6U 8-GPU NVIDIA HGX H200 AI server with 1.1TB of HBM3e, dual AMD EPYC 9005 processors, up to 9× 400G NICs, and 24× 2.5" U.2 NVMe bays for hyperscale AI training and LLM inference.
6U Rackmount
NVIDIA HGX H200 8-GPU (SXM)
1.1TB HBM3e total (141GB per GPU)
32 PFlops (FP4)
2× AMD EPYC 9005 (Intel Xeon variants available)
24× DDR5 DIMMs (up to 32 DIMMs on Intel variants)
24× 2.5" U.2 NVMe SSD + 2× M.2
Up to 10× PCIe 5.0 slots
Up to 9× 400G NICs
Air cooling
Available
Quote
The Aivres KR6288 is a 6U 8-GPU AI system purpose-built for hyperscale data centers around NVIDIA's HGX H200 baseboard. With eight Hopper-generation H200 SXM GPUs delivering a combined 1.1TB of HBM3e memory and 32 PFlops of AI performance, it accelerates GPT-class large language model training and high-throughput inference. ServeTheHome's review of the KR6288 — the first HGX H200 server they benchmarked — highlighted the directly accessible GPU tray design for fast field service.
The platform pairs the GPU subsystem with dual AMD EPYC 9005 processors (with Intel Xeon configurations also offered), up to 32 DIMMs of DDR5, 24× 2.5" U.2 NVMe bays plus 2× M.2, and up to ten PCIe 5.0 expansion slots for up to nine 400G NICs. Modular construction, a midplane for power and data delivery to the GPU compute tray, and optimized airflow make it a scalable building block for AI factories running training, fine-tuning, and inference at scale.
Need Aivres KR6288?
HGX H200 8-GPU AI infrastructure for hyperscale training and inference. Contact QS Compute for pricing, availability, and integration services.
Request Quote