GeekRadar / Il banco di confronto
CONFRONTA.
Sistemi testati
NVIDIA

NVIDIA GB200 NVL72

Sistema testato

Configurazione di sistema

Specifiche dettagliate

Specifiche · 12 campi
FP32
5,760 TFLOPS*
Ambito e condizioni
native_context
performance is theoretical not benchmark
Sì*
resource scope
rack_36_CPU_72_GPU*
sparse and dense figures not interchangeable
Sì*
CPU Memory | Bandwidth
values
  • unit
    TB*
    value
    amount
    17*
    memory_type
    LPDDR5X*
    type
    memory_quantity*
  • unit
    TB/s*
    value
    amount
    14*
    type
    quantity*
Ambito e condizioni
native_context
performance is theoretical not benchmark
Sì*
resource scope
rack_36_CPU_72_GPU*
sparse and dense figures not interchangeable
Sì*
NVFP4 Tensor Core 2
1,440 | 720 PFLOPS*
Ambito e condizioni
native_context
performance is theoretical not benchmark
Sì*
resource scope
rack_36_CPU_72_GPU*
sparse and dense figures not interchangeable
Sì*
NVLink Bandwidth
130* TB/s
Ambito e condizioni
native_context
performance is theoretical not benchmark
Sì*
resource scope
rack_36_CPU_72_GPU*
sparse and dense figures not interchangeable
Sì*
FP8/FP6 Tensor Core 2
720 PFLOPS*
Ambito e condizioni
native_context
performance is theoretical not benchmark
Sì*
resource scope
rack_36_CPU_72_GPU*
sparse and dense figures not interchangeable
Sì*
FP64 / FP64 Tensor Core
2,880 TFLOPS*
Ambito e condizioni
native_context
performance is theoretical not benchmark
Sì*
resource scope
rack_36_CPU_72_GPU*
sparse and dense figures not interchangeable
Sì*
GPU Memory | Bandwidth
values
  • unit
    TB*
    value
    amount
    13.4*
    memory_type
    HBM3E*
    type
    memory_quantity*
  • unit
    TB/s*
    value
    amount
    576*
    type
    quantity*
Ambito e condizioni
native_context
performance is theoretical not benchmark
Sì*
resource scope
rack_36_CPU_72_GPU*
sparse and dense figures not interchangeable
Sì*
TF32 Tensor Core 2
180 PFLOPS*
Ambito e condizioni
native_context
performance is theoretical not benchmark
Sì*
resource scope
rack_36_CPU_72_GPU*
sparse and dense figures not interchangeable
Sì*
FP16/BF16 Tensor Core 2
360 PFLOPS*
Ambito e condizioni
native_context
performance is theoretical not benchmark
Sì*
resource scope
rack_36_CPU_72_GPU*
sparse and dense figures not interchangeable
Sì*
CPU Core Count
2,592 Arm® Neoverse V2 cores*
Ambito e condizioni
native_context
performance is theoretical not benchmark
Sì*
resource scope
rack_36_CPU_72_GPU*
sparse and dense figures not interchangeable
Sì*
INT8 Tensor Core 2
720 POPS*
Ambito e condizioni
native_context
performance is theoretical not benchmark
Sì*
resource scope
rack_36_CPU_72_GPU*
sparse and dense figures not interchangeable
Sì*
Configuration
36 Grace CPU | 72 Blackwell GPUs*
Ambito e condizioni
native_context
performance is theoretical not benchmark
Sì*
resource scope
rack_36_CPU_72_GPU*
sparse and dense figures not interchangeable
Sì*

* Dati provvisori