{
 "source": "https://studiotvai.com/llm-vram-calculator",
 "gpus": [
  {
   "id": "b300",
   "name": "NVIDIA B300 / GB300 · 288 GB",
   "group": "NVIDIA datacenter",
   "memoryGB": 288,
   "bandwidthTBs": 8,
   "bf16Tflops": 2250,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "b200",
   "name": "NVIDIA B200 (HGX) · 180 GB",
   "group": "NVIDIA datacenter",
   "memoryGB": 180,
   "bandwidthTBs": 8,
   "bf16Tflops": 2250,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "h200",
   "name": "NVIDIA H200 · 141 GB",
   "group": "NVIDIA datacenter",
   "memoryGB": 141,
   "bandwidthTBs": 4.8,
   "bf16Tflops": 989,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "h20",
   "name": "NVIDIA H20 · 96 GB",
   "group": "NVIDIA datacenter",
   "memoryGB": 96,
   "bandwidthTBs": 4,
   "bf16Tflops": 148,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "h100-nvl",
   "name": "NVIDIA H100 NVL · 94 GB",
   "group": "NVIDIA datacenter",
   "memoryGB": 94,
   "bandwidthTBs": 3.9,
   "bf16Tflops": 835,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "h100",
   "name": "NVIDIA H100 · 80 GB",
   "group": "NVIDIA datacenter",
   "memoryGB": 80,
   "bandwidthTBs": 3.35,
   "bf16Tflops": 989,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "a100-80",
   "name": "NVIDIA A100 · 80 GB",
   "group": "NVIDIA datacenter",
   "memoryGB": 80,
   "bandwidthTBs": 2.04,
   "bf16Tflops": 312,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "a100-40",
   "name": "NVIDIA A100 · 40 GB",
   "group": "NVIDIA datacenter",
   "memoryGB": 40,
   "bandwidthTBs": 1.555,
   "bf16Tflops": 312,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "l40s",
   "name": "NVIDIA L40S · 48 GB",
   "group": "NVIDIA datacenter",
   "memoryGB": 48,
   "bandwidthTBs": 0.864,
   "bf16Tflops": 362,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "l4",
   "name": "NVIDIA L4 · 24 GB",
   "group": "NVIDIA datacenter",
   "memoryGB": 24,
   "bandwidthTBs": 0.3,
   "bf16Tflops": 121,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "mi355x",
   "name": "AMD MI355X / MI350X · 288 GB",
   "group": "AMD & Intel",
   "memoryGB": 288,
   "bandwidthTBs": 8,
   "bf16Tflops": 2300,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "mi325x",
   "name": "AMD MI325X · 256 GB",
   "group": "AMD & Intel",
   "memoryGB": 256,
   "bandwidthTBs": 6,
   "bf16Tflops": 1307,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "mi300x",
   "name": "AMD MI300X · 192 GB",
   "group": "AMD & Intel",
   "memoryGB": 192,
   "bandwidthTBs": 5.3,
   "bf16Tflops": 1307,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "gaudi3",
   "name": "Intel Gaudi 3 · 128 GB",
   "group": "AMD & Intel",
   "memoryGB": 128,
   "bandwidthTBs": 3.7,
   "bf16Tflops": 1835,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "rtx-pro-6000",
   "name": "RTX PRO 6000 Blackwell · 96 GB",
   "group": "Workstation & consumer",
   "memoryGB": 96,
   "bandwidthTBs": 1.792,
   "bf16Tflops": 500,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "rtx-6000-ada",
   "name": "RTX 6000 Ada · 48 GB",
   "group": "Workstation & consumer",
   "memoryGB": 48,
   "bandwidthTBs": 0.96,
   "bf16Tflops": 364,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "rtx-5090",
   "name": "GeForce RTX 5090 · 32 GB",
   "group": "Workstation & consumer",
   "memoryGB": 32,
   "bandwidthTBs": 1.792,
   "bf16Tflops": 210,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "rtx-4090",
   "name": "GeForce RTX 4090 · 24 GB",
   "group": "Workstation & consumer",
   "memoryGB": 24,
