RTX 4080 16GB

// weights Q3_K_M for Qwen 3.6 27B (27B params)
weights = params × bits ÷ 8
        = 27 × 3.44 ÷ 8
        = 11.61 GB

// budget on RTX 4080 (16GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.40 GB    (runtime, cuda, allocator)
safety    = 2.40 GB    (15% of 16GB)
budget    = vram − safety − kv − overhead
          = 16 − 2.40 − 0.05 − 1.40
          = 12.15 GB

// fit decision
11.61 ≤ 12.15  → FITS
headroom  = 0.54 GB of weights budget left

// weights Q3_K_M for Gemma 4 26B A4B (26B params)
weights = params × bits ÷ 8
        = 26 × 3.44 ÷ 8
        = 11.18 GB

// budget on RTX 4080 (16GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.40 GB    (runtime, cuda, allocator)
safety    = 2.40 GB    (15% of 16GB)
budget    = vram − safety − kv − overhead
          = 16 − 2.40 − 0.05 − 1.40
          = 12.15 GB

// fit decision
11.18 ≤ 12.15  → FITS
headroom  = 0.97 GB of weights budget left

// weights Q3_K_M for Mistral Small 3 (24B params)
weights = params × bits ÷ 8
        = 24 × 3.44 ÷ 8
        = 10.32 GB

// budget on RTX 4080 (16GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.40 GB    (runtime, cuda, allocator)
safety    = 2.40 GB    (15% of 16GB)
budget    = vram − safety − kv − overhead
          = 16 − 2.40 − 0.05 − 1.40
          = 12.15 GB

// fit decision
10.32 ≤ 12.15  → FITS
headroom  = 1.83 GB of weights budget left

// weights Q4_K_M for gpt-oss 20B (20.9B params)
weights = params × bits ÷ 8
        = 20.9 × 4.5 ÷ 8
        = 11.76 GB

// budget on RTX 4080 (16GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.40 GB    (runtime, cuda, allocator)
safety    = 2.40 GB    (15% of 16GB)
budget    = vram − safety − kv − overhead
          = 16 − 2.40 − 0.05 − 1.40
          = 12.15 GB

// fit decision
11.76 ≤ 12.15  → FITS
headroom  = 0.39 GB of weights budget left

// weights Q5_K_M for StarCoder2 15B (15B params)
weights = params × bits ÷ 8
        = 15 × 5.5 ÷ 8
        = 10.31 GB

// budget on RTX 4080 (16GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.40 GB    (runtime, cuda, allocator)
safety    = 2.40 GB    (15% of 16GB)
budget    = vram − safety − kv − overhead
          = 16 − 2.40 − 0.05 − 1.40
          = 12.15 GB

// fit decision
10.31 ≤ 12.15  → FITS
headroom  = 1.84 GB of weights budget left

// weights Q5_K_M for Phi-4 (14.7B params)
weights = params × bits ÷ 8
        = 14.7 × 5.5 ÷ 8
        = 10.11 GB

// budget on RTX 4080 (16GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.40 GB    (runtime, cuda, allocator)
safety    = 2.40 GB    (15% of 16GB)
budget    = vram − safety − kv − overhead
          = 16 − 2.40 − 0.05 − 1.40
          = 12.15 GB

// fit decision
10.11 ≤ 12.15  → FITS
headroom  = 2.04 GB of weights budget left

// weights Q8_0 for Qwen 3.5 9B (9B params)
weights = params × bits ÷ 8
        = 9 × 8.5 ÷ 8
        = 9.56 GB

// budget on RTX 4080 (16GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.40 GB    (runtime, cuda, allocator)
safety    = 2.40 GB    (15% of 16GB)
budget    = vram − safety − kv − overhead
          = 16 − 2.40 − 0.05 − 1.40
          = 12.15 GB

// fit decision
9.56 ≤ 12.15  → FITS
headroom  = 2.59 GB of weights budget left

// weights Q8_0 for Gemma 2 9B (9B params)
weights = params × bits ÷ 8
        = 9 × 8.5 ÷ 8
        = 9.56 GB

// budget on RTX 4080 (16GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.40 GB    (runtime, cuda, allocator)
safety    = 2.40 GB    (15% of 16GB)
budget    = vram − safety − kv − overhead
          = 16 − 2.40 − 0.05 − 1.40
          = 12.15 GB

// fit decision
9.56 ≤ 12.15  → FITS
headroom  = 2.59 GB of weights budget left

// weights Q8_0 for Llama 3.1 8B (8B params)
weights = params × bits ÷ 8
        = 8 × 8.5 ÷ 8
        = 8.50 GB

// budget on RTX 4080 (16GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.40 GB    (runtime, cuda, allocator)
safety    = 2.40 GB    (15% of 16GB)
budget    = vram − safety − kv − overhead
          = 16 − 2.40 − 0.05 − 1.40
          = 12.15 GB

// fit decision
8.50 ≤ 12.15  → FITS
headroom  = 3.65 GB of weights budget left

// weights Q8_0 for Granite 8B Code (8B params)
weights = params × bits ÷ 8
        = 8 × 8.5 ÷ 8
        = 8.50 GB

// budget on RTX 4080 (16GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.40 GB    (runtime, cuda, allocator)
safety    = 2.40 GB    (15% of 16GB)
budget    = vram − safety − kv − overhead
          = 16 − 2.40 − 0.05 − 1.40
          = 12.15 GB

// fit decision
8.50 ≤ 12.15  → FITS
headroom  = 3.65 GB of weights budget left

// weights Q8_0 for Mistral 7B v0.3 (7.2B params)
weights = params × bits ÷ 8
        = 7.2 × 8.5 ÷ 8
        = 7.65 GB

// budget on RTX 4080 (16GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.40 GB    (runtime, cuda, allocator)
safety    = 2.40 GB    (15% of 16GB)
budget    = vram − safety − kv − overhead
          = 16 − 2.40 − 0.05 − 1.40
          = 12.15 GB

// fit decision
7.65 ≤ 12.15  → FITS
headroom  = 4.50 GB of weights budget left

// weights Q8_0 for Qwen 2.5 7B (7B params)
weights = params × bits ÷ 8
        = 7 × 8.5 ÷ 8
        = 7.44 GB

// budget on RTX 4080 (16GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.40 GB    (runtime, cuda, allocator)
safety    = 2.40 GB    (15% of 16GB)
budget    = vram − safety − kv − overhead
          = 16 − 2.40 − 0.05 − 1.40
          = 12.15 GB

// fit decision
7.44 ≤ 12.15  → FITS
headroom  = 4.71 GB of weights budget left