RTX 3080 10GB

// weights Q3_K_M for StarCoder2 15B (15B params)
weights = params × bits ÷ 8
        = 15 × 3.44 ÷ 8
        = 6.45 GB

// budget on RTX 3080 (10GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.25 GB    (runtime, cuda, allocator)
safety    = 1.50 GB    (15% of 10GB)
budget    = vram − safety − kv − overhead
          = 10 − 1.50 − 0.05 − 1.25
          = 7.20 GB

// fit decision
6.45 ≤ 7.20  → FITS
headroom  = 0.75 GB of weights budget left

// weights Q3_K_M for Phi-4 (14.7B params)
weights = params × bits ÷ 8
        = 14.7 × 3.44 ÷ 8
        = 6.32 GB

// budget on RTX 3080 (10GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.25 GB    (runtime, cuda, allocator)
safety    = 1.50 GB    (15% of 10GB)
budget    = vram − safety − kv − overhead
          = 10 − 1.50 − 0.05 − 1.25
          = 7.20 GB

// fit decision
6.32 ≤ 7.20  → FITS
headroom  = 0.88 GB of weights budget left

// weights Q5_K_M for Qwen 3.5 9B (9B params)
weights = params × bits ÷ 8
        = 9 × 5.5 ÷ 8
        = 6.19 GB

// budget on RTX 3080 (10GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.25 GB    (runtime, cuda, allocator)
safety    = 1.50 GB    (15% of 10GB)
budget    = vram − safety − kv − overhead
          = 10 − 1.50 − 0.05 − 1.25
          = 7.20 GB

// fit decision
6.19 ≤ 7.20  → FITS
headroom  = 1.01 GB of weights budget left

// weights Q5_K_M for Gemma 2 9B (9B params)
weights = params × bits ÷ 8
        = 9 × 5.5 ÷ 8
        = 6.19 GB

// budget on RTX 3080 (10GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.25 GB    (runtime, cuda, allocator)
safety    = 1.50 GB    (15% of 10GB)
budget    = vram − safety − kv − overhead
          = 10 − 1.50 − 0.05 − 1.25
          = 7.20 GB

// fit decision
6.19 ≤ 7.20  → FITS
headroom  = 1.01 GB of weights budget left

// weights Q6_K for Llama 3.1 8B (8B params)
weights = params × bits ÷ 8
        = 8 × 6.56 ÷ 8
        = 6.56 GB

// budget on RTX 3080 (10GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.25 GB    (runtime, cuda, allocator)
safety    = 1.50 GB    (15% of 10GB)
budget    = vram − safety − kv − overhead
          = 10 − 1.50 − 0.05 − 1.25
          = 7.20 GB

// fit decision
6.56 ≤ 7.20  → FITS
headroom  = 0.64 GB of weights budget left

// weights Q6_K for Granite 8B Code (8B params)
weights = params × bits ÷ 8
        = 8 × 6.56 ÷ 8
        = 6.56 GB

// budget on RTX 3080 (10GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.25 GB    (runtime, cuda, allocator)
safety    = 1.50 GB    (15% of 10GB)
budget    = vram − safety − kv − overhead
          = 10 − 1.50 − 0.05 − 1.25
          = 7.20 GB

// fit decision
6.56 ≤ 7.20  → FITS
headroom  = 0.64 GB of weights budget left

// weights Q6_K for Mistral 7B v0.3 (7.2B params)
weights = params × bits ÷ 8
        = 7.2 × 6.56 ÷ 8
        = 5.90 GB

// budget on RTX 3080 (10GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.25 GB    (runtime, cuda, allocator)
safety    = 1.50 GB    (15% of 10GB)
budget    = vram − safety − kv − overhead
          = 10 − 1.50 − 0.05 − 1.25
          = 7.20 GB

// fit decision
5.90 ≤ 7.20  → FITS
headroom  = 1.29 GB of weights budget left

// weights Q6_K for Qwen 2.5 7B (7B params)
weights = params × bits ÷ 8
        = 7 × 6.56 ÷ 8
        = 5.74 GB

// budget on RTX 3080 (10GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.25 GB    (runtime, cuda, allocator)
safety    = 1.50 GB    (15% of 10GB)
budget    = vram − safety − kv − overhead
          = 10 − 1.50 − 0.05 − 1.25
          = 7.20 GB

// fit decision
5.74 ≤ 7.20  → FITS
headroom  = 1.46 GB of weights budget left

// weights Q8_0 for Gemma 4 E4B (4B params)
weights = params × bits ÷ 8
        = 4 × 8.5 ÷ 8
        = 4.25 GB

// budget on RTX 3080 (10GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.25 GB    (runtime, cuda, allocator)
safety    = 1.50 GB    (15% of 10GB)
budget    = vram − safety − kv − overhead
          = 10 − 1.50 − 0.05 − 1.25
          = 7.20 GB

// fit decision
4.25 ≤ 7.20  → FITS
headroom  = 2.95 GB of weights budget left

// weights Q8_0 for Phi-4 Mini (3.8B params)
weights = params × bits ÷ 8
        = 3.8 × 8.5 ÷ 8
        = 4.04 GB

// budget on RTX 3080 (10GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.25 GB    (runtime, cuda, allocator)
safety    = 1.50 GB    (15% of 10GB)
budget    = vram − safety − kv − overhead
          = 10 − 1.50 − 0.05 − 1.25
          = 7.20 GB

// fit decision
4.04 ≤ 7.20  → FITS
headroom  = 3.16 GB of weights budget left

// weights FP16/BF16 for Llama 3.2 3B (3.21B params)
weights = params × bits ÷ 8
        = 3.21 × 16 ÷ 8
        = 6.42 GB

// budget on RTX 3080 (10GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.25 GB    (runtime, cuda, allocator)
safety    = 1.50 GB    (15% of 10GB)
budget    = vram − safety − kv − overhead
          = 10 − 1.50 − 0.05 − 1.25
          = 7.20 GB

// fit decision
6.42 ≤ 7.20  → FITS
headroom  = 0.78 GB of weights budget left

// weights FP16/BF16 for Llama 3.2 1B (1.23B params)
weights = params × bits ÷ 8
        = 1.23 × 16 ÷ 8
        = 2.46 GB

// budget on RTX 3080 (10GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 1.25 GB    (runtime, cuda, allocator)
safety    = 1.50 GB    (15% of 10GB)
budget    = vram − safety − kv − overhead
          = 10 − 1.50 − 0.05 − 1.25
          = 7.20 GB

// fit decision
2.46 ≤ 7.20  → FITS
headroom  = 4.74 GB of weights budget left