NVIDIA
Operating mode
Use this to bias workload recommendations toward responsiveness, background autonomy, lighter serving, or multi-GPU scale-out.
Current mode
Balanced
Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.
The NVIDIA H20 is a China-export-compliant Hopper GPU purpose-built for LLM inference in the Chinese market. Its 96 GB of HBM3 at 4.0 TB/s significantly exceeds the H100's 80 GB, while compute is throttled to 148 TFLOPS FP16 — roughly 15% of the H100 — to stay under U.S. export performance thresholds. Despite the compute cap, inference for memory-bandwidth-bound LLM workloads is competitive: the H20 delivers up to 20% higher peak tokens-per-second than H100 at low batch sizes due to its superior memory capacity and bandwidth. Chinese cloud providers built massive H20 clusters for LLM serving workloads.
Beyond LLMs
What AI tasks this GPU can handle — from text generation to image and video creation.
| Capability | Status | Representative Model |
|---|---|---|
| LLM Chat (7B) | Runs natively | Llama 3.1 8B Q4 |
| LLM Coding (30B) | Runs natively | Qwen 3 30B Q4 |
| LLM Large (70B) | Runs natively | Llama 3.1 70B Q4 |
| Image Gen (SDXL) | Runs natively | SDXL 1.0 FP16 |
| Image Gen (Flux) | Runs natively | Flux.1 Dev FP16 |
| Image Gen (SD 3.5) | Runs natively | SD 3.5 Large FP16 |
| Video Short (25f) | Runs natively | LTX Video 2B |
| Video Long (100f) | Tight fit | Wan Video 14B |
Architecture
Hopper is NVIDIA's datacenter-focused architecture succeeding Ampere. Built on TSMC 4N, it introduces the Transformer Engine with automatic FP8/FP16 mixed-precision training, HBM3/HBM3e memory, and NVLink 4.0 for multi-GPU scaling. The H100 flagship delivers up to 3x the AI training performance of A100.
AI Relevance
The Transformer Engine automatically manages FP8 precision for optimal training speed without accuracy loss. With up to 141 GB HBM3e (H200), Hopper GPUs can hold the largest open-weight models entirely in GPU memory, making them the workhorse of AI datacenters.
Buying advice
Excellent choice for local AI
Runs 36 of 50 top models well — a strong all-rounder for local inference.
96.0 GB
VRAM
$12,000
MSRP
$125/GB
Cost per GB VRAM
Best models for this GPU
What will limit you first
This setup is broadly balanced for this model.
No major red flags
This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.
Best upgrade itinerary
Unlocks 2 additional models that do not fit on the current setup.
Want more headroom? NVIDIA H200 141GB (141.0 GB VRAM) is the next step up.
Chat
SThis model is a direct match for chat. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.
Coding
SThis model is a direct match for coding. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.
Agentic Coding
SThis model is still usable for agentic-coding, but it is not the most specialized pick. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.
Reasoning
SThis model is a direct match for reasoning. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.
RAG
SThis model is a direct match for rag. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.
Qwen 3.5 122B A10BQwen3-Coder-NextQwen 2.5 VL 72B
Devstral 2 123B InstructMistral Small 4 119B
Command A 111B
GPT-OSS 120BQwen 3.6 35B A3BQwen3-Coder 30B A3B InstructPixtral Large 124BQwen 3.5 27BQwen3-VL 30B A3B InstructQwen 3.5 35B A3BQwen 3.6 27BQwen 3 32BLeanstral 119B A6BMagistral Small 2507Devstral Small 2 24B InstructQwen 3 30B A3BNemotron 3 Nano 30B
