Shanghai AI Lab, China · Released September 11, 2026New
Atria Dawn Preview has 753.3B parameters (42B active per token). In BF16 its weights alone take 1385 GiB; quantized to 4-bit, about 418 GiB. Each 32K-token request adds 2.83 GiB of KV cache. Cheapest way to run it today: 7 × NVIDIA A100 80GB at $6.54 per hour.
Size your own setup in the calculator →Weights plus the KV cache of one request (BF16 cache), in GiB. Add 1 to 3 GiB for the inference engine itself.
| Weights format | Weights only | + 8K context | + 32K context | + 128K context | + 1M context |
|---|---|---|---|---|---|
| BF16 / FP16 | 1385 | 1385 | 1387 | 1396 | 1475 |
| FP8 / INT8 | 694 | 695 | 697 | 705 | 785 |
| 4-bit (GGUF Q4_K_M) | 418 | 419 | 421 | 429 | 509 |
Fewest GPUs that fit the weights plus one 8K-token request. Data-center GPUs use vLLM (90% of memory usable), the others llama.cpp. Several GPUs split the model with tensor or pipeline parallelism.
| GPU | BF16 / FP16 | FP8 / INT8 | 4-bit (GGUF Q4_K_M) |
|---|---|---|---|
| GeForce RTX 4090 24 GB | 96 (several servers) | 48 (several servers) | 24 (several servers) |
| GeForce RTX 5090 32 GB | 64 (several servers) | 32 (several servers) | 16 (several servers) |
| RTX PRO 6000 Blackwell 96 GB | 16 (several servers) | 8 | 5 |
| Mac M5 Max 128 GB 96 GB usable | 16 (several servers) | 8 | 5 |
| NVIDIA H100 80 GB | 24 (several servers) | 16 (several servers) | 7 |
| NVIDIA H200 141 GB | 16 (several servers) | 6 | 4 |
| NVIDIA B200 (HGX) 180 GB | 16 (several servers) | 5 | 3 |
For each GPU: the fewest cards that fit Atria Dawn Preview with one 8K-token request, the most faithful weight format at that count, and the cheapest on-demand price (2026-10-10).
On-demand prices from RunPod, Vast.ai, Verda and Azure, checked every hour. Serving many users needs more KV cache, so more memory: size it in the calculator. Some links are affiliate links: StudioTV may earn a commission, at no extra cost to you.
Architecture: 78 layers · MLA (512 + 64 latent) on 78 · indexer keys on 21. Its cache stays compact because MLA stores one compressed latent per token and layer instead of full keys and values. Only 42B parameters are active per token, but all 753B must sit in memory: VRAM depends on the total, speed on the active part.
| Context per request | BF16 cache | FP8 cache |
|---|---|---|
| 8K tokens | 0.71 GiB | 0.41 GiB |
| 32K tokens | 2.83 GiB | 1.65 GiB |
| 128K tokens | 11.3 GiB | 6.58 GiB |
| 1M tokens | 90.5 GiB | 52.7 GiB |
In BF16 the weights alone take 1385 GiB (1487 GB). In FP8 that is 694 GiB, and about 418 GiB with 4-bit quantization (Q4_K_M). Each request then adds KV cache: 0.71 GiB at 8K tokens and 2.83 GiB at 32K (BF16 cache).
No. Even in 4-bit it needs 24 (several servers) RTX 4090s. In 4-bit it needs 7 H100 GPUs.
On 2026-10-10, the cheapest on-demand setup is 7 × NVIDIA A100 80GB with 4-bit (GGUF Q4_K_M) weights on Vast.ai, at $6.54 per hour (about $4,772.74 per month). Next: 5 × RTX PRO 6000 Blackwell (96 GB) with 4-bit (GGUF Q4_K_M) weights on Vast.ai, at $6.67 per hour. Sized for one 8K-token request; prices are checked every hour.
With one request and an 8K-token context: 24 (several servers) in BF16, 16 (several servers) in FP8 and 7 in 4-bit (vLLM, 90% of memory usable). Serving many users at once needs more memory for their KV caches: the calculator sizes that for you.
For a model card or a README: the 4-bit size and the smallest GPU it fits on, linked to this page. Data as JSON: /api/models/atria-dawn-preview.json.
Shanghai AI Lab, China · Released September 11, 2026[](https://studiotvai.com/vram-requirements/atria-dawn-preview)Same family (large moe): MiniMax M2.7 · DeepSeek R1 · Kimi K2.6 · GLM-5.3 · GLM-5.3-Flash · DeepSeek V4 Flash · DeepSeek V4 Pro · DeepSeek V4.1 Flash · MiMo-V2.6 Pro · MiMo-V2.6 Flash · Kimi K3 · Mistral Large 4 · DeepSeek V4 Flash Vision Exp · Qwen3.8 2.4T-A95B · Ornith 1.5 397B · Intern S2 397B
All models: Llama 3.1 8B · Mistral Small 3.2 24B · Qwen3 32B · Llama 3.3 70B · Llama 3.1 405B · Qwen3.5 4B · Qwen3.5 9B · Qwen3.8 27B · Qwen3.6 35B-A3B · Qwen3.5 122B-A10B · Qwen3.8-Flash-Next · LFM2.5 8B-A1B · LFM2 24B-A2B · Gemma 4 12B · Gemma 4 31B · Gemma 4 26B-A4B · gpt-oss 20B · gpt-oss 120B · Llama 4 Scout 17B-16E · Llama 4 Maverick 17B-128E · Kolibri-1 · Nemotron 3 Nano 30B-A3B · Nemotron 3 Super 120B-A12B · MiniMax M2.7 · DeepSeek R1 · Kimi K2.6 · GLM-5.3 · GLM-5.3-Flash · DeepSeek V4 Flash · DeepSeek V4 Pro · DeepSeek V4.1 Flash · MiMo-V2.6 Pro · MiMo-V2.6 Flash · Kimi K3 · Mistral Large 4 · Qwen3-Coder-Next 80B-A3B · Qwen3-Coder 30B-A3B · Qwen3-Coder 480B-A35B · DeepSeek V4 Flash Vision Exp · Qwen3.8 2.4T-A95B · Ornith 1.5 397B · Ornith 1.5 35B-A3B · Ornith 1.5 9B · Intern S2 397B · Clef · Clef Flash · D1 3B · Humanizer · JEV 27B VL · Mellum2.1 12B-A2.5B · LightOnOCR-3 4B · Spark-X2.5 4B · Agnes-3.0-Qwen
Estimates, not guarantees: computed from the official config.json with the same engine as the LLM VRAM Calculator. Real usage depends on your engine version and settings.