V LLMVRAMCalc
EN中文ESDE
Calculator › MiniMax M3.1 Flash (428B-A23B) › B300 288GB Blackwell Ultra

MiniMax M3.1 Flash (428B-A23B) on B300 288GB Blackwell Ultra (288GB) — VRAM, KV Cache & Throughput

MiniMax M3.1 Flash (428B-A23B) (428B total, 23B active, Sparse MoE + GQA) uses 242.14 GiB of 4-bit weights (Q4_K_M), requiring 243.65 GiB total at 8K context and 246.46 GiB at 32K on B300 288GB Blackwell Ultra (~177.8 tok/s decode).

Interactive Sizing Workbench

Full Compatibility Report →
288 GB · 8000 GB/s
Fits with Safe Headroom
Config-derived formula HF config.json ↗
Memory Allocation Breakdown (GiB) 243.65 GiB (261.62 GB) / 288 GiB (85%)
Model Weights: 242.14 GiB KV Cache & State: 0.94 GiB Runtime & Activation Buffers: 0.57 GiB System RAM Offload: 0 GiB
Required GPU Memory
243.65 GiB
261.62 GB (decimal)
Usable Hardware Memory
273.6 GiB
288 GB (physical)
Safe Headroom
+29.95 GiB
85% utilization
Est. Decode Throughput
~177.8 tok/s
23B active params
Est. Prefill Throughput
~24000 tok/s
4K Prompt TTFT: ~171 ms
Est. Cost / 1M Tokens
$10.84
Sparse MoE + GQA · Custom

Recommended Hardware for Current Configuration

Best Single GPU
Instinct MI355X 288GB (288GB)
~$27,000
Best Value / Used GPU
2x H200 141GB SXM
282GB · ~$64,000
Apple Unified Memory
Mac Studio M3 Ultra 512GB
~$9,500
Cloud From
RunPod · $6.94 / hr
1x B300 288GB Blackwell Ultra

Production Launch Commands

llama.cpp (GGUF) 
llama-server -hf unsloth/MiniMax-M3.1-Flash-GGUF:Q4_K_M \
  --ctx-size 8192 -ngl 999 --flash-attn
vLLM 
vllm serve MiniMaxAI/MiniMax-M3.1-Flash \
  --max-model-len 8192 --gpu-memory-utilization 0.92
Live Token Speed Simulator (TTFT ~171ms · ~177.8 tok/s)
[MiniMax M3.1 Flash (428B-A23B) · 243.65 GiB] allocated on target GPU. Prefill completed in TTFT window; streaming autoregressive tokens at memory-bandwidth-bound rate...

Cloud GPU Comparison & Rent-vs-Buy TCO

Verified public rates across RunPod, Vast.ai, Vultr, and DigitalOcean.

Verified: 2026-11-04
Purchase break-even
~33.2 months
Est. local power: $29.79 / mo
Provider Recommended Instance Total VRAM Hourly Rate Monthly at Selected Hours Billing Mode Deploy
RunPod 1x B300 288GB Blackwell Ultra (Community Cloud) 288 GB $6.94 / hr $843.90 / mo Per-second Deploy GPU ↗
Vast.ai 2x H200 141GB SXM (Verified Host · NVL) 282 GB $7.74 / hr $940.94 / mo Spot / On-demand Deploy GPU ↗
Vultr ($300 Free Credit*) 8x A100 80GB SXM (Dedicated Instance · $300 Credit*) 640 GB $19.18 / hr $2331.80 / mo Hourly Deploy GPU ↗
Vultr Cloud GPU 8x A100 80GB SXM (Dedicated Instance) 640 GB $19.18 / hr $2331.80 / mo Hourly Deploy GPU ↗
DigitalOcean 8x Instinct MI300X 192GB (Dedicated Instance) 1536 GB $20.72 / hr $2519.55 / mo Per-second Deploy GPU ↗

Rates reflect public listings on the verification date and fluctuate with regional inventory.

Fine-Tuning VRAM Requirements (QLoRA / LoRA / Full) — MiniMax M3.1 Flash (428B-A23B)

QLoRA (4b) Fits with Safe Headroom
246.5 GiB
Model Weights: 242.14 GiB · Gradients + Optimizer: 3.22 GiB · Act: 1.14 GiB
Recommended GPU: Instinct MI355X 288GB (288GB)
LoRA (16b) Exceeds Available GPU Memory
803.79 GiB
Model Weights: 797.21 GiB · Gradients + Optimizer: 5.06 GiB · Act: 1.52 GiB
Recommended GPU: 8x H200 141GB SXM (1128GB)
Full BF16 Exceeds Available GPU Memory
3271.25 GiB
Model Weights: 797.21 GiB · Gradients + Optimizer: 2471.35 GiB · Act: 2.69 GiB
Recommended GPU: Multi-GPU Required

Engineering Sizing Notes & Optimization Playbook

Quick Hardware Verdict

B300 288GB Blackwell Ultra (288GB physical, 273.6 GiB usable) runs 4-bit (Q4_K_M) MiniMax M3.1 Flash (428B-A23B) at 8K context using 243.65 GiB with 29.95 GiB of safe headroom (~177.8 tok/s single-stream decode).

