B300 Specifications
Blackwell Ultra. 288 GB HBM3e is the headline — enough for very large models on a single module. Figures are from announcement and may move before volume shipping.
Announced — specifications may change before launch
Figures are taken from the NVIDIA announcement and are not yet confirmed against a shipping datasheet. Clocks, power and throughput commonly move between announcement and volume availability.
Buy it for
Frontier-scale training and long-context inference, where 288 GB per module and the higher FP4 throughput cut the GPU count a given model needs.
Look elsewhere if
Your facility cannot supply and cool it. At 1.4 kW per module this is liquid-cooled, high-density rack territory — air-cooled halls generally cannot take it.
Will it fit your system?
This is not a card you can install
SXM is a socketed module that mounts to an OEM baseboard, not a PCIe slot. You buy it inside a complete system — it cannot go into an existing desktop or a standard server under any circumstances.
- Form factor
- SXM
- Cooling
- liquid
- Power connectors
- None — SXM module, powered by the baseboard
- Board power
- 1400 W
- Interface
- PCIe Gen 6
Supported formats & features
What the B300 does and does not support in hardware.
Hardware FP4 — roughly doubles inference throughput over FP8
E4M3 and E5M2 — the standard format for modern inference
Wider exponent than FP16 — fewer overflow issues when training
Drop-in FP32 speedup with no code changes
Quantised inference supported
Not yet confirmed against vendor documentation.
Per-layer automatic FP8 precision management
1800 GB/s GPU-to-GPU
Compute-only die — no ray tracing hardware
Corrects single-bit errors — expected for long training runs
What fits in memory
Largest model that fits on one B300, assuming ~20% of the 288 GB is reserved for KV cache and activations.
| Model | FP16 / BF16 | FP8 | INT4 / FP4 |
|---|---|---|---|
| Llama 8B | Fits | Fits | Fits |
| Llama 70B | Fits | Fits | Fits |
| Mixtral 8x22B | 2× GPUs | Fits | Fits |
| Llama 405B | 4× GPUs | 2× GPUs | Fits |
Weights only. Long contexts, large batch sizes and optimiser state during training all need substantially more — full fine-tuning typically needs 3–4× the weight memory.
Workload character
Bandwidth and compute are matched closely enough that neither dominates across mixed training and inference.
Cost to run
Electricity for one card at 100% duty, including cooling and PSU overhead (17,170 kWh/year).
Where it sits in the line
A newer part existing does not make this one a bad buy — it makes it cheaper. What matters is price per unit of the throughput you actually need.
Availability & market
Announced; specifications may shift before volume shipment. Early supply goes to the largest buyers, so expect cloud rental before purchase.
Memory
- Capacity
- 288 GB
- Type
- HBM3e
- Bandwidth
- 8,000 GB/s
- Bus width
- —
- GB per TFLOPMemory relative to compute — higher favours large models
- 0.128
Compute
- FP64
- —
- FP32
- —
- TF32 dense
- —
- FP16 dense
- 2,250 TFLOPS
- FP16 sparseRequires structured sparsity; rarely reached in practice
- 4,500 TFLOPS
- FP8 dense
- 4,500 TFLOPS
- INT8 dense
- —
Silicon
- Codename
- GB300
- Process node
- TSMC 4NP
- Transistors
- —
- Die size
- —
- CUDA cores
- —
- Tensor cores
- —
- SM count
- —
- Boost clock
- —
Platform & power
- TDP
- 1,400 W
- TFLOPS per wattDense FP16 ÷ TDP — drives your power bill at scale
- 1.61
- Form factor
- SXM
- Cooling
- liquid
- Interconnect
- NVLink 5.0
- Interconnect speed
- 1,800 GB/s
- PCIe generation
- Gen 6
- MIG
- Yes — up to 7 instances
- Launch
- 2026
- MSRP at launch
- —
B300 — common questions
Answered from the specifications above. Questions this card's data cannot answer with confidence are left out rather than guessed.
- How much VRAM does the B300 have?
- The B300 has 288 GB of HBM3e memory with 8,000 GB/s of bandwidth. In practice plan on roughly 230 GB for weights: KV cache, activations and fragmentation take the rest, and long contexts or large batches take considerably more.
- What size model can the B300 run?
- On a single B300, about 115B parameters at FP16, or roughly 460B quantised to FP4. Anything larger has to be split across several GPUs. These are weight-only figures — a long context window can consume tens of gigabytes on its own.
- Can you put the B300 in a desktop PC?
- No. The B300 is a SXM module, not a PCIe card. It bolts to an OEM baseboard and is only sold inside a complete system, so there is no way to install one in a desktop or a standard server.
- How much power does the B300 use?
- The B300 is rated at 1400 W board power. It draws that from the baseboard rather than from PSU cables, so there is nothing to wire up yourself.
- How much does it cost to run the B300?
- At the US commercial average of $0.12/kWh, running one B300 continuously costs about $2,060 a year in electricity. That figure includes cooling and power-supply overhead at a factor of 1.4 over board power.
- Does the B300 support MIG?
- Yes. The B300 can be partitioned into up to 7 Multi-Instance GPU slices, each with isolated memory and compute. This is what makes it practical to share one card across several users or tenants without them affecting each other's latency.
- Does the B300 support FP8?
- Yes. The B300 has hardware FP8 tensor support, so FP8 weights and activations run at full tensor-core rate — 4,500 dense TFLOPS. This is the format most current inference servers default to for large models.
- Does the B300 support FP4?
- Yes. The B300 supports 4-bit floating point in hardware, which roughly doubles throughput again over FP8 and halves the memory a model occupies.
- Does the B300 support INT8?
- Yes. The B300 supports INT8 acceleration. INT8 remains the most widely supported quantised inference format across frameworks.
- Does the B300 support BF16?
- Yes. The B300 supports bfloat16. Its wider exponent range makes training far less prone to the overflow that plagues FP16, which is why most current training recipes assume it.
- Does the B300 have NVLink?
- Yes. The B300 supports NVLink (NVLink 5.0) at 1,800 GB/s, so GPU-to-GPU traffic bypasses PCIe. This matters for tensor-parallel serving, where every token crosses the link.
- Does the B300 have RT cores?
- No. The B300 is a compute-only part with no ray-tracing hardware. For rendering or virtual workstations, look at the L40S or an RTX workstation card instead.
- Does the B300 have ECC memory?
- Yes. The B300 has ECC, so single-bit memory errors are corrected rather than silently corrupting a long run.
- What replaced the B300?
- The Rubin R100 (2027) is its successor, with 288 GB of HBM4.. On the used market that gap is what sets a fair price for the B300.
- Is the B300 worth buying?
- Buy it for: Frontier-scale training and long-context inference, where 288 GB per module and the higher FP4 throughput cut the GPU count a given model needs. Look elsewhere if: Your facility cannot supply and cool it. At 1.4 kW per module this is liquid-cooled, high-density rack territory — air-cooled halls generally cannot take it.