NVIDIA Rubin SXM
Overview
NVIDIA Rubin is the post-Blackwell datacenter GPU announced at GTC 2025 and shipping in late 2026. Each Rubin GPU pairs with HBM4 stacks delivering 22 TB/s bandwidth, ships on the SXM form factor, and supports 6th-gen NVLink at 3.6 TB/s per GPU. Rubin debuts NVFP4, a microscaled 4-bit format that hits 50 PFLOPS dense for inference (35 PFLOPS dense for training) and adds tensor-core-emulated SGEMM/DGEMM that lifts FP32 to 400 TFLOPS and FP64 to 200 TFLOPS. Specifications are preliminary per NVIDIA and subject to change. NVFP4 training throughput on this GPU is 35 PFLOPS (dense).
Performance
Peak theoretical throughput by precision type
| Precision | Dense | 2:4 Sparse |
|---|---|---|
FP64 64-bit floating point | 33TFLOPS | Structured sparsity is a tensor-core feature; this vector precision has no sparse form |
FP64 Tensor 64-bit floating point with Tensor Cores | 200TFLOPS | Structured sparsity is a tensor-core feature; this vector precision has no sparse form |
FP32 32-bit floating point | 130TFLOPS | Structured sparsity is a tensor-core feature; this vector precision has no sparse form |
FP32 Tensor | 400TFLOPS | Structured sparsity is a tensor-core feature; this vector precision has no sparse form |
TF32 TensorFloat-32 | 2,000TFLOPS | The part supports 2:4 sparsity, but the vendor publishes no sparse figure for this precision |
BF16 Brain Float 16 | 4,000TFLOPS | The part supports 2:4 sparsity, but the vendor publishes no sparse figure for this precision |
FP16 16-bit floating point | 4,000TFLOPS | The part supports 2:4 sparsity, but the vendor publishes no sparse figure for this precision |
FP8 8-bit floating point | 17,500TFLOPS | The part supports 2:4 sparsity, but the vendor publishes no sparse figure for this precision |
FP6 | 17,500TFLOPS | The part supports 2:4 sparsity, but the vendor publishes no sparse figure for this precision |
FP4 | No verified data available | The part supports 2:4 sparsity, but the vendor publishes no sparse figure for this precision |
NVFP4 | 35,000TFLOPS | 50,000TFLOPS |
INT8 8-bit integer | 250TOPS | The part supports 2:4 sparsity, but the vendor publishes no sparse figure for this precision |
Specifications
Architecture
Rubin
Form Factor
SXM
Launch Year
2026
Process Node
No verified data available
Memory
288 GB HBM4
Bandwidth
22,000 GB/s
TDP
No verified data available
Max power
No verified data available
Interconnect
3.6 TB/s NVLink
bidirectional, per GPU
Spec Confidence
Vendor claimed
Full Specifications
| Memory | |
|---|---|
| Memory | 288 GB |
| Memory Type | HBM4 |
| Bandwidth | 22.0 TB/s |
| Interface Width | No verified data available |
| Interconnect & I/O | |
| GPU-to-GPU | NVLink |
| Interconnect Bandwidth | 3.6 TB/s bidirectional, per GPU |
| Power & Thermal | |
| TDP | No verified data available |
| Max power | No verified data available |
| Enterprise Features | |
| Sparsity | Yes |
| General | |
| Form Factor | SXM |
| Architecture | Rubin |
| Process Node | No verified data available |
| Launch Year | 2026 |
Datasheet & Resources
Data Provenance
Every figure traced to a source
Primary Source
- Publisher
- NVIDIA
- Published
- 2025-03-18
Data Quality
- Spec confidence
- Vendor claimed
- Clock basis
- Boost
- Core precisions with figures
- 8 of 9
- Normalization
- All values in TFLOPS
Compare cloud providers offering on-demand GPU instances for AI training, inference, and HPC workloads.
Browse GPU Cloud ProvidersSimilar GPUs
Systems Using This GPU
Pre-configured systems featuring the NVIDIA Rubin
| System | GPU Count | Peak Performance | Total Power | |
|---|---|---|---|---|
NVIDIA DGX Rubin NVL8 DGX · rack · 2026 | 8x Rubin | 280 PFLOPS | 24.0 kW | View System |
NVIDIA HGX Rubin NVL8 HGX · baseboard · 2026 | 8x Rubin | 280 PFLOPS | No verified data available | View System |
NVIDIA Vera Rubin NVL72 NVL · rack · 2026 | 72x Rubin | 2,520 PFLOPS | No verified data available | View System |
Inventec Artemis III Rack Artemis · rack · | 72x Rubin | No verified data available | 228.0 kW | View System |
Ingrasys IGS-HXR200 IGS · rack · | 8x Rubin | No verified data available | No verified data available | View System |
Frequently Asked Questions
How many TFLOPS does the NVIDIA Rubin have?
The NVIDIA Rubin delivers 130 TFLOPS FP32, 4,000 TFLOPS FP16 and 17,500 TFLOPS FP8 at peak.
What is the power consumption of the NVIDIA Rubin?
Flopper does not currently have a verified TDP (Thermal Design Power) figure for the NVIDIA Rubin.
How much memory does the NVIDIA Rubin have?
The NVIDIA Rubin is equipped with 288 GB of memory with 22,000 GB/s of memory bandwidth.
What architecture is the NVIDIA Rubin based on?
The NVIDIA Rubin is based on the Rubin architecture, launched in 2026.
How much does the NVIDIA Rubin cost to rent?
Flopper is not tracking any live rental listings for the NVIDIA Rubin. The provider directory lists the clouds that rent comparable accelerators on demand.
When did the NVIDIA Rubin come out?
The NVIDIA Rubin launched in 2026 on the Rubin architecture.
Stay Updated on GPU Releases
Get notified when new GPUs are added or specifications are updated.
No spam, unsubscribe anytime.