NVIDIA · Blackwell Ultra
Best AI Models for NVIDIA B300 (288 GB)
Runs everything up to 120B MoE well.
288 GB HBM3e8000 GB/s3750 TFLOPS FP161400 WBlackwell Ultra 2025
Datacenter Blackwell Ultra — rack/cloud only, not a desktop card. Listed for comparison.
| # | Model | Speed | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Q4_K_M · 186.5 GB · 229 t/s | 229 tok/s | Q4_K_M · 186.5 GB | |
| 2 | DeepSeek V4 Flash 0731DeepSeek Q4_K_M · 183.0 GB · 181 t/s | 181 tok/s | Q4_K_M · 183.0 GB | |
| 3 | GLM 5.3 FlashZ.ai Q4_K_M · 205.2 GB · 217 t/s | 217 tok/s | Q4_K_M · 205.2 GB | |
| 4 | Qwen3.8 27BAlibaba Q4_K_M · 17.9 GB · 299 t/s | 299 tok/s | Q4_K_M · 17.9 GB | |
| 5 | Hy3Tencent Q4_K_M · 182.3 GB · 147 t/s | 147 tok/s | Q4_K_M · 182.3 GB |
| # | Model | Speed | Coding Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Q4_K_M · 186.5 GB · 229 t/s | 229 tok/s | Q4_K_M · 186.5 GB | |
| 2 | DeepSeek V4 Flash 0731DeepSeek Q4_K_M · 183.0 GB · 181 t/s | 181 tok/s | Q4_K_M · 183.0 GB | |
| 3 | Qwen3.8 27BAlibaba Q4_K_M · 17.9 GB · 299 t/s | 299 tok/s | Q4_K_M · 17.9 GB | |
| 4 | GLM 5.3 FlashZ.ai Q4_K_M · 205.2 GB · 217 t/s | 217 tok/s | Q4_K_M · 205.2 GB | |
| 5 | Hy3Tencent Q4_K_M · 182.3 GB · 147 t/s | 147 tok/s | Q4_K_M · 182.3 GB |
| # | Model | Speed | Reasoning Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Q4_K_M · 183.0 GB · 181 t/s | 181 tok/s | Q4_K_M · 183.0 GB | |
| 2 | GLM 5.3 FlashZ.ai Q4_K_M · 205.2 GB · 217 t/s | 217 tok/s | Q4_K_M · 205.2 GB | |
| 3 | Qwen3.8 27BAlibaba Q4_K_M · 17.9 GB · 299 t/s | 299 tok/s | Q4_K_M · 17.9 GB | |
| 4 | Inkling SmallThinking Machines Q4_K_M · 162.7 GB · 399 t/s | 399 tok/s | Q4_K_M · 162.7 GB | |
| 5 | Hy3Tencent Q4_K_M · 182.3 GB · 147 t/s | 147 tok/s | Q4_K_M · 182.3 GB |
| # | Model | Speed ↓ | llmrun Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Qwen3.5 4BAlibabaHow to run → Q4_K_M · 3.4 GB · 1595 t/s | 1595 tok/s | Q4_K_M · 3.4 GB | |
| 2 | Qwen3.5 9BAlibaba Q4_K_M · 6.6 GB · 818 t/s | 818 tok/s | Q4_K_M · 6.6 GB | |
| 3 | GPT OSS 20BOpenAI Q4_K_M · 13.1 GB · 550 t/s | 550 tok/s | Q4_K_M · 13.1 GB | |
| 4 | Gemma 4 26B A4B ITGoogle Q4_K_M · 16.5 GB · 459 t/s | 459 tok/s | Q4_K_M · 16.5 GB | |
| 5 | Hy3Tencent Q4_K_M · 182.3 GB · 147 t/s | 147 tok/s | Q4_K_M · 182.3 GB |
| # | Model | Speed | Vision Score | Quant · VRAM |
|---|---|---|---|---|
| 1 | Q4_K_M · 205.2 GB · 217 t/s | 217 tok/s | Q4_K_M · 205.2 GB | |
| 2 | Qwen3.8 27BAlibaba Q4_K_M · 17.9 GB · 299 t/s | 299 tok/s | Q4_K_M · 17.9 GB | |
| 3 | Inkling SmallThinking Machines Q4_K_M · 162.7 GB · 399 t/s | 399 tok/s | Q4_K_M · 162.7 GB | |
| 4 | MiniMax M3MiniMax Q4_K_M · 258.0 GB · 179 t/s | 179 tok/s | Q4_K_M · 258.0 GB | |
| 5 | Qwen3.8 Flash NextAlibaba Q4_K_M · 109.0 GB · 102 t/s | 102 tok/s | Q4_K_M · 109.0 GB |
See all 433 models on the B300 →Best quant that fits · speeds estimated on this card
The B300 runs everything up to 120B MoE well. Blackwell Ultra generation, released 2025.
