VRAM TIER DIRECTORY โข 128 GB ENVELOPE10 of 12 Models Fit
Best Local LLMs for 128GB VRAM
Typical Hardware:Apple Mac Studio M4 Max 128GBAMD Strix Halo Ryzen AI Max+ 395 128GBNVIDIA DGX Spark 128GB
Workstation-class 110GBโ118GB usable memory. Runs 70B models at uncompressed FP16/Q8, Llama 4 Scout 109B with 128K context, and 552B MoE models (DeepSeek V4.1) at compact quants.
โ
Runs 100% in GPU VRAM (Optimal Headroom)71.66 GB (64%)~32 t/s
Meta Llama 4 Scout (109B-A17B MoE) on Apple Mac Studio M4 Max (128GB Unified Memory)
โ Fits Comfortably in VRAM
โ FITS IN VRAM
Required VRAM
71.66 GB
64% of hardware capacity (112 GB)Est. Decode Speed
~32 tok/s
17B active params @ 546 GB/sAvailable Headroom
+40.34 GB
Safe for OS display & full contextMemory Allocation & Headroom
71.66 GB allocated of 112 GB capacity
64% utilized
Weights: 65.4 GB
KV: (32K): 3.22 GB
Overhead: 3.04 GB
Free Headroom: +40.34 GB
โก Architecture & Throughput Transparency: Single-user decode speed (~32 tok/s) is memory-bandwidth bound at batch size = 1 on Apple Mac Studio M4 Max (128GB Unified Memory) (546 GB/s). Prompt prefill / TTFT (~464 tok/s) is compute-bound and utilizes GPU Tensor cores. Actual latency varies with FlashAttention-2 and runtime engine (Ollama/vLLM/llama.cpp).
OLLAMA_FLASH_ATTENTION=1 OLLAMA_KV_CACHE_TYPE=q8_0 ollama run llama4:scoutCloud Instance Options for Meta Llama 4 Scout (109B-A17B MoE)
On-DemandDeploy this model on an on-demand cloud GPU without local hardware constraints or power usage:
RunPod Cloud GPUsBest 1-Click Ollama & vLLM
$0.34/hr (RTX 4070 Ti) โ $2.39/hr (H100)Vast.ai GPU MarketplaceCheapest Dual RTX 3090 / 4090
$0.18/hr (RTX 3060) โ $0.55/hr (RTX 4090)Lambda GPU CloudEnterprise 8x H100 Clusters
$1.99/hr (GH200 / H100)Quantization Precision Matrix
32K ctx โข KV: Q8_0| Quantization | Weights | KV Cache | Total VRAM | Est. Speed | Fit Status |
|---|---|---|---|---|---|
| FP1616 bpw | 218 GB | 3.22 GB | 229.6 GB (205%) | 0 t/s | OOM |
| Q8_08.5 bpw | 115.54 GB | 3.22 GB | 123.55 GB (110%) | ~4 t/s | PARTIAL |
| Q6_K6.56 bpw | 89.38 GB | 3.22 GB | 96.48 GB (86%) | ~24 t/s | FITS |
| Q5_K_M5.68 bpw | 77.39 GB | 3.22 GB | 84.07 GB (75%) | ~27 t/s | FITS |
| Q4_K_M4.8 bpw | 65.4 GB | 3.22 GB | 71.66 GB (64%) | ~32 t/s | FITS |
| FP44.25 bpw | 57.77 GB | 3.22 GB | 63.76 GB (57%) | ~36 t/s | FITS |
| IQ3_XS3.3 bpw | 44.69 GB | 3.22 GB | 50.22 GB (45%) | ~45 t/s | FITS |
Local Hardware Configurations & Pricing
High-bandwidth VRAM specs and current market pricesApple Silicon112 GB VRAM
Apple Mac Studio M4 Max (128GB Unified Memory)
Running Llama 4 Scout 109B & 70B Q8 locally with 128K+ context
546 GB/s145W TDP
$3,499Amazon Prime โ
Consumer GPU32 GB VRAM
NVIDIA GeForce RTX 5090 (32GB GDDR7)
Maximum tokens/sec on 27Bโ35B models & Blackwell FP4 quantization
1792 GB/s575W TDP
$1,999Amazon Prime โ
Consumer GPU24 GB VRAM
NVIDIA GeForce RTX 4090 (24GB GDDR6X)
The 24GB CUDA enthusiast standard for 14Bโ32B Q4_K_M models
1008 GB/s450W TDP
$1,699Amazon Prime โ
Multi-GPU Rig48 GB VRAM
2x NVIDIA RTX 3090 Dual Rig (48GB Total VRAM)
Best price-to-VRAM homelab rig for 70B Q4_K_M & Qwen 3.8 27B Q8
936 GB/s700W TDP
$1,550Amazon Prime โ
Export Full Architecture & Infrastructure Spec
Get an executive-ready PDF report with exact memory math, hardware benchmarks, and cost comparisons.