Gemma 4 31BDevstral Small 1.1Qwen 3.5 9BQwen 3 14BNemotron Cascade 2 30B A3B
Phi-4-reasoning-plus 14BGPT-OSS 20BQwen 3 8B
EXAONE 4.0 32BQwen 3.5 4BGemma 4 26B A4BMinistral 3 14BNemotron Nano 8BPhi-4 Mini Reasoning 4B
Jina Embeddings v3
BGE M3Qwen 3.5 397B A17B
Kimi K2.5Kimi K2.6
DeepSeek V4 ProDeepSeek V4 Flash
GLM-5.1GLM-5DeepSeek V3.2Qwen 3 235B A22BQwen3-Coder 480B A35B InstructDeepSeek Coder V2 236BDeepSeek R1 671BDeepSeek V3.1 671BJust out of reach
High-quality models that need a bit more memory
Image & Video Generation
51 of 52 models can generate images or video on your NVIDIA H20 96GB
| Model | Max Resolution | Gen Time | Grade |
|---|---|---|---|
| SD TurboImage | 512×512 | 300ms | S |
| Stable Diffusion 1.5Image | 512×768 | 500ms | S |
| Realistic Vision v5.1Image | 512×768 | 500ms | S |
| DreamShaper 8Image | 512×768 | 500ms | S |
| LCM DreamShaper v7Image | 512×768 | 200ms | S |
| PixArt-SigmaImage | 1024×1024 | ~2.1s | S |
| FramePack I2VVideo | 1280×720 | ~3.8s/frame | S |
| SDXL TurboImage | 512×512 | 300ms | S |
| SDXL LightningImage | 1024×1024 | 800ms | S |
| Stable Diffusion XL 1.0Image | 1024×1024 | ~2.1s | S |
| Playground v2.5Image | 1024×1024 | ~3.1s | S |
| RealVisXL v5.0Image | 1024×1024 | ~2.3s | S |
| DreamShaper XLImage | 1024×1024 | ~2.3s | S |
| Juggernaut XL v9Image | 1024×1024 | ~2.3s | S |
| Animagine XL 3.1Image | 1024×1024 | ~2.3s | S |
| Pony Diffusion V6 XLImage | 1024×1024 | ~2.3s | S |
| Animagine XL 4.0Image | 1024×1024 | ~2.3s | S |
| Illustrious XLImage | 1024×1024 | ~2.3s | S |
| Wan Video 2.1 1.3BVideo | 480×832 | ~1.5s/frame | S |
| Stable Diffusion 3.5 MediumImage | 1024×1024 | ~3.6s | S |
| Flux.2 Klein 4BImage | 1024×1024 | 600ms | S |
| LTX Video 2BVideo | 1280×720 | ~1.8s/frame | S |
| KolorsImage | 1024×1024 | ~4.2s | S |
| Stable CascadeImage | 1024×1024 | ~5.2s | S |
| AuraFlow v0.3Image | 1536×1536 | ~9.4s | S |
| Stable Diffusion 3.5 LargeImage | 1024×1024 | ~11.4s | S |
| Stable Diffusion 3.5 Large TurboImage | 1024×1024 | ~2.1s | S |
| CogVideoX 2BVideo | 720×480 | ~1.8s/frame | S |
| HunyuanVideoVideo | 720×1280 | ~3.8s/frame | S |
| ChromaImage | 1024×1024 | ~2.1s | S |
| Z-Image TurboImage | 1536×1536 | ~2.1s | S |
| Flux.1 DevImage | 1024×1024 | ~9.4s | S |
| Flux.1 SchnellImage | 1024×1024 | ~1.8s | S |
| LTX Video 13BVideo | 1280×720 | ~3.8s/frame | S |
| Flux.1 Kontext DevImage | 1024×1024 | ~10.4s | S |
| AnimateDiff v1.5.3Video | 512×768 | 900ms/frame | S |
| Cosmos Diffusion 7BVideo | 1024×576 | ~3s/frame | S |
| CogVideoX 5BVideo | 720×480 | ~2.6s/frame | S |
| Wan2.2 TI2V 5BVideo | 832×480 | ~2.6s/frame | S |
| Flux.2 Klein 9BImage | 1024×1024 | ~1s | S |
| Flux.1 Fill DevImage | 1024×1024 | ~8.8s | S |
| Mochi 1 PreviewVideo | 848×480 | ~3.4s/frame | S |
| HunyuanVideo 1.5Video | 720×1280 | ~3.2s/frame | S |
| Helios 14BVideo | 1280×720 | ~3.9s/frame | S |
| SkyReels V2 14BVideo | 1280×720 | ~3.9s/frame | S |
| Wan Video 2.1 14BVideo | 720×1280 | ~3.9s/frame | S |
| Wan Video 2.2 14BVideo | 720×1280 | ~3.9s/frame | S |
| Qwen ImageImage | 1024×1024 | ~3.5s | S |
| Qwen Image EditImage | 1024×1024 | ~3.5s | S |
| Flux.2 DevImage | 1024×1024 | ~1m 38s | S |
| MAGI-1Video | 1280×720 | ~4.9s/frame | S |
| HunyuanImage 3.0Image | 256×256 | ~6.2s | F |
Image models estimated at 1024×1024 (28 steps, FP16). Video models estimated at 768×512 (25 frames, 30 steps, FP16). Actual performance varies with runtime and system load.