   "bandwidthTBs": 1.008,
   "bf16Tflops": 165,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "rtx-3090",
   "name": "GeForce RTX 3090 · 24 GB",
   "group": "Workstation & consumer",
   "memoryGB": 24,
   "bandwidthTBs": 0.936,
   "bf16Tflops": 71,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "rtx-5080",
   "name": "GeForce RTX 5080 · 16 GB",
   "group": "Workstation & consumer",
   "memoryGB": 16,
   "bandwidthTBs": 0.96,
   "bf16Tflops": 113,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "rtx-5070-ti",
   "name": "GeForce RTX 5070 Ti · 16 GB",
   "group": "Workstation & consumer",
   "memoryGB": 16,
   "bandwidthTBs": 0.896,
   "bf16Tflops": 88,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "rtx-5070",
   "name": "GeForce RTX 5070 · 12 GB",
   "group": "Workstation & consumer",
   "memoryGB": 12,
   "bandwidthTBs": 0.672,
   "bf16Tflops": 62,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "rtx-5060-ti",
   "name": "GeForce RTX 5060 Ti · 16 GB",
   "group": "Workstation & consumer",
   "memoryGB": 16,
   "bandwidthTBs": 0.448,
   "bf16Tflops": 47,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "rtx-5060",
   "name": "GeForce RTX 5060 · 8 GB",
   "group": "Workstation & consumer",
   "memoryGB": 8,
   "bandwidthTBs": 0.448,
   "bf16Tflops": 38,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "rtx-4080",
   "name": "GeForce RTX 4080 / 4080 SUPER · 16 GB",
   "group": "Workstation & consumer",
   "memoryGB": 16,
   "bandwidthTBs": 0.717,
   "bf16Tflops": 97,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "rtx-4070-ti-super",
   "name": "GeForce RTX 4070 Ti SUPER · 16 GB",
   "group": "Workstation & consumer",
   "memoryGB": 16,
   "bandwidthTBs": 0.672,
   "bf16Tflops": 88,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "rtx-4070",
   "name": "GeForce RTX 4070 / 4070 SUPER · 12 GB",
   "group": "Workstation & consumer",
   "memoryGB": 12,
   "bandwidthTBs": 0.504,
   "bf16Tflops": 58,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "rtx-4060-ti",
   "name": "GeForce RTX 4060 Ti · 16 GB",
   "group": "Workstation & consumer",
   "memoryGB": 16,
   "bandwidthTBs": 0.288,
   "bf16Tflops": 44,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "rtx-4060",
   "name": "GeForce RTX 4060 · 8 GB",
   "group": "Workstation & consumer",
   "memoryGB": 8,
   "bandwidthTBs": 0.272,
   "bf16Tflops": 30,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "rtx-3080",
   "name": "GeForce RTX 3080 · 10 GB",
   "group": "Workstation & consumer",
   "memoryGB": 10,
   "bandwidthTBs": 0.76,
   "bf16Tflops": 60,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "rtx-3060",
   "name": "GeForce RTX 3060 · 12 GB",
   "group": "Workstation & consumer",
   "memoryGB": 12,
   "bandwidthTBs": 0.36,
   "bf16Tflops": 25,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "rtx-5090-laptop",
   "name": "GeForce RTX 5090 Laptop · 24 GB",
   "group": "Laptop GPUs",
   "memoryGB": 24,
   "bandwidthTBs": 0.896,
   "bf16Tflops": 114,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "rtx-5080-laptop",
   "name": "GeForce RTX 5080 Laptop · 16 GB",
   "group": "Laptop GPUs",
   "memoryGB": 16,
   "bandwidthTBs": 0.896,
   "bf16Tflops": 83,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "rtx-5070-ti-laptop",
   "name": "GeForce RTX 5070 Ti Laptop · 12 GB",
   "group": "Laptop GPUs",
   "memoryGB": 12,
   "bandwidthTBs": 0.672,
   "bf16Tflops": 62,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "rtx-5070-laptop",
   "name": "GeForce RTX 5070 Laptop · 8 GB",
   "group": "Laptop GPUs",
