V LLMVRAMCalc
EN中文ESDE
Calculator › DeepSeek V4 Flash › Ryzen AI Max+ 395 128GB

DeepSeek V4 Flash on Ryzen AI Max+ 395 128GB (128GB) — VRAM, KV Cache & Throughput

DeepSeek V4 Flash (284B total, 11B active, Sparse MoE + GQA) uses 144.44 GiB of 4-bit weights (UD-Q4_K_XL), requiring 145.06 GiB total at 8K context and 145.22 GiB at 32K on Ryzen AI Max+ 395 128GB (~19 tok/s decode).

Interactive Sizing Workbench

Full Compatibility Report →
128 GB · 256 GB/s
Exceeds Available GPU Memory
Measured GGUF shard size HF config.json ↗
Memory Allocation Breakdown (GiB) 145.06 GiB (155.76 GB) / 128 GiB (113%)
Model Weights: 144.44 GiB KV Cache & State: 0.06 GiB Runtime & Activation Buffers: 0.56 GiB System RAM Offload: 0 GiB
Required GPU Memory
145.06 GiB
155.76 GB (decimal)
Usable Hardware Memory
99.8 GiB
128 GB (physical)
Safe Headroom
-45.26 GiB
113% utilization
Est. Decode Throughput
~19 tok/s
11B active params
Est. Prefill Throughput
~1126 tok/s
4K Prompt TTFT: ~3636 ms
Est. Cost / 1M Tokens
$37.87
Sparse MoE + GQA · MIT

Recommended Hardware for Current Configuration

Best Single GPU
Instinct MI300X 192GB (192GB)
~$15,000
Best Value / Used GPU
2x A100 80GB SXM
160GB · ~$24,000
Apple Unified Memory
Mac Studio M4 Ultra 256GB
~$5,999
Cloud From
DigitalOcean · $2.59 / hr
1x Instinct MI300X 192GB

Production Launch Commands

llama.cpp (GGUF) 
llama-server -hf unsloth/DeepSeek-V4-Flash-GGUF:UD-Q4_K_XL \
  --ctx-size 8192 -ngl 999 --flash-attn
vLLM 
vllm serve deepseek-ai/DeepSeek-V4-Flash-0731 \
  --max-model-len 8192 --gpu-memory-utilization 0.92
Live Token Speed Simulator (TTFT ~3636ms · ~19 tok/s)
[DeepSeek V4 Flash · 145.06 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
~21.1 months
Est. local power: $15.96 / mo
Provider Recommended Instance Total VRAM Hourly Rate Monthly at Selected Hours Billing Mode Deploy
RunPod 1x B200 180GB (Community Cloud) 180 GB $5.98 / hr $727.17 / mo Per-second Deploy GPU ↗
Vast.ai 1x B200 180GB (Verified Host) 180 GB $7.50 / hr $912.24 / mo Spot / On-demand Deploy GPU ↗
Vultr ($300 Free Credit*) 4x L40S 48GB (Dedicated Instance · $300 Credit*) 192 GB $6.68 / hr $812.77 / mo Hourly Deploy GPU ↗
Vultr Cloud GPU 4x L40S 48GB (Dedicated Instance) 192 GB $6.68 / hr $812.77 / mo Hourly Deploy GPU ↗
DigitalOcean 1x Instinct MI300X 192GB (Dedicated Instance) 192 GB $2.59 / hr $314.94 / 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) — DeepSeek V4 Flash

QLoRA (4b) Exceeds Available GPU Memory
147.06 GiB
Model Weights: 144.44 GiB · Gradients + Optimizer: 1.54 GiB · Act: 1.08 GiB
Recommended GPU: Instinct MI300X 192GB (192GB)
LoRA (16b) Exceeds Available GPU Memory
150.65 GiB
Model Weights: 146.8 GiB · Gradients + Optimizer: 2.42 GiB · Act: 1.43 GiB
Recommended GPU: Instinct MI300X 192GB (192GB)
Full BF16 Exceeds Available GPU Memory
604.4 GiB
Model Weights: 146.8 GiB · Gradients + Optimizer: 455.08 GiB · Act: 2.52 GiB
Recommended GPU: 4x B200 180GB (720GB)

Engineering Sizing Notes & Optimization Playbook

Quick Hardware Verdict

A single Ryzen AI Max+ 395 128GB (128GB) cannot hold 4-bit DeepSeek V4 Flash (145.06 GiB at 8K) entirely in VRAM. Use multi-GPU tensor parallelism, offload layers to system RAM, or deploy on a larger cloud GPU.

