Can I run Muse Spark Open 816B (MoE) locally?
Muse Spark Open 816B (MoE) has 822.04 billion parameters (Hugging Face ). At the popular Q4_K_M quantization it needs roughly 524.5 GB of GPU memory ESTIMATE . It has been on Hugging Face since 2026-09-15. It's a mixture-of-experts model: all weights must fit in memory, but it generates faster than a dense model of the same size.
Check your exact hardware →
Memory needed by quantization Quant Weights Total needed (est.) IQ1_S 160.3 GB 169.6 GB IQ2_XXS 211.7 GB 223.5 GB IQ2_M 277.4 GB 292.6 GB Q2_K 325.7 GB 343.3 GB IQ3_XXS 314.4 GB 331.4 GB Q3_K_S 375.1 GB 395.1 GB IQ3_M 376.1 GB 396.1 GB Q3_K_M 411.0 GB 432.8 GB IQ4_XS 436.7 GB 459.8 GB Q4_0 462.4 GB 486.8 GB Q4_K_M 498.4 GB 524.5 GB MXFP4 475.8 GB 500.8 GB Q5_K_M 584.7 GB 615.2 GB Q6_K 674.1 GB 709.0 GB Q8_0 873.4 GB 918.3 GB F16 1644.1 GB 1727.5 GB
KV cache by context length ESTIMATE Context f16 q8_0 q4_0 4k 0.65 GB 0.35 GB 0.18 GB 8k 0.92 GB 0.49 GB 0.26 GB 16k 1.46 GB 0.78 GB 0.41 GB 32k 2.53 GB 1.35 GB 0.71 GB 64k 4.68 GB 2.49 GB 1.32 GB 128k 8.98 GB 4.77 GB 2.52 GB
From config.json : 16 KV-caching attention layers, 16 KV heads × 64 head dim, plus 46 sliding-window layers (2048 tokens). Max context 1024k. The tables above assume 4k context.
Which GPUs and Macs can run it? Hardware Memory Q4_K_M Q8_0 F16 RTX 5090 32 GB ❌ ❌ ❌ RTX 5080 16 GB ❌ ❌ ❌ RTX 5070 Ti 16 GB ❌ ❌ ❌ RTX 5070 12 GB ❌ ❌ ❌ RTX 5060 Ti 16GB 16 GB ❌ ❌ ❌ RTX 5060 Ti 8GB 8 GB ❌ ❌ ❌ RTX 5060 8 GB ❌ ❌ ❌ RTX 5050 8 GB ❌ ❌ ❌ RTX 4090 24 GB ❌ ❌ ❌ RTX 4080 Super 16 GB ❌ ❌ ❌ RTX 4080 16 GB ❌ ❌ ❌ RTX 4070 Ti Super 16 GB ❌ ❌ ❌ RTX 4070 Ti 12 GB ❌ ❌ ❌ RTX 4070 Super 12 GB ❌ ❌ ❌ RTX 4070 12 GB ❌ ❌ ❌ RTX 4060 Ti 16GB 16 GB ❌ ❌ ❌ RTX 4060 Ti 8GB 8 GB ❌ ❌ ❌ RTX 4060 8 GB ❌ ❌ ❌ RTX 3090 Ti 24 GB ❌ ❌ ❌ RTX 3090 24 GB ❌ ❌ ❌ RTX 3080 Ti 12 GB ❌ ❌ ❌ RTX 3080 12GB 12 GB ❌ ❌ ❌ RTX 3080 10GB 10 GB ❌ ❌ ❌ RTX 3070 Ti 8 GB ❌ ❌ ❌ RTX 3070 8 GB ❌ ❌ ❌ RTX 3060 Ti 8 GB ❌ ❌ ❌ RTX 3060 12GB 12 GB ❌ ❌ ❌ RTX 3060 8GB 8 GB ❌ ❌ ❌ RTX 3050 8GB 8 GB ❌ ❌ ❌ RTX 3050 6GB 6 GB ❌ ❌ ❌ RTX 2080 Ti 11 GB ❌ ❌ ❌ RTX 2080 Super 8 GB ❌ ❌ ❌ RTX 2080 8 GB ❌ ❌ ❌ RTX 2070 Super 8 GB ❌ ❌ ❌ RTX 2070 8 GB ❌ ❌ ❌ RTX 2060 Super 8 GB ❌ ❌ ❌ RTX 2060 6GB 6 GB ❌ ❌ ❌ RTX 2060 12GB 12 GB ❌ ❌ ❌ GTX 1080 Ti 11 GB ❌ ❌ ❌ GTX 1660 Super 6 GB ❌ ❌ ❌ RTX 5090 Laptop 24 GB ❌ ❌ ❌ RTX 4090 Laptop 16 GB ❌ ❌ ❌ RTX PRO 6000 Blackwell 96 GB ❌ ❌ ❌ RTX 6000 Ada 48 GB ❌ ❌ ❌ RTX A6000 48 GB ❌ ❌ ❌ RTX A5000 24 GB ❌ ❌ ❌ RTX A4000 16 GB ❌ ❌ ❌ Tesla P40 24 GB ❌ ❌ ❌ DGX Spark (GB10) 128 GB ❌ ❌ ❌ Radeon RX 9070 XT 16 GB ❌ ❌ ❌ Radeon RX 9070 16 GB ❌ ❌ ❌ Radeon RX 9060 XT 16GB 16 GB ❌ ❌ ❌ Radeon RX 9060 XT 8GB 8 GB ❌ ❌ ❌ Radeon RX 7900 XTX 24 GB ❌ ❌ ❌ Radeon RX 7900 XT 20 GB ❌ ❌ ❌ Radeon RX 7900 GRE 16 GB ❌ ❌ ❌ Radeon RX 7800 XT 16 GB ❌ ❌ ❌ Radeon RX 7700 XT 12 GB ❌ ❌ ❌ Radeon RX 7600 XT 16 GB ❌ ❌ ❌ Radeon RX 7600 8 GB ❌ ❌ ❌ Radeon RX 6950 XT 16 GB ❌ ❌ ❌ Radeon RX 6900 XT 16 GB ❌ ❌ ❌ Radeon RX 6800 XT 16 GB ❌ ❌ ❌ Radeon RX 6800 16 GB ❌ ❌ ❌ Radeon RX 6750 XT 12 GB ❌ ❌ ❌ Radeon RX 6700 XT 12 GB ❌ ❌ ❌ Radeon RX 6650 XT 8 GB ❌ ❌ ❌ Radeon RX 6600 XT 8 GB ❌ ❌ ❌ Radeon RX 6600 8 GB ❌ ❌ ❌ Radeon AI Pro R9700 32 GB ❌ ❌ ❌ Intel Arc B580 12 GB ❌ ❌ ❌ Intel Arc B570 10 GB ❌ ❌ ❌ Intel Arc A770 16GB 16 GB ❌ ❌ ❌ Intel Arc A770 8GB 8 GB ❌ ❌ ❌ Intel Arc A750 8 GB ❌ ❌ ❌ Intel Arc Pro B60 24 GB ❌ ❌ ❌ GTX 1650 4 GB ❌ ❌ ❌ GTX 1650 Super 4 GB ❌ ❌ ❌ GTX 1650 Laptop 4 GB ❌ ❌ ❌ GTX 1050 Ti 4 GB ❌ ❌ ❌ RTX 3050 Laptop 4GB 4 GB ❌ ❌ ❌ RTX 2050 Laptop 4 GB ❌ ❌ ❌ Quadro T1000 4 GB ❌ ❌ ❌ RX 6500 XT 4GB 4 GB ❌ ❌ ❌ RX 6400 4 GB ❌ ❌ ❌ Intel Arc A380 6 GB ❌ ❌ ❌ Apple M1 (16 GB) 16 GB ❌ ❌ ❌ Apple M1 Pro (16 GB) 16 GB ❌ ❌ ❌ Apple M1 Pro (32 GB) 32 GB ❌ ❌ ❌ Apple M1 Max (32 GB) 32 GB ❌ ❌ ❌ Apple M1 Max (64 GB) 64 GB ❌ ❌ ❌ Apple M1 Ultra (64 GB) 64 GB ❌ ❌ ❌ Apple M1 Ultra (128 GB) 128 GB ❌ ❌ ❌ Apple M2 (16 GB) 16 GB ❌ ❌ ❌ Apple M2 (24 GB) 24 GB ❌ ❌ ❌ Apple M2 Pro (16 GB) 16 GB ❌ ❌ ❌ Apple M2 Pro (32 GB) 32 GB ❌ ❌ ❌ Apple M2 Max (32 GB) 32 GB ❌ ❌ ❌ Apple M2 Max (64 GB) 64 GB ❌ ❌ ❌ Apple M2 Max (96 GB) 96 GB ❌ ❌ ❌ Apple M2 Ultra (64 GB) 64 GB ❌ ❌ ❌ Apple M2 Ultra (128 GB) 128 GB ❌ ❌ ❌ Apple M2 Ultra (192 GB) 192 GB ❌ ❌ ❌ Apple M3 (16 GB) 16 GB ❌ ❌ ❌ Apple M3 (24 GB) 24 GB ❌ ❌ ❌ Apple M3 Pro (36 GB) 36 GB ❌ ❌ ❌ Apple M3 Max (36 GB) 36 GB ❌ ❌ ❌ Apple M3 Max (48 GB) 48 GB ❌ ❌ ❌ Apple M3 Max (64 GB) 64 GB ❌ ❌ ❌ Apple M3 Max (96 GB) 96 GB ❌ ❌ ❌ Apple M3 Max (128 GB) 128 GB ❌ ❌ ❌ Apple M3 Ultra (96 GB) 96 GB ❌ ❌ ❌ Apple M3 Ultra (256 GB) 256 GB ❌ ❌ ❌ Apple M3 Ultra (512 GB) 512 GB ❌ ❌ ❌ Apple M4 (16 GB) 16 GB ❌ ❌ ❌ Apple M4 (24 GB) 24 GB ❌ ❌ ❌ Apple M4 (32 GB) 32 GB ❌ ❌ ❌ Apple M4 Pro (24 GB) 24 GB ❌ ❌ ❌ Apple M4 Pro (48 GB) 48 GB ❌ ❌ ❌ Apple M4 Pro (64 GB) 64 GB ❌ ❌ ❌ Apple M4 Max (36 GB) 36 GB ❌ ❌ ❌ Apple M4 Max (48 GB) 48 GB ❌ ❌ ❌ Apple M4 Max (64 GB) 64 GB ❌ ❌ ❌ Apple M4 Max (128 GB) 128 GB ❌ ❌ ❌ Apple M5 (16 GB) 16 GB ❌ ❌ ❌ Apple M5 (24 GB) 24 GB ❌ ❌ ❌ Apple M5 (32 GB) 32 GB ❌ ❌ ❌ Apple M5 Pro (24 GB) 24 GB ❌ ❌ ❌ Apple M5 Pro (48 GB) 48 GB ❌ ❌ ❌ Apple M5 Pro (64 GB) 64 GB ❌ ❌ ❌ Apple M5 Max (36 GB) 36 GB ❌ ❌ ❌ Apple M5 Max (48 GB) 48 GB ❌ ❌ ❌ Apple M5 Max (64 GB) 64 GB ❌ ❌ ❌ Apple M5 Max (128 GB) 128 GB ❌ ❌ ❌ AMD Ryzen AI Max+ 395 Strix Halo 32GB (Linux, ~24 GB for GPU) · Radeon 8060S · Framework Desktop / GMKtec EVO-X2 / HP Z2 Mini G1a 24 GB ❌ ❌ ❌ AMD Ryzen AI Max+ 395 Strix Halo 32GB (Windows, ~24 GB for GPU) · Radeon 8060S · Framework Desktop / GMKtec EVO-X2 / HP Z2 Mini G1a 24 GB ❌ ❌ ❌ AMD Ryzen AI Max+ 395 Strix Halo 64GB (Linux, ~56 GB for GPU) · Radeon 8060S · Framework Desktop / GMKtec EVO-X2 / HP Z2 Mini G1a 56 GB ❌ ❌ ❌ AMD Ryzen AI Max+ 395 Strix Halo 64GB (Windows, ~48 GB for GPU) · Radeon 