Koi Computers’ HPC Servers Feature AMD Instinct MI100 Accelerator
The MI100 accelerators deliver compute and interconnect performance, offering a boost for HPC (FP32 Matrix) performance and a boost for AI throughput (FP16) performance.
March 25, 2021
Koi Computers announces its turnkey high-performance computing servers and clusters feature the AMD Instinct MI100 Accelerator graphics processing unit.
Powered by the first AMD CDNA architecture, the MI100 accelerators deliver compute and interconnect performance, offering a boost for HPC (FP32 Matrix) performance and a boost for AI throughput (FP16) performance compared to AMD’s prior generation accelerators. Features and benefits include:
- New CDNA architecture delivering density to accelerate HPC and AI workloads compared with AMD prior gen (Learn more at AMD.com/en/technologies/CDNA;)
- New CDNA architecture delivering higher FP64 performance to accelerate HPC workloads compared with AMD prior gen;
- New Matrix Core technology delivering aAI matrix performance with FP32 and FP16—introducing bFloat16 operations;
- 2nd-generation Infinity Architecture bringing advanced platform connectivity and scalability;
- enhanced HBM2 memory interface with more theoretical memory throughput and improved latencies vs. AMD’s prior gen accelerators;
- Enhanced RAS and security capabilities; and
- AMD ROCm Open ecosystem.
MI100 accelerators offer an open environment that helps eliminate vendor lock-in, enabling developers to enhance existing GPU codes to run everywhere. Combined with the AMD EPYC processors and AMD Infinity Fabric technology, MI100-powered systems provide scientists and researchers with powerful platforms.
Sources: Press materials received from the company and additional information gleaned from the company’s website.