Fujitsu PRIMERGY GX2580 M8s vs NVIDIA HGX B300 8-GPU

Accelerator configuration, aggregate memory, bandwidth and power, side by side.

Specifications

Specification	Fujitsu PRIMERGY GX2580 M8s	NVIDIA HGX B300 8-GPU
Vendor	Fujitsu	NVIDIA
Part number	GX2580 M8s	—
Form factor	rack	baseboard
Released	—	2025
Accelerator	8x B300	8x B300
Aggregate memory	—	2.1 TB
Aggregate bandwidth	—	64.0 TB/s
System power	—	—
Interconnect	One NVIDIA SXM HGX B300 8-GPU baseboard integrating eight Blackwell Ultra SXM modules with high-speed interconnect; scale-out is eight NVIDIA ConnectX-8 OSFP 800GbE ports, alongside two 10GbE RJ45 on an Intel X710	NVLink 5.0 / NVSwitch 4.0

Rows shown in grey are identical between the two.

Fujitsu PRIMERGY GX2580 M8s

Fujitsu's Blackwell Ultra machine, built on one NVIDIA SXM HGX B300 8-GPU baseboard integrating eight Blackwell Ultra SXM modules. Networking is the notable part: eight NVIDIA ConnectX-8 OSFP ports at 800GbE apiece, one per accelerator, so scale-out matches the baseboard rather than bottlenecking it. Hosting is two Intel Xeon 6700 or 6500P processors reaching 86 cores and 172 threads, with 32 DIMM slots taking up to 8 TB of DDR5-6400. Storage is eight hot-plug E1.S NVMe bays holding up to 123 TB plus two M.2 devices, and there are two front full-height half-length PCIe 5.0 x16 slots. The 8U chassis measures 483 by 980 by 356 mm and weighs up to 137 kg with rails, drawing from six 6,600 W Titanium supplies in 3+3 redundancy. Fujitsu states no total GPU memory, no throughput figure and no system power draw; baseboard performance is on the linked HGX B300 entry.

Vendor product page

NVIDIA HGX B300 8-GPU

The eight-GPU Blackwell Ultra baseboard OEMs build around. NVIDIA publishes both a dense and a sparse figure for every tensor precision here, and for FP4 the two are not a factor of two apart: dense rose by half over the B200, from 72 to 108 PFLOPS, while sparse stayed at 144. Total memory is NVIDIA's own 2.1 TB rather than the 2.304 TB that eight 288 GB GPUs would suggest. Memory bandwidth of 64 TB/s is the one derived number, being eight times the B300's published 8 TB/s, because NVIDIA gives no board-level HBM figure. No INT8 row is recorded: NVIDIA's table gives 3 POPS sparse for this board while our B300 chip entry implies 36 POPS dense, the two cannot both be right, and the question is open rather than guessed at. NVIDIA publishes no baseboard power figure, so total power is blank.