8060S · Framework Desktop / GMKtec EVO-X2 / HP Z2 Mini G1a 48 GB ❌ ❌ ❌ AMD Ryzen AI Max+ 395 Strix Halo 96GB (Linux, ~88 GB for GPU) · Radeon 8060S · Framework Desktop / GMKtec EVO-X2 / HP Z2 Mini G1a 88 GB ❌ ❌ ❌ AMD Ryzen AI Max+ 395 Strix Halo 96GB (Windows, ~72 GB for GPU) · Radeon 8060S · Framework Desktop / GMKtec EVO-X2 / HP Z2 Mini G1a 72 GB ❌ ❌ ❌ AMD Ryzen AI Max+ 395 Strix Halo 128GB (Linux, ~120 GB for GPU) · Radeon 8060S · Framework Desktop / GMKtec EVO-X2 / HP Z2 Mini G1a 120 GB ❌ ❌ ❌ AMD Ryzen AI Max+ 395 Strix Halo 128GB (Windows, ~96 GB for GPU) · Radeon 8060S · Framework Desktop / GMKtec EVO-X2 / HP Z2 Mini G1a 96 GB ❌ ❌ ❌ AMD Ryzen AI Max 390/385 Strix Halo 32GB (Linux, ~24 GB for GPU) · Radeon 8050S · Framework Desktop / GMKtec EVO-X2 / HP Z2 Mini G1a 24 GB ❌ ❌ ❌ AMD Ryzen AI Max 390/385 Strix Halo 32GB (Windows, ~24 GB for GPU) · Radeon 8050S · Framework Desktop / GMKtec EVO-X2 / HP Z2 Mini G1a 24 GB ❌ ❌ ❌ AMD Ryzen AI Max 390/385 Strix Halo 64GB (Linux, ~56 GB for GPU) · Radeon 8050S · Framework Desktop / GMKtec EVO-X2 / HP Z2 Mini G1a 56 GB ❌ ❌ ❌ AMD Ryzen AI Max 390/385 Strix Halo 64GB (Windows, ~48 GB for GPU) · Radeon 8050S · Framework Desktop / GMKtec EVO-X2 / HP Z2 Mini G1a 48 GB ❌ ❌ ❌ AMD Ryzen AI Max 390/385 Strix Halo 128GB (Linux, ~120 GB for GPU) · Radeon 8050S · Framework Desktop / GMKtec EVO-X2 / HP Z2 Mini G1a 120 GB ❌ ❌ ❌ AMD Ryzen AI Max 390/385 Strix Halo 128GB (Windows, ~96 GB for GPU) · Radeon 8050S · Framework Desktop / GMKtec EVO-X2 / HP Z2 Mini G1a 96 GB ❌ ❌ ❌
✅ fits · ⚠️ fits with little headroom · ❌ doesn't fit on this one device (multiple GPUs or RAM offload may work; check your setup ). Apple Silicon assumes macOS lets the GPU use ~67–75% of unified memory. All fit verdicts are estimates.
Get the weekly "new models & what it takes to run them" email Free. One email a week. Unsubscribe anytime.
