GeekRadar / The comparison desk

COMPARE.

GPU

NVIDIA

NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

Graphics processor

No matched benchmark results are available for this entry yet.

Processing & platform

Architecture: Blackwell Ultra^*
Video memory per GPU: 279 GB^*

Memory & storage

Video memory type: HBM3E^*

System configuration

Detailed specifications

Specifications 1 · 27 fields

GPU architecture — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

Blackwell Ultra^*

Calculated theoretical peak: BF16 tensor dense per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

2,500^* TFLOPS

Scope and conditions

Calculation: 5 / 2^*

Theoretical peak: BF16 tensor with sparsity per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

5,000^* TFLOPS

Calculated theoretical peak: FP16 tensor dense per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

2,500^* TFLOPS

Scope and conditions

Calculation: 5 / 2^*

Theoretical peak: FP16 tensor with sparsity per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

5,000^* TFLOPS

Theoretical peak: FP32 non-tensor dense per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

80^* TFLOPS

Theoretical peak: FP4 tensor dense per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

15,000^* TFLOPS

Theoretical peak: FP4 tensor with sparsity per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

20,000^* TFLOPS

Theoretical peak: FP64 non-tensor dense per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

1.3^* TFLOPS

Theoretical peak: FP64 tensor dense per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

1.3^* TFLOPS

Calculated theoretical peak: FP6 tensor dense per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

5,000^* TFLOPS

Scope and conditions

Calculation: 10 / 2^*

Theoretical peak: FP6 tensor with sparsity per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

10,000^* TFLOPS

Calculated theoretical peak: FP8 tensor dense per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

5,000^* TFLOPS

Scope and conditions

Calculation: 10 / 2^*

Theoretical peak: FP8 tensor with sparsity per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

10,000^* TFLOPS

Calculated theoretical peak: INT8 tensor dense per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

165^* TOPS

Scope and conditions

Calculation: 330 / 2^*

Theoretical peak: INT8 tensor with sparsity per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

330^* TOPS

Calculated theoretical peak: TF32 tensor dense per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

1,250^* TFLOPS

Scope and conditions

Calculation: 2.5 / 2^*

Theoretical peak: TF32 tensor with sparsity per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

2,500^* TFLOPS

Physical configuration — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

Grace Blackwell Ultra Superchip, NVL72 rack^*

GPU interconnect bandwidth (NVLink 5) bidirectional — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

1,800^* GB/s

GPU interconnect bandwidth per direction — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

900^* GB/s

PCIe interface declaration — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

PCIe Gen6: 256 GB/s^*

Maximum board power — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

1,400^* W

Peak memory bandwidth per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

8,000^* GB/s

Memory capacity per GPU — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

279^* GB

Memory technology — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

HBM3E^*

GPU model — NVIDIA GB300 NVL72 GPU component (279 GB per GPU)

NVIDIA Blackwell Ultra GPU (GB300 NVL72, individual GPU)^*

Specifications 2 · 1 fields

Nominal GPU Memory

288^* GB · HBM3e^*

Scope and conditions

native_context

Component scope: GPU component only; CPU and complete system are separate^*
Memory scope: Nominal per-GPU memory; not the total CPU + GPU module memory or runtime-available capacity^*
memory role: nominal_release_inventory^*
not dedicated VRAM for shared memory systems: Yes^*
not runtime available or HGX usable capacity: Yes^*

* Provisional data