Rubin R100 Specifications
Successor architecture to Blackwell, paired with the Vera CPU. HBM4 is confirmed by NVIDIA roadmap; throughput figures are not published, so none are listed here rather than repeating leaks as specifications.
Unreleased — specifications not published
NVIDIA has not published specifications for this part. Only details confirmed on the public roadmap are listed below — figures circulating elsewhere are unverified and are deliberately omitted here rather than reproduced as fact.
Buy it for
Nothing yet — planning only. It is the first part to pair HBM4 with the Vera CPU generation, so it is worth tracking if you are sizing a 2027 build.
Look elsewhere if
You need numbers to plan against. Throughput figures are not published, and anything circulating now is inference from roadmaps rather than datasheet fact.
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
- —
- Interface
- PCIe Gen 6
Supported formats & features
What the Rubin R100 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
Direct GPU-to-GPU link
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 Rubin R100, 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.
Where it sits in the line
Availability & market
Not purchasable. Treat every performance claim about it, including capacity, as provisional.
Memory
- Capacity
- 288 GB
- Type
- HBM4
- Bandwidth
- —
- Bus width
- —
Compute
- FP64
- —
- FP32
- —
- TF32 dense
- —
- FP16 dense
- —
- FP16 sparseRequires structured sparsity; rarely reached in practice
- —
- FP8 dense
- —
- INT8 dense
- —
Silicon
- Codename
- R100
- Process node
- —
- Transistors
- —
- Die size
- —
- CUDA cores
- —
- Tensor cores
- —
- SM count
- —
- Boost clock
- —
Platform & power
- TDP
- —
- Form factor
- SXM
- Cooling
- liquid
- Interconnect
- NVLink 6.0
- Interconnect speed
- —
- PCIe generation
- Gen 6
- MIG
- No
- Launch
- 2027
- MSRP at launch
- —