NVIDIA · Hopper (Grace Hopper)
Best AI Models for NVIDIA GH200 Grace Hopper Superchip (144 GB)
Runs everything up to 120B MoE well.
144 GB HBM3e4900 GB/s990 TFLOPS FP161000 WHopper (Grace Hopper) 2024
Grace Hopper superchip: 144 GB HBM3e GPU pool plus a large NVLink-coherent LPDDR5X extension. Server/cloud part, listed for comparison.
| # | Model | Speed | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Inkling SmallThinking Machines Q3_K_M · 132.8 GB · 389 t/s | 389 tok/s | Q3_K_M · 132.8 GB | |
| 3 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 4 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 156 t/s | 156 tok/s | Q4_K_M · 21.7 GB | |
| 5 | Q4_K_M · 19.3 GB · 302 t/s | 302 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Coding Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 156 t/s | 156 tok/s | Q4_K_M · 21.7 GB | |
| 3 | Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 322 t/s | 322 tok/s | Q4_K_M · 22.0 GB | |
| 4 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 398 t/s | 398 tok/s | Q4_K_M · 16.5 GB | |
| 5 | Q4_K_M · 19.3 GB · 302 t/s | 302 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Reasoning Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Inkling SmallThinking Machines Q3_K_M · 132.8 GB · 389 t/s | 389 tok/s | Q3_K_M · 132.8 GB | |
| 3 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 156 t/s | 156 tok/s | Q4_K_M · 21.7 GB | |
| 4 | Qwen3.5 27BAlibaba Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 5 | Q4_K_M · 19.3 GB · 302 t/s | 302 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Math Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Inkling SmallThinking Machines Q3_K_M · 132.8 GB · 389 t/s | 389 tok/s | Q3_K_M · 132.8 GB | |
| 3 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 4 | Qwen3.6 35B A3BAlibaba Q4_K_M · 22.0 GB · 322 t/s | 322 tok/s | Q4_K_M · 22.0 GB | |
| 5 | Q4_K_M · 19.3 GB · 302 t/s | 302 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Long context Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 3 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 156 t/s | 156 tok/s | Q4_K_M · 21.7 GB | |
| 4 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 398 t/s | 398 tok/s | Q4_K_M · 16.5 GB | |
| 5 | Q4_K_M · 19.3 GB · 302 t/s | 302 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed ↓ | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 977 t/s | 977 tok/s | Q4_K_M · 3.4 GB | |
| 2 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB · 501 t/s | 501 tok/s | Q4_K_M · 6.6 GB | |
| 3 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 469 t/s | 469 tok/s | Q4_K_M · 13.1 GB | |
| 4 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 398 t/s | 398 tok/s | Q4_K_M · 16.5 GB | |
| 5 | Q4_K_M · 19.3 GB · 302 t/s | 302 tok/s | Q4_K_M · 19.3 GB |
| # | Model | Speed | Vision Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.8 27BAlibabaHow to run → Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 2 | Inkling SmallThinking Machines Q3_K_M · 132.8 GB · 389 t/s | 389 tok/s | Q3_K_M · 132.8 GB | |
| 3 | Qwen3.6 27BAlibaba Q4_K_M · 17.9 GB · 183 t/s | 183 tok/s | Q4_K_M · 17.9 GB | |
| 4 | Gemma 4 31B ITGoogle Q4_K_M · 21.7 GB · 156 t/s | 156 tok/s | Q4_K_M · 21.7 GB | |
| 5 | Q4_K_M · 87.7 GB · 195 t/s | 195 tok/s | Q4_K_M · 87.7 GB |
See all 405 models on the GH200 Grace Hopper Superchip →Best quant that fits · speeds estimated on this card
The GH200 Grace Hopper Superchip runs everything up to 120B MoE well. Hopper (Grace Hopper) generation, released 2024.
GenerationHopper (Grace Hopper)released 2024
GH200 Grace Hopper Superchip strengths and weaknesses
Strengths
- Runs 120B MoE models: GPT OSS 120B at 326 tok/s in 70.8 GB
- Room for long context: Llama 3.3 70B Instruct at Q4 uses 51.3 GB, leaving 92.7 GB free
- Reads long prompts fast: A 4K-token prompt on Qwen3.8 27B starts answering in 511 ms
Weaknesses
- 1000 W under load: Plan a 2,000 W PSU; ≈$219 a year at 4 h a day
Output speed depends on memory bandwidth (4900 GB/s).