KV Cache & Runtime Tuning

MiniMax M3.1 Flash (428B-A23B) uses Sparse MoE + GQA (428B total, 23B active per token). Moving from 8K to 32K context expands FP16 KV cache from 0.94 GiB to 3.75 GiB; enabling 8-bit KV cache (llama.cpp -ctk q8_0 -ctv q8_0 or vLLM --kv-cache-dtype fp8) saves 1.87 GiB at 32K (reducing total VRAM to 244.59 GiB).

VRAM Scaling Across Context Windows (4K to 128K, 4-bit Weights)

MiniMax M3.1 Flash (428B-A23B) VRAM scaling from 4K to 128K context (4K: 243.17 GiB, 8K: 243.65 GiB, 32K: 246.46 GiB, 128K: 257.71 GiB) 4K 243.17 GiB (KV 0.47 GiB) 8K 243.65 GiB (KV 0.94 GiB) 32K 246.46 GiB (KV 3.75 GiB) 128K 257.71 GiB (KV 15 GiB) B300 288GB Blackwell Ultra (288GB)
Model weights remain constant while attention KV cache grows with context length and batch size.

Quantization Precision & Memory Matrix — MiniMax M3.1 Flash (428B-A23B) @ 8K

Quantization Format Weights (GiB) KV Cache Total VRAM (GiB) Decode Speed Compatibility (B300 288GB Blackwell Ultra) Recommended GPU Data Source
Native Weights (BF16 / FP8 / MXFP4 — Lossless) (BF16) 797.21 GiB 0.94 GiB 798.72 GiB (857.62 GB) ~1.8 tok/s Exceeds Available GPU Memory Multi-GPU Required Official checkpoint size
FP8 / GGUF Q8_0 (8-bit — Near-Lossless) (Q8_0) 423.52 GiB 0.94 GiB 425.03 GiB (456.37 GB) ~6.6 tok/s Exceeds Available GPU Memory Multi-GPU Required Config-derived formula
GGUF Q6_K (6-bit — High Precision) (Q6_K) 326.86 GiB 0.94 GiB 328.37 GiB (352.58 GB) ~20.1 tok/s Exceeds Available GPU Memory Multi-GPU Required Config-derived formula
GGUF Q5_K_M (5-bit — Balanced) (Q5_K_M) 283.76 GiB 0.94 GiB 285.27 GiB (306.31 GB) ~161.2 tok/s Tight Fit (<8% Headroom) Multi-GPU Required Config-derived formula
Q4_K_M / UD-Q4_K_XL / AWQ (4-bit — Recommended) (Q4_K_M) 242.14 GiB 0.94 GiB 243.65 GiB (261.62 GB) ~177.8 tok/s Fits with Safe Headroom Instinct MI355X 288GB (288GB) Config-derived formula
GGUF IQ4_XS (4.25-bit — Compact) (IQ4_XS) 212.12 GiB 0.94 GiB 213.63 GiB (229.38 GB) ~192 tok/s Fits with Safe Headroom Instinct MI325X 256GB (256GB) Config-derived formula
GGUF Q3_K_M (3-bit — Low VRAM) (Q3_K_M) 195.13 GiB 0.94 GiB 196.64 GiB (211.14 GB) ~201.2 tok/s Fits with Safe Headroom Instinct MI325X 256GB (256GB) Config-derived formula
GGUF UD-Q2_K_XL (2-bit — Extreme MoE) (UD-Q2_K_XL) 137.52 GiB 0.94 GiB 139.03 GiB (149.28 GB) ~239.8 tok/s Fits with Safe Headroom Instinct MI300X 192GB (192GB) Config-derived formula

Technical Specifications & FAQ

Does MiniMax M3.1 Flash (428B-A23B) fit on B300 288GB Blackwell Ultra (288GB)?

In 4-bit (Q4_K_M) at 8,192 context, MiniMax M3.1 Flash (428B-A23B) requires 243.65 GiB total GPU memory (242.14 GiB weights + 0.94 GiB KV cache + 0.57 GiB runtime buffers) against 273.6 GiB usable on B300 288GB Blackwell Ultra, delivering ~177.8 tok/s decode.

How much VRAM does MiniMax M3.1 Flash (428B-A23B) need at 32K context?

At 32,768 context tokens, MiniMax M3.1 Flash (428B-A23B) (Sparse MoE + GQA) allocates 3.75 GiB for KV cache and state, bringing total 4-bit VRAM to 246.46 GiB (264.63 GB decimal).