All Open Models Evaluated on 128GB VRAM
Standard Q4_K_M quantization with 32K context window and Q8_0 KV cache
| Model | Architecture | VRAM Required | Headroom | Est. Decode Speed | Compatibility | Ollama Command |
|---|---|---|---|---|---|---|
| Qwen 3.6 35B-A3B MoE Ultra-Fast 100+ tok/s Coding & Local RAG on 24GB GPUs | 35B MoE | 24.1 GB | +87.9 GB free | ~159 t/s | FITS | ollama run qwen3.6:35b-a3b |
| Microsoft Phi-4 Mini 3.8B Ultra-Light Laptop & CPU/Integrated GPU Inference | 3.8B Dense | 3.92 GB | +108.1 GB free | ~141 t/s | FITS | ollama run phi4-mini:3.8b |
| Google Gemma 4 E4B (Edge 4B) 8GB Laptop GPUs, Local Autocomplete & Fast Summarization | 4.1B Dense | 4.24 GB | +107.8 GB free | ~130 t/s | FITS | ollama run gemma4:4b |
| Google Gemma 4 26B MoE High-Speed Consumer GPU & Apple Silicon Daily Assistant | 26.4B MoE | 20.23 GB | +91.8 GB free | ~106 t/s | FITS | ollama run gemma4:26b |
| Meta Llama 3.1 8B Instruct Fast 8GBโ12GB GPU Chat, Classification & Local Agents | 8B Dense | 7.87 GB | +104.1 GB free | ~65 t/s | FITS | ollama run llama3.1:8b |
| DeepSeek-R1 Distill Qwen 14B Mainstream 12GBโ16GB Gaming GPUs Reasoning Assistant | 14.8B Dense | 13.16 GB | +98.8 GB free | ~36 t/s | FITS | ollama run deepseek-r1:14b |
| Meta Llama 4 Scout (109B-A17B MoE) Long-Context Codebase Indexing & Multi-Document Analysis | 109B MoE | 71.66 GB | +40.3 GB free | ~32 t/s | FITS | ollama run llama4:scout |
| Qwen 3.8 27B Instruct Agentic Coding, Terminal Workflows & Multi-File Refactoring | 27.2B Dense | 21.93 GB | +90.1 GB free | ~20 t/s | FITS | ollama run qwen3.8:27b |
| DeepSeek-R1 Distill Qwen 32B Local Step-by-Step Math, Logic & Code Verification | 32.8B Dense | 25.41 GB | +86.6 GB free | ~17 t/s | FITS | ollama run deepseek-r1:32b |
| Meta Llama 3.3 70B Instruct Standard 70B Dense Enterprise Baseline for 48GBโ64GB Rigs | 70.6B Dense | 49.96 GB | +62 GB free | ~8 t/s | FITS | ollama run llama3.3:70b |
| DeepSeek-V4.1-Flash (552B Engram MoE) Frontier SWE Coding (77.0 DeepSWE), 1M Context & 3D/Game Synthesis | 552B MoE | 345.59 GB | -233.6 GB short | 0 (OOM) | OOM | Deploy on Cloud GPU โ |
| Meta Muse Spark Open (816B-A42B MoE) Frontier Multi-Step Agentic Reasoning & Native Multimodal | 816B MoE | 512.86 GB | -400.9 GB short | 0 (OOM) | OOM | Deploy on Cloud GPU โ |
Need to Run Models Larger than 128GB?
If your preferred model exceeds your local 128GB GPU, you can rent high-memory on-demand cloud GPUs with zero upfront hardware costs:
24GBโ32GB Cloud VRAM
Deploy on RunPod ($0.59/hr) โRent RTX 4090 / RTX 5090
Boot an on-demand RTX 4090 (24GB) from $0.59/hr or RTX 5090 (32GB) from $0.89/hr on RunPod.
48GBโ80GB High VRAM
Rent on Vast.ai ($0.78/hr) โRent Dual 3090 or H100
Dual 3090 (48GB) from $0.78/hr on Vast.ai, or H100 (80GB HBM3) from $2.39/hr for 70B+ models.