← All hardware
Grace CPU + Hopper GPU fused over NVLink-C2C
Pros
- Huge coherent memory pool
- Excellent for recommender/graph AI
- Tight CPU-GPU coupling
- Available on several clouds
Cons
- Niche vs pure GPU racks
- Hopper-gen GPU
- Limited cloud availability
- Superseded by Grace Blackwell
✓ Where it shines / best for
- Memory-bound inference
- HPC + AI hybrid
- Recommender systems
- Large embedding tables
✕ Not the best fit for
- Pure dense training at frontier scale
- Budget deployments
- Edge
- Air-cooled small clusters
Features
- ✓ Grace Arm CPU + Hopper GPU
- ✓ NVLink-C2C 900 GB/s coherent link
- ✓ 96GB HBM3 GPU memory
- ✓ Up to 480GB LPDDR5X CPU memory
- ✓ Unified memory addressing
- ✓ Strong for memory-bound inference
Pricing
| Plan | Price | Billing | Notes |
|---|---|---|---|
| GH200 superchip | ~$40,000+ | one-time | Cloud from ~$1.99-$6.50/hr; outright price varies by OEM |
Pricing verified from the official source. Prices change often — confirm on the vendor's site before buying.
Specifications
| use | Memory-bound AI and HPC |
| power | ~1,000W |
| memory | 96GB HBM3 + up to 480GB LPDDR5X |
| performance | ~4 PFLOPS FP8 (GPU) |
| architecture | Grace (Arm) + Hopper |
Sponsored
A full review is being generated for this product and will appear here shortly.