NVIDIA Rubin
288 GB decides what fits. 22000 GB/s decides how fast it runs once it does.
- Largest popular model that fitsMiniMax-M3427B params · Q4_K_M · needs 264.0 GB~927 tok/s Faster than you readOpen in the calculator →
- Best fast pickMiniMax-Text-01456B params · Q4_K_M · needs 283.5 GB~589 tok/s Faster than you readOpen in the calculator →
- Most downloaded that fitsGLM-5.3-Flash321B params · Q4_K_M · needs 198.3 GB~1583 tok/s Faster than you readOpen in the calculator →
At 8,192 tokens of context with the whole model in device memory. Speeds are estimates from memory bandwidth, not benchmarks run on this card.
286 of 320 fit entirely
286 of the 320 models the engine can size fit entirely in device memory at Q4_K_M where it is published, otherwise the nearest published format, and 8,192 tokens. 1 run with some layers on system memory (32 GB assumed), and 33 do not run at all.
Featured models · Q4_K_M at 8,192 tokens, including offload
| Model | Needs | Verdict | Decode | Calculator |
|---|---|---|---|---|
| Qwen3.5-2B2.3B parameters | 2.32 GB | fits | ~12127 tok/s | Open → |
| NVIDIA-Nemotron-3-Nano-4B-BF164.0B parameters | 3.46 GB | fits | ~5619 tok/s | Open → |
| Qwen3-VL-8B-Instruct8.8B parameters | 7.04 GB | fits | ~2542 tok/s | Open → |
| Qwen3.5-9B9.7B parameters | 7.05 GB | fits | ~3078 tok/s | Open → |
| gemma-4-12B-it12B parameters | 9.54 GB | fits | ~1830 tok/s | Open → |
| gpt-oss-20b21B parameters | 13.8 GB | fits | ~5981 tok/s | Open → |
| Qwen3.8-27B28B parameters | 18.6 GB | fits | ~1011 tok/s | Open → |
| Qwen3.6-27B28B parameters | 18.6 GB | fits | ~1011 tok/s | Open → |
| Kimi-Linear-48B-A3B-Instruct49B parameters | 30.4 GB | fits | ~8002 tok/s | Open → |
| Qwen3.8-Flash-Next180B parameters | 111.6 GB | fits | ~4158 tok/s | Open → |
| DeepSeek-V4-Flash-Vision-Exp305B parameters | 188.1 GB | fits | ~1329 tok/s | Open → |
| GLM-5.3-Flash321B parameters | 198.3 GB | fits | ~1583 tok/s | Open → |
These examples are selected from prominent labs using the catalogue’s latest Hugging Face 30-day downloads and repository-creation freshness signal, with newer releases guaranteed a place. Offloaded rows assume 32 GB of system RAM, and a speed is only shown for a row that runs. Fitting in memory is not the same as loading: whether the runtime and version you have supports each architecture and format on this machine has not been tested here. Each “Open” link carries the same model, format, context and RAM into the calculator.
Fits entirely in NVIDIA Rubin memory — 286 of 320 sized models
The most demanding model that fits is MiniMax-Text-01 at Q4_K_M: 283.5 GB of the 288 GB, leaving 4.53 GB spare.
Fully resident at 8,192 tokens (or the model’s own maximum, where that is shorter), offload off, at 22000 GB/s. Each row is a run of the engine for this configuration; the rows start with current, prominent releases and “fits” is memory, not a tested runtime. Older or less prominent models remain available through this search and “Show all”.
