RTX 6000 Ada 48GB

// weights Q3_K_M for Qwen 2.5 72B (72.7B params)
weights = params × bits ÷ 8
        = 72.7 × 3.44 ÷ 8
        = 31.26 GB

// budget on RTX 6000 Ada (48GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 2.20 GB    (runtime, cuda, allocator)
safety    = 7.20 GB    (15% of 48GB)
budget    = vram − safety − kv − overhead
          = 48 − 7.20 − 0.05 − 2.20
          = 38.55 GB

// fit decision
31.26 ≤ 38.55  → FITS
headroom  = 7.29 GB of weights budget left

// weights Q3_K_M for Llama 3.3 70B (70.6B params)
weights = params × bits ÷ 8
        = 70.6 × 3.44 ÷ 8
        = 30.36 GB

// budget on RTX 6000 Ada (48GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 2.20 GB    (runtime, cuda, allocator)
safety    = 7.20 GB    (15% of 48GB)
budget    = vram − safety − kv − overhead
          = 48 − 7.20 − 0.05 − 2.20
          = 38.55 GB

// fit decision
30.36 ≤ 38.55  → FITS
headroom  = 8.19 GB of weights budget left

// weights Q5_K_M for Mixtral 8x7B (46.7B params)
weights = params × bits ÷ 8
        = 46.7 × 5.5 ÷ 8
        = 32.11 GB

// budget on RTX 6000 Ada (48GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 2.20 GB    (runtime, cuda, allocator)
safety    = 7.20 GB    (15% of 48GB)
budget    = vram − safety − kv − overhead
          = 48 − 7.20 − 0.05 − 2.20
          = 38.55 GB

// fit decision
32.11 ≤ 38.55  → FITS
headroom  = 6.44 GB of weights budget left

// weights FP8/INT8 for Qwen 3.6 35B A3B (35B params)
weights = params × bits ÷ 8
        = 35 × 8 ÷ 8
        = 35.00 GB

// budget on RTX 6000 Ada (48GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 2.20 GB    (runtime, cuda, allocator)
safety    = 7.20 GB    (15% of 48GB)
budget    = vram − safety − kv − overhead
          = 48 − 7.20 − 0.05 − 2.20
          = 38.55 GB

// fit decision
35.00 ≤ 38.55  → FITS
headroom  = 3.55 GB of weights budget left

// weights Q8_0 for Yi 34B (34B params)
weights = params × bits ÷ 8
        = 34 × 8.5 ÷ 8
        = 36.13 GB

// budget on RTX 6000 Ada (48GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 2.20 GB    (runtime, cuda, allocator)
safety    = 7.20 GB    (15% of 48GB)
budget    = vram − safety − kv − overhead
          = 48 − 7.20 − 0.05 − 2.20
          = 38.55 GB

// fit decision
36.13 ≤ 38.55  → FITS
headroom  = 2.42 GB of weights budget left

// weights Q8_0 for Qwen 2.5 32B (32.5B params)
weights = params × bits ÷ 8
        = 32.5 × 8.5 ÷ 8
        = 34.53 GB

// budget on RTX 6000 Ada (48GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 2.20 GB    (runtime, cuda, allocator)
safety    = 7.20 GB    (15% of 48GB)
budget    = vram − safety − kv − overhead
          = 48 − 7.20 − 0.05 − 2.20
          = 38.55 GB

// fit decision
34.53 ≤ 38.55  → FITS
headroom  = 4.02 GB of weights budget left

// weights Q8_0 for Qwen 2.5 Coder 32B (32.5B params)
weights = params × bits ÷ 8
        = 32.5 × 8.5 ÷ 8
        = 34.53 GB

// budget on RTX 6000 Ada (48GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 2.20 GB    (runtime, cuda, allocator)
safety    = 7.20 GB    (15% of 48GB)
budget    = vram − safety − kv − overhead
          = 48 − 7.20 − 0.05 − 2.20
          = 38.55 GB

// fit decision
34.53 ≤ 38.55  → FITS
headroom  = 4.02 GB of weights budget left

// weights Q8_0 for Qwen3 30B A3B (30.5B params)
weights = params × bits ÷ 8
        = 30.5 × 8.5 ÷ 8
        = 32.41 GB

// budget on RTX 6000 Ada (48GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 2.20 GB    (runtime, cuda, allocator)
safety    = 7.20 GB    (15% of 48GB)
budget    = vram − safety − kv − overhead
          = 48 − 7.20 − 0.05 − 2.20
          = 38.55 GB

// fit decision
32.41 ≤ 38.55  → FITS
headroom  = 6.14 GB of weights budget left

// weights Q8_0 for Qwen 3.6 27B (27B params)
weights = params × bits ÷ 8
        = 27 × 8.5 ÷ 8
        = 28.69 GB

// budget on RTX 6000 Ada (48GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 2.20 GB    (runtime, cuda, allocator)
safety    = 7.20 GB    (15% of 48GB)
budget    = vram − safety − kv − overhead
          = 48 − 7.20 − 0.05 − 2.20
          = 38.55 GB

// fit decision
28.69 ≤ 38.55  → FITS
headroom  = 9.86 GB of weights budget left

// weights Q8_0 for Gemma 4 26B A4B (26B params)
weights = params × bits ÷ 8
        = 26 × 8.5 ÷ 8
        = 27.63 GB

// budget on RTX 6000 Ada (48GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 2.20 GB    (runtime, cuda, allocator)
safety    = 7.20 GB    (15% of 48GB)
budget    = vram − safety − kv − overhead
          = 48 − 7.20 − 0.05 − 2.20
          = 38.55 GB

// fit decision
27.63 ≤ 38.55  → FITS
headroom  = 10.92 GB of weights budget left

// weights Q8_0 for Mistral Small 3 (24B params)
weights = params × bits ÷ 8
        = 24 × 8.5 ÷ 8
        = 25.50 GB

// budget on RTX 6000 Ada (48GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 2.20 GB    (runtime, cuda, allocator)
safety    = 7.20 GB    (15% of 48GB)
budget    = vram − safety − kv − overhead
          = 48 − 7.20 − 0.05 − 2.20
          = 38.55 GB

// fit decision
25.50 ≤ 38.55  → FITS
headroom  = 13.05 GB of weights budget left

// weights Q8_0 for gpt-oss 20B (20.9B params)
weights = params × bits ÷ 8
        = 20.9 × 8.5 ÷ 8
        = 22.21 GB

// budget on RTX 6000 Ada (48GB) at ctx 8K, conc 1, 15% safety
kv_cache  = 0.05 GB    (1× at ctx 8K)
overhead  = 2.20 GB    (runtime, cuda, allocator)
safety    = 7.20 GB    (15% of 48GB)
budget    = vram − safety − kv − overhead
          = 48 − 7.20 − 0.05 − 2.20
          = 38.55 GB

// fit decision
22.21 ≤ 38.55  → FITS
headroom  = 16.34 GB of weights budget left