NVIDIA GB200 NVL72 GPU 186GB vs NVIDIA GB200 Grace Blackwell Superchip 372GB

The GB200 Grace Blackwell Superchip 372GB delivers 2.0x the FP8 throughput of the GB200 NVL72 GPU 186GB (10,000 vs 5,000 TFLOPS dense). The GB200 Grace Blackwell Superchip 372GB also carries 186 GB more memory (372 GB vs 186 GB).

GB200 NVL72 GPU 186GB: Blackwell, 2024 GB200 Grace Blackwell Superchip 372GB: Blackwell, 2024
FP8 dense lead
2.0x
GB200 Grace Blackwell Superchip 372GB: 10,000 vs 5,000 TFLOPS
VRAM
186 vs 372 GB
186 GB more for the GB200 Grace Blackwell Superchip 372GB
Bandwidth
8.0 TB/s vs 16.0 TB/s
GB200 Grace Blackwell Superchip 372GB moves data faster

Specifications

GB200 NVL72 GPU 186GB
GB200 Grace Blackwell Superchip 372GB
Architecture
BlackwellBlackwell
Launch Year
20242024
Form Factor
SXMSuperchip
VRAM
186 GB372 GB
Memory Bandwidth
8.0 TB/s16.0 TB/s
TDP
Process Node
4nm4nm

Performance (TFLOPS)

GB200 NVL72 GPU 186GB
GB200 Grace Blackwell Superchip 372GB
FP64
40.0 TFlops 80.0 TFlops
FP32
80.0 TFlops 160.0 TFlops
TF32
1250.0 TFlops
2500.0 TFLOPS sparse
2500.0 TFlops
5000.0 TFLOPS sparse
BF16
2500.0 TFlops
5000.0 TFLOPS sparse
5000.0 TFlops
10000.0 TFLOPS sparse
FP16
2500.0 TFlops
5000.0 TFLOPS sparse
5000.0 TFlops
10000.0 TFLOPS sparse
FP8
5000.0 TFlops
10000.0 TFLOPS sparse
10000.0 TFlops
20000.0 TFLOPS sparse
FP6
5000.0 TFlops
10000.0 TFLOPS sparse
10000.0 TFlops
20000.0 TFLOPS sparse
FP4
10000.0 TFlops
20000.0 TFLOPS sparse
20000.0 TFlops
40000.0 TFLOPS sparse
INT8
5000.0 TFlops
10000.0 TOPS sparse
10000.0 TFlops
20000.0 TOPS sparse

FLOPS by Precision

What actually differs

Both GPUs launched in 2024: the GB200 NVL72 GPU 186GB on Blackwell and the GB200 Grace Blackwell Superchip 372GB on Blackwell.

Dense throughput for the GB200 NVL72 GPU 186GB against the GB200 Grace Blackwell Superchip 372GB: FP64 40 vs 80 TFLOPS, FP32 80 vs 160 TFLOPS, FP16 2,500 vs 5,000 TFLOPS, FP8 5,000 vs 10,000 TFLOPS. Sparse figures, where the vendor publishes them, appear under each dense number in the performance table.

Memory is 186 GB against 372 GB, fed at 8.0 TB/s versus 16.0 TB/s. More memory per GPU means larger models fit before you have to shard across cards, which often matters more than raw TFLOPS for inference.

For LLM training and inference, weight the FP8 and FP16 rows and memory capacity most heavily. For scientific and HPC workloads, the FP64 row is the one to read.

Frequently asked questions

Is the GB200 NVL72 GPU 186GB faster than the GB200 Grace Blackwell Superchip 372GB?

At FP8 precision the GB200 Grace Blackwell Superchip 372GB reaches 10,000 TFLOPS dense against 5,000 TFLOPS for the GB200 NVL72 GPU 186GB. The performance table on this page lists every published precision for both GPUs.

Which has more memory, the GB200 NVL72 GPU 186GB or the GB200 Grace Blackwell Superchip 372GB?

The GB200 Grace Blackwell Superchip 372GB carries 372 GB of VRAM versus 186 GB for the GB200 NVL72 GPU 186GB. Memory bandwidth is 8.0 TB/s for the GB200 NVL72 GPU 186GB and 16.0 TB/s for the GB200 Grace Blackwell Superchip 372GB.

Get Comparison Updates

New GPUs added weekly. Be the first to see how they compare.