Multi-GPU scaling
Scale out with multiple GPUs for larger models. PCIe interconnect with 22% scaling overhead.
| Config | Effective memory | Models that fit | Est. bandwidth |
|---|---|---|---|
| 1× NVIDIA | 96 GB | 351/374 | 4,000 GB/s |
| 2× NVIDIA | 192 GB | 359/374 | 6,240 GB/s |
| 4× NVIDIA | 384 GB | 366/374 | 12,480 GB/s |
| 8× NVIDIA | 768 GB | 373/374 | 24,960 GB/s |
Model counts use default quantization at coding workload settings. Multi-GPU scaling factor: 0.78× per additional GPU.
Upgrade paths
See what you unlock with more powerful hardware
Upgrade options
8× NVIDIA H20 96GBMulti-GPUUnlocks 22 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 111%.
NVLink gives this scale-out path a cleaner inter-GPU story than PCIe-only builds.
~$12,000 MSRP
NVIDIA H200 141GBNext step upUnlocks 2 additional models that do not fit on the current setup.
~$30,000 MSRP
NVIDIA B200 180GBNVIDIA upgradeUnlocks 8 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 34%.
~$30,000 MSRP
Unlocks 12 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 12%.
~$20,000 MSRP
Unlocks 13 additional models that do not fit on the current setup.
Lifts average decode speed across fitting models by about 25%.
~$8,000 MSRP
NVIDIA H20 96GB (96 GB VRAM) can run these top models: Qwen 3.5 122B A10B (score: 96/100), Qwen3-Coder-Next (score: 95/100), Qwen 2.5 VL 72B (score: 95/100). See the full compatibility list above.
NVIDIA H20 96GB has 96 GB of VRAM available for AI model inference. This determines which models and quantization levels you can run locally.
Yes, NVIDIA H20 96GB is excellent for running LLMs locally with top compatibility scores above 80/100.
For coding on NVIDIA H20 96GB, we recommend Qwen3-Coder-Next. It achieves 218.8 tokens per second with 256K context window. This model is a direct match for coding. It belongs to a current frontier family for local AI. It fits natively with comfortable headroom. Known channels: huggingface, ollama, lm-studio.
There are 5 upgrade path(s) from NVIDIA H20 96GB: NVIDIA H20 96GB, NVIDIA H200 141GB. Upgrading would unlock larger models and faster inference speeds.
Yes, NVIDIA H20 96GB with 96 GB of usable memory can run Flux.1 Dev at FP16 natively. Flux is a 12B parameter diffusion transformer that produces high-quality images. You can also run the Schnell variant for faster generation.
NVIDIA H20 96GB (96 GB VRAM) can handle various AI generation tasks beyond LLMs. For image generation, SDXL and Stable Diffusion 3.5 run well. Flux.1 Dev also runs natively for state-of-the-art image quality. For video, LTX Video 2.3 can generate short clips. Check the AI Capability Matrix above for detailed compatibility.
NVIDIA H20 96GB is excellent for AI image generation. With 96 GB of usable memory, it runs all major diffusion models including Flux.1, SDXL, and Stable Diffusion 3.5 at full precision. You can generate high-resolution images quickly and even handle video generation models.
Yes, NVIDIA H20 96GB with 96 GB of usable memory can run Qwen 3.5 27B at Q8 (near-lossless, ~28.9 GB) or even FP16 (~55.4 GB) depending on your context needs. This setup provides an excellent experience with this model. Use Ollama or vLLM for best results.
With 96 GB VRAM on NVIDIA H20 96GB, use Q8_0 for most models — it is near-lossless and you have the memory for it. For 70B+ models, Q6_K offers excellent quality. Reserve Q4_K_M for 100B+ models or when you need maximum context length.
NVIDIA H20 96GB already has strong memory bandwidth, so the next limit is often memory capacity and context headroom rather than raw decode speed. For local LLMs, fit first and bandwidth second is the right mental model.
NVIDIA H20 96GB supports up to 8× GPU scaling via NVLink. With 8× GPUs, you get 768 GB effective memory with a 0.78× scaling factor per GPU. This enables running models like Qwen 3.5 397B A17B and Kimi K2.5 that don't fit on a single card.
NVIDIA H20 96GB uses PCIe for multi-GPU communication, which has approximately 22% scaling overhead. For best multi-GPU performance, consider NVLink-equipped variants.
Compare with similar