- 9BQwen3.5 9B · Q4_K_M · 6.6 GB501tok/s
- 14BQwen3 14B · Q4_K_M · 11.9 GB335tok/s
- 27–31BQwen3.8 27B · Q4_K_M · 17.9 GB183tok/s
- 35B MoEQwen3.6 35B A3B · Q4_K_M · 22.0 GB322tok/s
- 70BLlama 3.3 70B Instruct · Q4_K_M · 51.3 GB68.4tok/s
- 120B MoEGPT OSS 120B · Q4_K_M · 70.8 GB326tok/s
Estimated speeds: Q4_K_M, 4K context, one user. How we estimate
All 1296 models on the GH200 Grace Hopper Superchip
Featured onlyBest quant that fits, ordered by llmrun Score. Fit grades how well it runs here, S best. How ranking works
| Model | Quant | VRAM | Speed | Context | Score | Fit |
|---|---|---|---|---|---|---|
| BF16 | 1.8 GB | 4136.4 t/s | — | S100, Runs great | ||
Aya Expanse 32BChat | Q4_K_M | 25.1 GB | 149.4 t/s | — | S97, Runs great | |
Tiny DeepseekV3ForCausalLMChatFunctions | Q4_K_M | 0.3 GB | 10616.7 t/s | 164K | S100, Runs great | |
Dhara 70MChatCode | BF16 | 0.5 GB | 6500.0 t/s | 1K | S100, Runs great | |
Llava 1.5 7B HFVision | Q4_K_M | 5.0 GB | 683.5 t/s | 4K | S100, Runs great | |
VibeThinker 1.5BChatMathCode | Q4_K_M | 1.8 GB | 2243.0 t/s | 131K | S100, Runs great | |
UserLM 8BChat | Q4_K_M | 6.2 GB | 590.9 t/s | 8K | S100, Runs great | |
Devstral 2 123B Instruct 2512ChatFunctions | Q4_K_M | 81.2 GB | 41.9 t/s | 262K | A76, Runs well | |
| Q4_K_M | 11.9 GB | 312.3 t/s | 4K | S100, Runs great | ||
| Q4_K_M | 24.7 GB | 279.3 t/s | — | S100, Runs great | ||
| Q4_K_M | 1.6 GB | 4083.3 t/s | 41K | S100, Runs great | ||
| Q4_K_M | 46.6 GB | 71.4 t/s | 8K | S86, Runs great | ||
| Q4_K_M | 7.6 GB | 577.0 t/s | 41K | S100, Runs great | ||
| Q4_K_M | 7.3 GB | 590.9 t/s | 131K | S100, Runs great | ||
Granite 4.0 1BChatFunctions | Q4_K_M | 2.6 GB | 2196.6 t/s | 131K | S100, Runs great |
Specs
- Architecture
- Hopper (Grace Hopper)
- VRAM
- 144 GB HBM3e
- Memory bandwidth
- 4900.0 GB/s
- FP16 (dense)
- 989.50 TFLOPS
- CUDA cores
- 16,896
- Tensor cores
- 528
- Compute capability
- 9.0
- TDP
- 1000 W
- Released
- 2024-01-01
GH200 Grace Hopper Superchip FAQ
What is the largest model the GH200 Grace Hopper Superchip can run?
Qwen3 235B A22B Thinking 2507 (235.1B) is the largest at 4-bit or better: at Q4_K_M it uses 142.9 GB of 144 GB, and runs at about 107 tok/s. At 3-bit, a bigger model fits too: DeepSeek V4 Flash 0731 (304.2B) at IQ3_M, 137.4 GB.
Can the GH200 Grace Hopper Superchip run Qwen3.8 27B?
Yes. Qwen3.8 27B fits at Q4_K_M, using 17.9 GB of 144 GB, and it is the best model that runs well on this card.
How many tokens per second does the GH200 Grace Hopper Superchip give?
About 183 tok/s on Qwen3.8 27B at Q4_K_M, with the first token after 0.5 s on a 4K-token prompt.
What power supply does the GH200 Grace Hopper Superchip need?
Plan a 2000 W PSU (2× the 1000 W TDP).
Is 144 GB enough for coding and agents?
Yes for coding: Qwen3.8 27B, the best of the 14 coding models that fit, runs at about 183 tok/s.