NVIDIA Grace CPU, mathematical optimization efficiency and performance improvements

According to the NVIDIA Technology Blog, in recent developments, NVIDIA’s Grace CPU has made significant progress in mathematical optimization performance and energy efficiency. These improvements are expected to benefit industries that require high computational power and energy-saving solutions.

Improved optimization features

Mathematical optimization is a critical tool for businesses to make smarter decisions, improve operational efficiency, and reduce costs. However, the complexity of models and the size of data sets require sophisticated AI algorithms and high-performance computing. NVIDIA’s new Grace CPU aims to meet these demands with its exceptional computing capabilities.

Gurobi Optimization, a leading mathematical optimization solver founded in 2008, has received a Supermicro NVIDIA MGX-based system powered by the NVIDIA GH200 Grace Hopper Superchip. The system promises high performance with low power consumption, addressing the need for efficient and fast optimization solutions.

Benchmarking Performance

Benchmark testing utilized a single NVIDIA Grace Hopper Superchip server and a cluster of four AMD EPYC 7313P servers. The test setup included Gurobi Optimizer 11.0 on Ubuntu 22.04, with the Grace Hopper Superchip combining an Arm-based NVIDIA Grace CPU and an NVIDIA Hopper GPU.

Performance evaluations were performed using Mixed Integer Programming Library (MIPLIB) 2017, which includes 240 real-world optimized instances. Results on NVIDIA Grace CPUs were compared to commonly used AMD EPYC servers.

Important discoveries

Early benchmarks show that the NVIDIA Grace Hopper Superchip outperforms AMD EPYC servers in most hard-core models, achieving an average runtime of 80 seconds compared to AMD’s 130 seconds. That’s a 38 percent improvement. Additionally, the NVIDIA Grace CPU delivered 23 percent faster throughput than the AMD EPYC 7313P while consuming 46 percent less energy.

Further analysis showed energy consumption benefits, with Grace Hopper using about 1.4kWh at 8 threads, compared to 1.75kWh for AMD, a 20% improvement. At 12 threads, Grace Hopper used 1.6kWh, compared to 2.6kWh for AMD, a 38% improvement.

Future outlook

According to preliminary benchmarks, Gurobi Optimizer running on NVIDIA Grace Hopper Superchip delivers faster computational performance with lower energy consumption. This development provides hope to a variety of industries that want to improve energy efficiency while solving complex business challenges with improved performance.

For a more in-depth look at the testing and results, interested readers can watch the on-demand session at NVIDIA GTC. More insight into how mathematical optimization can solve complex problems can be found in the Gurobi Resource Center.

Image source: Shutterstock