NVIDIA · Blackwell Ultra
Canada · 224 vCPU · 1920 GB RAM · 1000GB root + 40000GB ephemeral
Stock and price checked 6 min ago.
You start it yourself once there is stock.
The hardware type or the provider confirms a fast GPU-to-GPU link. Software still has to split the work across GPUs. They never appear as one big memory.
n3-B300-SXM6x8
France · 224 vCPU · 3840 GB RAM · 20 Gbit/s · Block, Scratch
Stock and price checked 6 min ago.
The hardware type or the provider confirms a fast GPU-to-GPU link. Software still has to split the work across GPUs. They never appear as one big memory.
B300-SXM-8-288G
Excluding VAT.
Each provider counts once. Full price and spot are separate.
Full price. Offers we can't say are really sold there are left out until they've been in stock once.
No provider reports a number for this GPU.
In stock, yes or no, for every provider selling this GPU. A filled band means in stock: green at full price, orange as spot.
Built to be linked. All GPUs in a server are connected with NVLink 5 at up to 1800 GB/s. Software still has to split the work, so the memory doesn't act as one big pool.
NVIDIA states up to 160 SMs at 128 CUDA cores per SM, varying by SKU, so core count is null. NVIDIA node memory totals conflict (2.1 vs 2.3 TB). Per-GPU compute figures are derived by dividing NVIDIA's 8-GPU HGX B300 totals by 8; FP4, FP8, FP16, BF16 and INT8 are with sparsity. memoryDataRateGbps is derived (8 TB/s x 8 / 8192 bits). pcieBandwidthGBs is PCIe Gen5 x16, 128 GB/s per NVIDIA.