Strix HaloOut of stock
Framework Desktop DIY
Ryzen AI Max+ 395, 128GB
$3,449at Framework store
256 GB/s
Check priceMemory guide · 128GB unified memory
What can you run on 128GB unified memory?
14 of the 16 open models we track fit in 128GB unified memory without spilling into system RAM, the biggest being Qwen3.5 122B-A10B. Speeds range from about 5 to 72 tokens/sec depending on the model and machine. Prices & stock checked 2026-10-23. Prices change daily and some retailers block our checker, so always confirm on the store page.
Models that fit in 128GB unified memory
| Model | Needs at 4-bit | Strix Halo256 GB/s | DGX Spark273 GB/s |
|---|---|---|---|
| Qwen3.5 9Bdense · 9B active · chat, small agents | ~7 GB | Fits~21 tok/sQ8_0 | Fits~22 tok/sQ8_0 |
| Gemma 4 12Bdense · 12B active · chat, vision | ~9 GB | Fits~16 tok/sQ8_0 | Fits~16 tok/sQ8_0 |
| GPT-OSS 20BMoE · 3.6B active · reasoning, tools | ~13 GB | Fits~42 tok/sQ8_0 | Fits~46 tok/sQ8_0 |
| Gemma 4 26B-A4BMoE · 3.8B active · fast general chat | ~18 GB | Fits~40 tok/sQ8_0 | Fits~44 tok/sQ8_0 |
| Qwen3.8 27Bdense · 27B active · coding, vision, general | ~19 GB | Fits~12 tok/sQ4_K_M | Fits~13 tok/sQ4_K_M |
| GLM-4.7-Flash 30B-A3BMoE · 3B active · coding agents | ~21 GB | Fits~48 tok/sQ8_0 | Fits~54 tok/sQ8_0 |
| Gemma 4 31Bdense · 31B active · general, vision | ~21 GB | Fits~11 tok/sQ4_K_M | Fits~11 tok/sQ4_K_M |
| Qwen3.6 35B-A3BMoE · 3B active · fast coding & chat | ~24 GB | Fits~48 tok/sQ8_0 | Fits~54 tok/sQ8_0 |
| Llama 3.3 70Bdense · 70B active · general | ~46 GB | Fits~5 tok/sQ4_K_M | Fits~5 tok/sQ4_K_M |
| Qwen3-Next 80B-A3BMoE · 3B active · long context, fast | ~52 GB | Fits~72 tok/sQ4_K_M | Fits~54 tok/sQ8_0 |
| Llama 4 Scout 109B-A17BMoE · 17B active · multimodal, long context | ~71 GB | Fits~18 tok/sQ4_K_M | Fits~19 tok/sQ4_K_M |
| GPT-OSS 120BMoE · 5.1B active · reasoning, tools | ~67 GB | Fits~55 tok/sMXFP4 (native) | Fits~61 tok/sMXFP4 (native) |
| Nemotron 3 Super 120B-A12BMoE · 12B active · agents | ~78 GB | Fits~25 tok/sQ4_K_M | Fits~26 tok/sQ4_K_M |
| Qwen3.5 122B-A10BMoE · 10B active · strong general / coding | ~79 GB | Fits~29 tok/sQ4_K_M | Fits~31 tok/sQ4_K_M |
Each column shows the quant we'd use (8-bit when it fits comfortably and still runs 10+ tok/s, otherwise 4-bit) and the estimated decode speed for one user at ~8K context. How we estimate
Just out of reach: Qwen3 235B-A22B (~151 GB at 4-bit), Muse Spark Open 816B-A42B (~521 GB at 4-bit).
9 tracked rigs with 128GB unified memory
Strix HaloAvailable
GMKtec EVO-X2
Ryzen AI Max+ 395, 128GB / 1TB
$3,500at GMKtec store
256 GB/s
Check price
Strix HaloAvailable
Minisforum MS-S1 Max
Ryzen AI Max+ 395, 128GB / 2TB
$3,919at Minisforum store
256 GB/s
Check price
Strix HaloPre-order
Beelink GTR9 Pro
Ryzen AI Max+ 395, 128GB / 2TB
$4,349at Beelink store
256 GB/s
Check price
DGX SparkAvailable
NVIDIA DGX Spark
Founders Edition, 4TB
$6,950at NVIDIA Marketplace
273 GB/s
Check price
FAQ
What's the biggest model I can run on 128GB unified memory?
Of the models we track, Qwen3.5 122B-A10B is the largest that fits fully in 128GB unified memory (Q4_K_M, about 79 GB including runtime and ~8K context).
What's the cheapest rig with 128GB unified memory?
Of the machines we track, the Framework Desktop DIY at $3,449 (Framework store, checked 2026-10-23).
Can I run Llama 3.3 70B on 128GB unified memory?
Yes, at 4-bit it needs about 46 GB including runtime and ~8K context.
How are fit and speed estimated?
Memory needed = weights + 1.5 GB + 5% of weights. Decode speed ≈ 1 ÷ (active weights in GB ÷ (efficiency × memory bandwidth) + a small fixed per-token overhead), with efficiency and overhead fitted per platform to 43 public llama.cpp/MLX benchmarks (median error about 5%). They're estimates, not benchmarks.