NVIDIA HGX B300 8-GPU vs QCT QuantaGrid D75H-10U
Accelerator configuration, aggregate memory, bandwidth and power, side by side.
Specifications
| Specification | NVIDIA HGX B300 8-GPU | QCT QuantaGrid D75H-10U |
|---|---|---|
| Vendor | NVIDIA | QCT |
| Part number | — | D75H-10U |
| Form factor | baseboard | rack |
| Released | 2025 | — |
| Accelerator | 8x B300 | 8x B300 |
| Aggregate memory | 2.1 TB | — |
| Aggregate bandwidth | 64.0 TB/s | — |
| System power | — | — |
| Interconnect | NVLink 5.0 / NVSwitch 4.0 | One NVIDIA HGX B300 8-GPU baseboard; scale-out is eight ConnectX-8 OSFP ports carrying 800G Ethernet or InfiniBand at one port per GPU for GPU-Direct RDMA |
Rows shown in grey are identical between the two.
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.
Vendor product pageQCT QuantaGrid D75H-10U
QCT's 10U Blackwell Ultra server, built on the NVIDIA HGX B300 platform with two Intel Xeon 6 processors. Networking is eight ConnectX-8 OSFP ports at 800G, one per accelerator, for one-to-one GPU-Direct RDMA, and there is room for four full-height full-length PCIe Gen 5 slots. QCT quotes 72 PFLOPS of FP8 and 144 PFLOPS of FP4 without saying whether either is dense or sparse. Both match NVIDIA's sparse figures for this baseboard exactly, and neither matches the dense figures of 36 and 108, so they are recorded as sparse. Worth noting QCT calls the FP8 number a training figure, which would normally imply dense; the number itself settles the basis, not the label. Power is twelve 3,200 W supplies in N+N redundancy, which is capacity rather than draw, so no system power is shown.
Vendor product page