KV Cache & Runtime Tuning

DeepSeek V4 Flash uses Sparse MoE + GQA (284B total, 11B active per token). Moving from 8K to 32K context expands FP16 KV cache from 0.06 GiB to 0.22 GiB; enabling 8-bit KV cache (llama.cpp -ctk q8_0 -ctv q8_0 or vLLM --kv-cache-dtype fp8) saves 0.11 GiB at 32K (reducing total VRAM to 145.11 GiB).

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

DeepSeek V4 Flash VRAM scaling from 4K to 128K context (4K: 145.03 GiB, 8K: 145.06 GiB, 32K: 145.22 GiB, 128K: 145.85 GiB) 4K 145.03 GiB (KV 0.03 GiB) 8K 145.06 GiB (KV 0.06 GiB) 32K 145.22 GiB (KV 0.22 GiB) 128K 145.85 GiB (KV 0.85 GiB) Ryzen AI Max+ 395 128GB (128GB)
Model weights remain constant while attention KV cache grows with context length and batch size.

Quantization Precision & Memory Matrix — DeepSeek V4 Flash @ 8K

Quantization Format Weights (GiB) KV Cache Total VRAM (GiB) Decode Speed Compatibility (Ryzen AI Max+ 395 128GB) Recommended GPU Data Source
Native Weights (BF16 / FP8 / MXFP4 — Lossless) (FP4FP8) 146.8 GiB 0.06 GiB 147.42 GiB (158.29 GB) ~18.3 tok/s Exceeds Available GPU Memory Instinct MI300X 192GB (192GB) Official checkpoint size
FP8 / GGUF Q8_0 (8-bit — Near-Lossless) (Q8_0) 146.5 GiB 0.06 GiB 147.12 GiB (157.97 GB) ~18.4 tok/s Exceeds Available GPU Memory Instinct MI300X 192GB (192GB) Measured GGUF shard size
GGUF Q6_K (6-bit — High Precision) (Q6_K) 145.8 GiB 0.06 GiB 146.42 GiB (157.22 GB) ~18.6 tok/s Exceeds Available GPU Memory Instinct MI300X 192GB (192GB) Measured GGUF shard size
GGUF Q5_K_M (5-bit — Balanced) (Q5_K_M) 145.1 GiB 0.06 GiB 145.72 GiB (156.47 GB) ~18.8 tok/s Exceeds Available GPU Memory Instinct MI300X 192GB (192GB) Measured GGUF shard size
Q4_K_M / UD-Q4_K_XL / AWQ (4-bit — Recommended) (UD-Q4_K_XL) 144.44 GiB 0.06 GiB 145.06 GiB (155.76 GB) ~19 tok/s Exceeds Available GPU Memory Instinct MI300X 192GB (192GB) Measured GGUF shard size
GGUF IQ4_XS (4.25-bit — Compact) (IQ4_XS) 128.43 GiB 0.06 GiB 129.05 GiB (138.57 GB) ~25.3 tok/s Exceeds Available GPU Memory Instinct MI300X 192GB (192GB) Measured GGUF shard size
GGUF Q3_K_M (3-bit — Low VRAM) (Q3_K_M) 120.44 GiB 0.06 GiB 121.06 GiB (129.99 GB) ~27.3 tok/s Tight Fit (<8% Headroom) Instinct MI300X 192GB (192GB) Measured GGUF shard size
GGUF UD-Q2_K_XL (2-bit — Extreme MoE) (UD-Q2_K_XL) 90.18 GiB 0.06 GiB 90.8 GiB (97.5 GB) ~35.7 tok/s Fits with Safe Headroom Instinct MI300X 192GB (192GB) Measured GGUF shard size

Technical Specifications & FAQ

Does DeepSeek V4 Flash fit on Ryzen AI Max+ 395 128GB (128GB)?

In 4-bit (UD-Q4_K_XL) at 8,192 context, DeepSeek V4 Flash requires 145.06 GiB total GPU memory (144.44 GiB weights + 0.06 GiB KV cache + 0.56 GiB runtime buffers) against 99.8 GiB usable on Ryzen AI Max+ 395 128GB, delivering ~19 tok/s decode.

How much VRAM does DeepSeek V4 Flash need at 32K context?

At 32,768 context tokens, DeepSeek V4 Flash (Sparse MoE + GQA) allocates 0.22 GiB for KV cache and state, bringing total 4-bit VRAM to 145.22 GiB (155.93 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)B300 288GB Blackwell Ultra (288GB)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)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 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 M3.1 Flash (428B-A23B)MiniMax 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