GenerationBlackwell Ultrareleased 2025
B300 strengths and weaknesses
Strengths
- Runs 120B MoE models: GPT OSS 120B at 378 tok/s in 70.8 GB
- Room for long context: Llama 3.3 70B Instruct at Q4 uses 51.3 GB, leaving 236.7 GB free
- Reads long prompts fast: A 4K-token prompt on MiMo V2.6 Flash RL starts answering in 82 ms
- FP8 and NVFP4 run natively: Blackwell takes FP8 and NVFP4 checkpoints in vLLM as published
Weaknesses
- 1400 W under load: Plan a 2,800 W PSU; ≈$307 a year at 4 h a day
Output speed depends on memory bandwidth (8000 GB/s).
- 9BQwen3.5 9B · Q4_K_M · 6.6 GB818tok/s
- 14BQwen3 14B · Q4_K_M · 11.9 GB547tok/s
- 27–31BQwen3.8 27B · Q4_K_M · 17.9 GB299tok/s
- 35B MoEQwen3.6 35B A3B · Q4_K_M · 22.0 GB363tok/s
- 70BLlama 3.3 70B Instruct · Q4_K_M · 51.3 GB112tok/s
- 120B MoEGPT OSS 120B · Q4_K_M · 70.8 GB378tok/s
Estimated speeds: Q4_K_M, 4K context, one user. How we estimate
All 1341 models on the B300
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 |
|---|---|---|---|---|---|---|
| Q4_K_M | 19.1 GB | 334.2 t/s | 203K | S100, Runs great | ||
| Q4_K_M | 1.0 GB | 5416.7 t/s | 33K | S100, Runs great | ||
InternVL2 1BVision | Q4_K_M | 2.0 GB | 8387.1 t/s | — | S100, Runs great | |
Ilama 3.2 1BChat | BF16 | 3.3 GB | 1831.0 t/s | 131K | S100, Runs great | |
PaddleOCR VL 1.6Vision | Q4_K_M | 1.0 GB | 5842.7 t/s | 131K | S100, Runs great | |
MN VelvetCafe RP 12BChatRoleplay | Q4_K_M | 11.0 GB | 644.4 t/s | 131K | S100, Runs great | |
| Q4_K_M | 4.2 GB | 1793.1 t/s | 262K | S100, Runs great | ||
| Q4_K_M | 4.2 GB | 1793.1 t/s | 262K | S100, Runs great | ||
Granite 4.1 8BChatFunctions | Q4_K_M | 8.3 GB | 879.9 t/s | 131K | S100, Runs great | |
GPT OSS BF16ChatFunctions | Q4_K_M | 0.3 GB | 17333.3 t/s | 131K | S100, Runs great | |
LightOnOCR 3 4BVisionFunctions | Q4_K_M | 3.4 GB | 1630.1 t/s | 262K | S100, Runs great | |
GhostFace 24B V1Chat | Q4_K_M | 17.8 GB | 349.9 t/s | 131K | S100, Runs great | |
LightOnOCR 3 0.8BVisionFunctions | Q4_K_M | 0.9 GB | 6046.5 t/s | 262K | S100, Runs great | |
GPT OSS 20B HereticChatFunctions | Q4_K_M | 13.1 GB | 550.0 t/s | 131K | S100, Runs great | |
Qwen3 4B Instruct 2507ChatFunctions | Q4_K_M | 4.2 GB | 1793.1 t/s | 262K | S100, Runs great |
Specs
- Architecture
- Blackwell Ultra
- VRAM
- 288 GB HBM3e
- Memory bandwidth
- 8000.0 GB/s
- FP16 (dense)
- 3750.00 TFLOPS
- CUDA cores
- 20,480
- Tensor cores
- 640
- Compute capability
- 10.0
- TDP
- 1400 W
- Released
- 2025-08-01
B300 FAQ
What is the largest model the B300 can run?
Beam 1.0 501B A23B (501.3B) is the largest at 4-bit or better: at Q4_K_S it uses 285.2 GB of 288 GB, and runs at about 185 tok/s.
Can the B300 run MiMo V2.6 Flash RL?
Yes. MiMo V2.6 Flash RL fits at Q4_K_M, using 186.5 GB of 288 GB, and it is the best model that runs well on this card.
How many tokens per second does the B300 give?
About 229 tok/s on MiMo V2.6 Flash RL at Q4_K_M, with the first token after 0.1 s on a 4K-token prompt.
What power supply does the B300 need?
Plan a 2800 W PSU (2× the 1400 W TDP).
Is 288 GB enough for coding and agents?
Yes for coding: MiMo V2.6 Flash RL, the best of the 25 coding models that fit, runs at about 229 tok/s.