this fact guarantees to be a greater match for respective prospects. AMD expects to launch its first CDNA-based Radeon Intuition MI100 GPU within the second half of the year.

“We anticipate [datacenter GPU] income to extend within the second half of the year, as extra cloud-based visible computing wins ramp and we launch our new CDNA information heart GPU structure optimized for next-generation exascale and machine studying workloads,” mentioned Lisa Su, chief govt of AMD.

AMD’s first CDNA GPU reportedly derives from the Vega II (GCN 1.5) processor, options new vector ALUs, Bfloat16 help, and is codenamed Arcturus. The corporate doesn’t challenge its first CDNA GPU to considerably contribute to its revenues in 2020. AMD says that its datacenter GPUs will grow to be a major progress alternative in 2021 and past, which isn’t stunning as such processors will probably be used within the upcoming Frontier and El Capitan supercomputers.

“We see [datacenter GPUs] as a multi-year opportunity, however, it’s not an enormous income contributor right here in 2020,” mentioned Su. “But we see progress alternative as we go into 2021 and past.”