Showing 40 of 286 models that fit.
| Model | Format | Needs | Spare | Decode | Downloads / 30d | Calculator |
|---|---|---|---|---|---|---|
| GLM-5.3-FlashNEWzai-org · 321B params | Q4_K_M | 198.3 GB | 89.7 GB | ~1583 tok/s | 5.4M | Open → |
| Qwen3.8-27BQwen · 28B params | Q4_K_M | 18.6 GB | 269.4 GB | ~1011 tok/s | 6.9M | Open → |
| DeepSeek-V4-Flash-0731DeepSeek · 304B params | Q4_K_M | 187.5 GB | 100.5 GB | ~1329 tok/s | 4.5M | Open → |
| Qwen3.8-Flash-NextNEWQwen · 180B params | Q4_K_M | 111.6 GB | 176.4 GB | ~4158 tok/s | 1.4M | Open → |
| gemma-4-26B-A4B-itGoogle · 26B params | Q4_K_M | 16.8 GB | 271.2 GB | ~5619 tok/s | 13M | Open → |
| DeepSeek-V4-Flash-Vision-ExpNEWDeepSeek · 305B params | Q4_K_M | 188.1 GB | 99.9 GB | ~1329 tok/s | 915.3K | Open → |
| gemma-4-31B-itGoogle · 31B params | Q4_K_M | 22.2 GB | 265.8 GB | ~727 tok/s | 9.9M | Open → |
| Qwen3.5-9BQwen · 9.7B params | Q4_K_M | 7.05 GB | 281.0 GB | ~3078 tok/s | 9M | Open → |
| Qwen3.5-4BQwen · 4.7B params | Q4_K_M | 3.98 GB | 284.0 GB | ~5539 tok/s | 7.8M | Open → |
| Qwen3.6-35B-A3BQwen · 36B params | Q4_K_M | 23.1 GB | 264.9 GB | ~8069 tok/s | 3.3M | Open → |
| gemma-4-12B-itGoogle · 12B params | Q4_K_M | 9.54 GB | 278.5 GB | ~1830 tok/s | 1.9M | Open → |
| Inkling-Smallthinkingmachines · 266B params | Q4_K_M | 165.5 GB | 122.5 GB | ~2100 tok/s | 657.4K | Open → |
| Qwen3.6-27BQwen · 28B params | Q4_K_M | 18.6 GB | 269.4 GB | ~1011 tok/s | 2.5M | Open → |
| Qwen3.5-2BQwen · 2.3B params | Q4_K_M | 2.32 GB | 285.7 GB | ~12127 tok/s | 4.9M | Open → |
| gemma-4-E4B-itGoogle · 8.0B params | Q4_K_M | 5.86 GB | 282.1 GB | ~4945 tok/s | 4.4M | Open → |
| NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16NVIDIA · 32B params | Q4_K_M | 20.3 GB | 267.7 GB | ~767 tok/s | 530K | Open → |
| NVIDIA-Nemotron-3-Nano-4B-BF16NVIDIA · 4.0B params | Q4_K_M | 3.46 GB | 284.5 GB | ~5619 tok/s | 3.4M | Open → |
| gemma-4-E2B-itGoogle · 5.1B params | Q4_K_M | 4.02 GB | 284.0 GB | ~13063 tok/s | 3M | Open → |
| Qwen3.5-0.8BQwen · 873M params | Q4_K_M | 1.46 GB | 286.5 GB | ~26567 tok/s | 2.6M | Open → |
| DeepSeek-V4-FlashDeepSeek · 284B params | Q4_K_M | 175.9 GB | 112.1 GB | ~1329 tok/s | 1.1M | Open → |
| Qwen3-VL-8B-InstructQwen · 8.8B params | Q4_K_M | 7.04 GB | 281.0 GB | ~2542 tok/s | 14.6M | Open → |
| MiniMax-M2.7MiniMaxAI · 229B params | Q4_K_M | 141.2 GB | 146.8 GB | ~1741 tok/s | 1.1M | Open → |
| Qwen3.5-27BQwen · 28B params | Q4_K_M | 18.6 GB | 269.4 GB | ~1011 tok/s | 1.9M | Open → |
| North-Micro-Vision-InstructCohereLabs · 2.5B params | Q4_K_M | 2.99 GB | 285.0 GB | ~7975 tok/s | 180.7K | Open → |
| Hy3Tencent · 299B params | Q4_K_M | 181.2 GB | 106.8 GB | ~1060 tok/s | 304.9K | Open → |
| Qwen3.5-35B-A3BQwen · 36B params | Q4_K_M | 23.1 GB | 264.9 GB | ~8069 tok/s | 1.6M | Open → |
| NVIDIA-Nemotron-3-Super-120B-A12B-BF16NVIDIA · 124B params | Q4_K_M | 76.9 GB | 211.1 GB | ~196 tok/s | 1.2M | Open → |
