ZenDNN 5.2: Accelerating vLLM V1 Engine and Recommender Systems Inference on AMD EPYC™ CPUs
… By deploying multiple vLLM instances using numactl and interleaving memory access for each instance, we’ve effectively maximized DRAM memory bandwidth. …
Search
ZenDNN 5.2.1: Deepening Quantization and Expanding the AI Inference Frontier on AMD EPYC™ CPUs
… Run More with Less The practical impact of quantization is straightforward: models that previously required BFloat16 precision and the memory footprint that comes with it can now run in INT4 or INT8 with low to minimal accuracy loss Table-1 . …
A Look Ahead: Extending Server Energy Efficiency with LPDDR5X Memory
… One of the most promising developments is Low Power Double Data Rate 5X LPDDR5X memory deployed using the recently JEDEC approved Small Outline Compression Attached Memory Module SOCAMM2 form factor, which promises to combine the proven power efficiency of LPDDR5 mobile memory with the modularity a… …
7 Questions to Ask Before Building Your AI Infrastructure
… April 30, 2026 A Look Ahead: Extending Server Energy Efficiency with LPDDR5X Memory Learn how LPDDR5X memory can help improve server energy efficiency while delivering high bandwidth and modular serviceability for next-gen data centers. …
How To: Migrate your Cloud Instance to AMD EPYC
… The patterns in this guide, such as rolling replacement, blue/green, and canary, are the same deployment strategies many organizations already use for version upgrades, and they allow the transition to happen gradually, with full observability and instant rollback at every step. …