GeekRadar / The comparison desk
COMPARE.
Accelerators
NVIDIA

NVIDIA Rubin Ultra GPU

Accelerators

Process node
3 nm*

Processing & platform

Process node
3 nm*
Reported chip die area
2,912 mm²*
Transistors
670,000 million*
GPU L2 cache
128 MB*
Texture mapping units
1,792*
Raster output units
24*
Streaming multiprocessors
448*
Tensor cores
1,792*
Chip foundry
TSMC*
Card or module format
SXM Module*
Suggested system PSU
4,000 W*
OpenCL
3.0*
CUDA compute capability
10.7*
Compute processor
GR110*
Reported accelerator cores
57,344*
Device memory
1,024 GB*
Device memory type
HBM4e*
Host interface
PCIe 6.0 x16*
Declared accelerator TDP
3,600 W*

Memory & storage

Memory bus
32,768 bit*
Rated memory bandwidth
31,100 GB/s*

System configuration

Detailed specifications

Specifications · 29 fields
CUDA compute capability
10.7*
OpenCL
3.0*
Host interface
PCIe 6.0 x16*
Memory bus
32,768* bit
Reported accelerator cores
57,344*
Reported theoretical FP16 peak
520.0* TFLOPS
Scope and conditions
Arithmetic density
Not stated*
Reported theoretical FP32 peak
260.0* TFLOPS
Scope and conditions
Arithmetic density
Not stated*
Reported theoretical FP64 peak
65.0* TFLOPS
Scope and conditions
Arithmetic density
Not stated*
Declared L1 cache configuration
256 KB (per SM)*
Declared GPU L2 cache
128 MB*
Reported display outputs
No outputs*
Theoretical pixel fill rate
54.41 GPixel/s*
Reported chip die area
2912 mm²*
Chip foundry
TSMC*
Process node declaration
3 nm*
Transistors
670,000 million*
Compute processor
GR110*
Declared device memory
1,024* GB
Rated memory bandwidth
31.1 TB/s*
Device memory type
HBM4e*
Raster output units
24*
Reported shading units
57,344*
Card or module format
SXM Module*
Streaming multiprocessors
448*
Suggested system PSU
4000 W*
Declared accelerator TDP
3600 W*
Tensor cores
1,792*
Theoretical texture fill rate
4,062.5 GTexel/s*
Texture mapping units
1,792*

* Provisional data