GeekRadar / The comparison desk

COMPARE.

Component references

NVIDIA

NVIDIA GB200 Grace Blackwell Superchip

Component references

System configuration

Detailed specifications

Specifications 1 · 7 fields

Graphics model

GB200^*

Scope and conditions

native_context

Physical scope: 1 CPU + 2 GPU superchip^*

Full graphics designation

NVIDIA GB200^*

Scope and conditions

native_context

Physical scope: 1 CPU + 2 GPU superchip^*

Manufacturer

NVIDIA^*

Scope and conditions

native_context

Physical scope: 1 CPU + 2 GPU superchip^*

Form factor declaration

Superchip module (Grace CPU + 2 Blackwell GPUs)^*

Scope and conditions

native_context

Physical scope: 1 CPU + 2 GPU superchip^*

Scoped memory capacity

372^* GB

Scope and conditions

native_context

Capacity unit base: Not independently established^*
Physical scope: 1 CPU + 2 GPU superchip^*

Memory technology

HBM3E^*

Scope and conditions

native_context

Physical scope: 1 CPU + 2 GPU superchip^*

Package power declaration

2,700^* W

Scope and conditions

native_context

Physical scope: 1 CPU + 2 GPU superchip^*

Specifications 2 · 13 fields

INT8 Tensor Core 2

20 POPS^*

Scope and conditions

native_context

performance is theoretical not benchmark: Yes^*
resource scope: superchip_1_CPU_2_GPU^*
sparse and dense figures not interchangeable: Yes^*

TF32 Tensor Core 2

5 PFLOPS^*

Scope and conditions

native_context

performance is theoretical not benchmark: Yes^*
resource scope: superchip_1_CPU_2_GPU^*
sparse and dense figures not interchangeable: Yes^*

FP16/BF16 Tensor Core 2

10 PFLOPS^*

Scope and conditions

native_context

performance is theoretical not benchmark: Yes^*
resource scope: superchip_1_CPU_2_GPU^*
sparse and dense figures not interchangeable: Yes^*

NVFP4 Tensor Core 2

40 | 20 PFLOPS^*

Scope and conditions

native_context

performance is theoretical not benchmark: Yes^*
resource scope: superchip_1_CPU_2_GPU^*
sparse and dense figures not interchangeable: Yes^*

FP8/FP6 Tensor Core 2

20 PFLOPS^*

Scope and conditions

native_context

performance is theoretical not benchmark: Yes^*
resource scope: superchip_1_CPU_2_GPU^*
sparse and dense figures not interchangeable: Yes^*

Configuration

1 Grace CPU | 2 Blackwell GPU^*

Scope and conditions

native_context

performance is theoretical not benchmark: Yes^*
resource scope: superchip_1_CPU_2_GPU^*
sparse and dense figures not interchangeable: Yes^*

FP32

160 TFLOPS^*

Scope and conditions

native_context

performance is theoretical not benchmark: Yes^*
resource scope: superchip_1_CPU_2_GPU^*
sparse and dense figures not interchangeable: Yes^*

CPU Memory | Bandwidth

values

unit
GB^*
value
amount
480^*
memory_type
LPDDR5X^*
qualifier
up_to^*
type
memory_quantity^*
unit
GB/s^*
value
amount
512^*
qualifier
up_to^*
type
quantity^*

Scope and conditions

native_context

performance is theoretical not benchmark: Yes^*
resource scope: superchip_1_CPU_2_GPU^*
sparse and dense figures not interchangeable: Yes^*

GPU Memory | Bandwidth

values

unit
GB^*
value
amount
372^*
memory_type
HBM3E^*
type
memory_quantity^*
unit
TB/s^*
value
amount
16^*
type
quantity^*

Scope and conditions

native_context

performance is theoretical not benchmark: Yes^*
resource scope: superchip_1_CPU_2_GPU^*
sparse and dense figures not interchangeable: Yes^*

Nominal GPU Memory

192^* GB · HBM3e^*

Scope and conditions

native_context

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^*

NVLink Bandwidth

3.6^* TB/s

Scope and conditions

native_context

performance is theoretical not benchmark: Yes^*
resource scope: superchip_1_CPU_2_GPU^*
sparse and dense figures not interchangeable: Yes^*

CPU Core Count

72 Arm Neoverse V2 cores^*

Scope and conditions

native_context

performance is theoretical not benchmark: Yes^*
resource scope: superchip_1_CPU_2_GPU^*
sparse and dense figures not interchangeable: Yes^*

FP64 / FP64 Tensor Core

80 TFLOPS^*

Scope and conditions

native_context

performance is theoretical not benchmark: Yes^*
resource scope: superchip_1_CPU_2_GPU^*
sparse and dense figures not interchangeable: Yes^*

* Provisional data