Intel’s 5th Generation Xeon Emerald Rapids CPUs have demonstrated a significant performance boost when utilizing AVX-512 instructions, according to benchmarks by Phoronix. In tests with a server equipped with two Xeon Platinum 8592+ 64-core processors and 1TB of DDR5 memory, the use of AVX-512 doubled the average performance, with some workloads like OpenVINO showing increases of over tenfold. This improvement is particularly beneficial for AI applications due to support for AVX-VNNI and BF16 instructions.
Despite the performance gains, the power consumption remained relatively unchanged on average, although some tasks did require up to 10% more power. The peak frequency of the CPUs was slightly reduced when AVX-512 was enabled, which is likely a result of the increased power draw and temperatures. However, the CPUs still reached their 3.9 GHz boost clock with AVX-512 either on or off.
The inclusion of AVX-512 instruction support is a key feature of the Emerald Rapids chips, setting them apart from AMD’s 4th Generation EPYC Genoa processors, which lead in raw performance. This capability is particularly relevant in the context of AI workloads and may have influenced Microsoft’s decision to opt for last-generation Sapphire Rapids chips for their pairing with AMD’s MI300X GPUs.
Read more at Tom’s Hardware…