NVIDIA GH200 144GB HBM3e
Overview
The 144 GB HBM3e configuration of the NVIDIA GH200 Grace Hopper Superchip, and the one most cloud providers rent. It keeps the 72-core Grace CPU, the Hopper H100 GPU and the 900 GB/s NVLink-C2C link of the 96 GB part, and changes only the memory system: 144 GB of HBM3e at 4.9 TB/s in place of 96 GB of HBM3 at 4 TB/s. With up to 480 GB of LPDDR5X on the CPU side that gives 624 GB of fast-access memory in one coherent address space. Compute is unchanged, because NVIDIA publishes a single performance table for both configurations, so the 50 percent extra capacity and 23 percent extra bandwidth are the whole of the difference. That matters most where the working set rather than the arithmetic is the constraint, which is why this part shows up under long-context inference and large-embedding retrieval rather than under training benchmarks.
Performance
Peak theoretical throughput by precision type
| Precision | Dense | 2:4 Sparse |
|---|---|---|
FP64 64-bit floating point | 34TFLOPS | Structured sparsity is a tensor-core feature; this vector precision has no sparse form |
FP64 Tensor 64-bit floating point with Tensor Cores | 67TFLOPS | Structured sparsity is a tensor-core feature; this vector precision has no sparse form |
FP32 32-bit floating point | 67TFLOPS | Structured sparsity is a tensor-core feature; this vector precision has no sparse form |
TF32 TensorFloat-32 | 494TFLOPS | 989TFLOPS |
BF16 Brain Float 16 | 990TFLOPS | 1,979TFLOPS |
FP16 16-bit floating point | 990TFLOPS | 1,979TFLOPS |
FP8 8-bit floating point | 1,979TFLOPS | 3,958TFLOPS |
FP6 | No verified data available | The part supports 2:4 sparsity, but the vendor publishes no sparse figure for this precision |
FP4 | No verified data available | The part supports 2:4 sparsity, but the vendor publishes no sparse figure for this precision |
INT8 8-bit integer | 1,979TOPS | 3,958TOPS |
Dense peak divided by accelerator TDP. Board power only: excludes host CPUs, networking, cooling and facility overhead. TDP for this part: 1000 W.
The GH200 144GB HBM3e in the GPU landscape
Peak FP16 TFLOPS (dense) against TDP, single-GPU parts tracked by Flopper
Higher and further left is better: more half-precision throughput for less power.
Specifications
Architecture
Hopper
Form Factor
Superchip
Launch Year
2024
Process Node
4nm
Memory
144 GB HBM3e
Bandwidth
4,900 GB/s
TDP
1.0 kW
Max power (Flopper estimate)
~1.1 kW est. Flopper estimate: 1000 W TDP x 1.15. The vendor publishes no maximum board power for this part.
Interconnect
900 GB/s NVLink-C2C
bidirectional, per GPU
Spec Confidence
Official
Full Specifications
| Memory | |
|---|---|
| Memory | 144 GB |
| Memory Type | HBM3e |
| Bandwidth | 4.9 TB/s |
| Interface Width | No verified data available |
| Interconnect & I/O | |
| GPU-to-GPU | NVLink-C2C |
| Interconnect Bandwidth | 900 GB/s bidirectional, per GPU |
| Power & Thermal | |
| TDP | 1.0 kW |
| Max power (Flopper estimate) | ~1.1 kW est. |
| Enterprise Features | |
| Sparsity | Yes |
| General | |
| Form Factor | Superchip |
| Architecture | Hopper |
| Process Node | 4nm |
| Launch Year | 2024 |
Datasheet & Resources
Flopper Datasheet
NVIDIA GH200 144GB HBM3e specifications, generated from our database. Printable.
Vendor product page
NVIDIA documentation for the GH200 144GB HBM3e
NVIDIA GH200 Grace Hopper Superchip Datasheet (doc 3154100, rev Mar24)
NVIDIA · 2024-03-01
NVIDIA Developer Documentation
Technical specs, programming guides
Data Provenance
Every figure traced to a source
Primary Source
- Publisher
- NVIDIA
- Published
- 2024-03-01
Data Quality
- Spec confidence
- Official
- Clock basis
- Boost
- Core precisions with figures
- 7 of 9
- Normalization
- All values in TFLOPS
Compare cloud providers offering on-demand GPU instances for AI training, inference, and HPC workloads.
Browse GPU Cloud ProvidersSimilar GPUs
Frequently Asked Questions
How many TFLOPS does the NVIDIA GH200 have?
The NVIDIA GH200 delivers 67 TFLOPS FP32, 990 TFLOPS FP16 and 1,979 TFLOPS FP8 at peak.
What is the power consumption of the NVIDIA GH200?
The NVIDIA GH200 has a TDP (Thermal Design Power) rating of 1,000 watts.
How much memory does the NVIDIA GH200 have?
The NVIDIA GH200 is equipped with 144 GB of memory with 4,900 GB/s of memory bandwidth.
What architecture is the NVIDIA GH200 based on?
The NVIDIA GH200 is based on the Hopper architecture, launched in 2024.
Stay Updated on GPU Releases
Get notified when new GPUs are added or specifications are updated.
No spam, unsubscribe anytime.