| granite-4.2-8bIBM · 8.8B params | Q4_K_M | 7.49 GB | 280.5 GB | ~2315 tok/s | 128.4K | Open → |
| GLM-4.7-Flashzai-org · 31B params | Q4_K_M | 20.4 GB | 267.6 GB | ~762 tok/s | 1.8M | Open → |
| granite-4.1-3bIBM · 3.4B params | Q4_K_M | 3.72 GB | 284.3 GB | ~5401 tok/s | 521.8K | Open → |
| LFM2.5-2.6BLiquidAI · 2.7B params | Q4_K_M | 2.61 GB | 285.4 GB | ~8662 tok/s | 108.1K | Open → |
| MiMo-V2.6-Flash-RLNEWXiaomiMiMo · 309B params | Q4_K_M | 192.2 GB | 95.8 GB | ~1542 tok/s | 48.6K | Open → |
| MiniMax-M3MiniMaxAI · 427B params | Q4_K_M | 264.0 GB | 24.0 GB | ~927 tok/s | 177.2K | Open → |
| MiMo-V2.5XiaomiMiMo · 311B params | Q4_K_M | 192.2 GB | 95.8 GB | ~1542 tok/s | 245.2K | Open → |
| Qwen3-VL-4B-InstructQwen · 4.4B params | Q4_K_M | 4.72 GB | 283.3 GB | ~4040 tok/s | 3.5M | Open → |
| granite-4.1-30bIBM · 29B params | Q4_K_M | 20.4 GB | 267.6 GB | ~761 tok/s | 301.5K | Open → |
| Qwen3.5-122B-A10BQwen · 125B params | Q4_K_M | 77.9 GB | 210.1 GB | ~2954 tok/s | 512.6K | Open → |
| Qwen3-VL-2B-InstructQwen · 2.1B params | Q4_K_M | 3.05 GB | 285.0 GB | ~7322 tok/s | 2.8M | Open → |
| granite-4.2-3bIBM · 3.7B params | Q4_K_M | 3.72 GB | 284.3 GB | ~5401 tok/s | 50.6K | Open → |
| LFM2.5-230MLiquidAI · 230M params | Q4_K_M | 1.05 GB | 286.9 GB | ~60379 tok/s | 88.6K | Open → |
Sized at the format most people actually download, not at FP16. The catalogue ordered by downloads →
The specification behind every figure
What the manufacturer publishes for this device, and the pages it was read from.
Manufacturer specification
| Memory scope | dedicated |
|---|---|
| Capacity | 288 GB |
| Published options | 288 GB |
| Bandwidth | 22000 GB/s |
| Memory type | HBM4 |
| Bus width | Not published |
| FP32 peak | Not published |
| Dense matrix peak | Not published without sparsity |
| Power | Not published |
Source ledger
Caveats
- Announced, not verified as purchasable. NVIDIA published this architecture on 21 July 2026; this catalogue has not confirmed a shipping part, a price or a delivery date, and the entry exists so the device can be planned against rather than as evidence that it can be bought.
- NVIDIA states both figures as "up to". They are the ceiling of the part as described, and a specific product built on it may publish less.
- These are per-GPU figures. The Vera Rubin NVL72 in the same article is a rack of them with its own networking and cooling; its rack-level numbers are not this device's and are not carried here.
- No per-GPU power figure is published in this article, so none is recorded. The rack-level power discussion is about the NVL72, not one GPU.
- No dense matrix throughput is published for this device in a form this catalogue accepts, so no compute roof is priced and time to first token is withheld.
Published capacity is a hardware ceiling, not guaranteed free runtime memory. The calculator shows the runtime reserve separately rather than folding it into a single number.
Run the diagnostic on the NVIDIA Rubin →