AMD is teaming up with Rapt AI to enhance AI model training and inference on its Instinct processors. On Wednesday, they announced their plans to combine Rapt AI’s workload automation platform with AMD’s MI300X, MI325X, and the upcoming MI350 series GPUs.
This partnership aims to help businesses better allocate resources and manage GPUs while addressing performance bottlenecks. Rapt AI also plans to focus on optimizing memory utilization in the future.
According to Jack Gold from J.Gold Associates, enterprise IT teams often struggle to maximize their data center resources. Monitoring utilization is crucial, and that’s exactly what Rapt AI’s software does for GPUs. The technology tracks workloads on each GPU, ensuring they operate at full capacity. This means companies will invest in additional GPUs only when necessary, ultimately leading to cost savings.
“It’s about avoiding overspending on compute resources that aren’t fully utilized,” Gold explains.
Rapt AI’s software is compatible with both AMD and Nvidia GPUs in private or cloud data centers. It also works with tensor processing units from Google Cloud and AWS’ Trainium AI accelerator—specialized chips designed for high-volume AI tasks.
This collaboration means Rapt AI will optimize its software to ensure peak performance on AMD’s Instinct processors. “They’ll work closely together to fine-tune everything,” Gold says.
In the competitive landscape, AMD’s GPUs, including the Instinct series, go head-to-head with Nvidia’s H100, H200, and Blackwell GPUs for AI model tasks. In the last quarter of 2022, over half of AMD’s revenue came from Instinct GPUs and Epyc CPUs, which cater to data centers.
AMD plans to ramp up shipments of the MI350 around mid-year. Designed specifically for AI workloads, this GPU targets cloud providers and major enterprises, including pharmaceutical companies, financial firms, and large retail operations. Following the MI350, AMD aims to release the MI400 in 2026, featuring a new CDNA 4 microarchitecture that integrates networking, CPU, and GPU capabilities at the silicon level. The programming platform for these Instinct processors is called Radeon Open Compute, or ROCm.
Antone Gonsalves, an editor at Informa TechTarget, has been reporting on tech trends for 25 years, offering insights valuable to enterprise technology buyers. If you have any news tips, reach out to him via email.