Evaluate This Model on Other GPUs

RTX 3060 12GB (12GB)RTX 4060 Ti 16GB (16GB)RTX 4070 Ti Super (16GB)RTX 4080 Super (16GB)RTX 5070 12GB (12GB)RTX 5070 Ti (16GB)RTX 5080 (16GB)RTX 3090 (24GB)RTX 4090 (24GB)RTX 4090 D (24GB)RTX 5090 (32GB)RTX 5090 D (32GB)RTX A6000 (48GB)RTX 6000 Ada (48GB)RTX PRO 6000 Blackwell (96GB)DGX Spark (GB10 128GB) (128GB)L4 24GB (24GB)A40 48GB (48GB)L20 48GB (48GB)L40S 48GB (48GB)A800 80GB SXM (80GB)A100 80GB SXM (80GB)H800 80GB SXM (80GB)H100 80GB SXM (80GB)H20 96GB HBM3 (96GB)H20 141GB HBM3e (141GB)H200 141GB SXM (141GB)B200 180GB (180GB)Ascend 910B 64GB (昇腾) (64GB)Ascend 910C 128GB (昇腾) (128GB)Atlas 300I Duo 96GB (310P) (96GB)MLU590 80GB (寒武纪思元) (80GB)DCU K100-AI 64GB (海光深算) (64GB)MetaX C500 64GB (沐曦曦云) (64GB)Kunlunxin P800 64GB (昆仑芯) (64GB)MTT S4000 48GB (摩尔线程) (48GB)Radeon RX 7900 XTX 24GB (24GB)Radeon AI PRO R9700 32GB (32GB)Ryzen AI Max+ 395 128GB (128GB)Instinct MI300X 192GB (192GB)Instinct MI325X 256GB (256GB)Instinct MI355X 288GB (288GB)Arc Pro B60 24GB (24GB)Gaudi 3 128GB HBM2e (128GB)Mac mini M4 Pro 64GB (64GB)MacBook Pro M4 Max 128GB (128GB)Mac Studio M4 Ultra 256GB (256GB)Mac Studio M3 Ultra 512GB (512GB)

Evaluate Other Models on This Hardware

Qwen3.8 Flash Next (125B-A6B)Qwen3.8 27BQwen3.8 2.4T-A95BQwen3.6 35B-A3BQwen3.5 397B-A17BQwen3.5 122B-A10BQwen3.5 35B-A3BQwen3.5 27BQwen3.5 9BQwen3.5 4BQwen3 Coder 480B-A35BQwen3 Coder Next 80B-A3BQwen3 Coder 30B-A3BQwen3 30B-A3B Instruct 2507Qwen3 235B-A22B Instruct 2507Qwen3 32BQwen3 14BQwen3 8BQwen3 4B Instruct 2507Qwen3 1.7BQwen3 0.6BQwen2.5 72B InstructQwen2.5 Coder 32B InstructQwQ 32B ReasoningDeepSeek V4.1 Flash (552B-A16B)DeepSeek V4 FlashDeepSeek V3.2 (685B)DeepSeek R1-0528 671BDeepSeek V3-0324 671BDeepSeek R1 Distill Llama 70BDeepSeek R1 Distill Qwen 32BDeepSeek R1 Distill Qwen 14BDeepSeek R1 Distill Llama 8Bgpt-oss-20bgpt-oss-120bGemma 4 31B ITGemma 4 26B-A4B ITGemma 4 12B ITGemma 3 27B ITGemma 3 12B ITGemma 3 4B ITGLM-5.3 Flash (320B-A18B)GLM-5.3 (753B)GLM-5 (744B)GLM-4.5 (355B-A32B)GLM-4.5 Air (110B-A12B)GLM-4 32B-0414Kimi K3 (2.8T-A104B)Kimi K2 Instruct (1T-A32B)Kimi-Dev 72BMiniMax M2.5 (229B-A10B)Step-3.7 Flash (198B-A11B)Step-3.5 Flash (197B-A11B)ERNIE 4.5 300B-A47B (文心)ERNIE 4.5 21B-A3B (文心)Hunyuan A13B Instruct (混元 80B)MiniCPM5 2B (面壁)Llama 4 Maverick 17B-128E (400B)Llama 4 Scout 17B-16E (109B)Llama 3.1 405B InstructLlama 3.3 70B InstructLlama 3.1 8B InstructLlama 3.2 3B InstructMistral Large 4 (1T-A52B)Mistral Large 2 123BMixtral 8x22B InstructMixtral 8x7B InstructMistral Small 3.2 24BDevstral Small 24BNemotron 3 Ultra 550B-A55BNemotron 3 Super 120B-A12BNemotron 3 Nano 30B-A3BCohere Command A 111BPhi-4 14BLiquid LFM3 1.7B