   "memoryGB": 8,
   "bandwidthTBs": 0.384,
   "bf16Tflops": 50,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "rtx-4090-laptop",
   "name": "GeForce RTX 4090 Laptop · 16 GB",
   "group": "Laptop GPUs",
   "memoryGB": 16,
   "bandwidthTBs": 0.576,
   "bf16Tflops": 86,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "rtx-4080-laptop",
   "name": "GeForce RTX 4080 Laptop · 12 GB",
   "group": "Laptop GPUs",
   "memoryGB": 12,
   "bandwidthTBs": 0.432,
   "bf16Tflops": 68,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "rtx-4070-laptop",
   "name": "GeForce RTX 4070 Laptop · 8 GB",
   "group": "Laptop GPUs",
   "memoryGB": 8,
   "bandwidthTBs": 0.256,
   "bf16Tflops": 40,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "rtx-4060-laptop",
   "name": "GeForce RTX 4060 Laptop · 8 GB",
   "group": "Laptop GPUs",
   "memoryGB": 8,
   "bandwidthTBs": 0.256,
   "bf16Tflops": 29,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "radeon-ai-pro-r9700",
   "name": "Radeon AI PRO R9700 · 32 GB",
   "group": "Radeon & Arc",
   "memoryGB": 32,
   "bandwidthTBs": 0.64,
   "bf16Tflops": 191,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "rx-7900-xtx",
   "name": "Radeon RX 7900 XTX · 24 GB",
   "group": "Radeon & Arc",
   "memoryGB": 24,
   "bandwidthTBs": 0.96,
   "bf16Tflops": 123,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "rx-9070-xt",
   "name": "Radeon RX 9070 XT · 16 GB",
   "group": "Radeon & Arc",
   "memoryGB": 16,
   "bandwidthTBs": 0.64,
   "bf16Tflops": 195,
   "fp8": true,
   "fp4": false
  },
  {
   "id": "arc-pro-b60",
   "name": "Intel Arc Pro B60 · 24 GB",
   "group": "Radeon & Arc",
   "memoryGB": 24,
   "bandwidthTBs": 0.456,
   "bf16Tflops": 98,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "arc-b580",
   "name": "Intel Arc B580 · 12 GB",
   "group": "Radeon & Arc",
   "memoryGB": 12,
   "bandwidthTBs": 0.456,
   "bf16Tflops": 116,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "mac-m5-ultra-512",
   "name": "Mac Studio M5 Ultra 512 GB · ~384 GB GPU",
   "group": "Unified memory",
   "memoryGB": 384,
   "bandwidthTBs": 1.2,
   "bf16Tflops": 70,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "mac-m5-max-128",
   "name": "Mac M5 Max 128 GB · ~96 GB GPU",
   "group": "Unified memory",
   "memoryGB": 96,
   "bandwidthTBs": 0.614,
   "bf16Tflops": 35,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "mac-m5-max-64",
   "name": "Mac M5 Max 64 GB · ~48 GB GPU",
   "group": "Unified memory",
   "memoryGB": 48,
   "bandwidthTBs": 0.614,
   "bf16Tflops": 35,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "mac-m5-max-36",
   "name": "Mac M5 Max 36 GB · ~24 GB GPU",
   "group": "Unified memory",
   "memoryGB": 24,
   "bandwidthTBs": 0.46,
   "bf16Tflops": 28,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "mac-m5-pro-64",
   "name": "Mac M5 Pro 64 GB · ~48 GB GPU",
   "group": "Unified memory",
   "memoryGB": 48,
   "bandwidthTBs": 0.307,
   "bf16Tflops": 17.5,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "mac-m5-pro-48",
   "name": "Mac M5 Pro 48 GB · ~36 GB GPU",
   "group": "Unified memory",
   "memoryGB": 36,
   "bandwidthTBs": 0.307,
   "bf16Tflops": 17.5,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "mac-m5-pro-24",
   "name": "Mac M5 Pro 24 GB · ~16 GB GPU",
   "group": "Unified memory",
   "memoryGB": 16,
   "bandwidthTBs": 0.307,
   "bf16Tflops": 17.5,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "mac-m5-32",
   "name": "Mac M5 32 GB · ~21 GB GPU",
   "group": "Unified memory",
   "memoryGB": 21,
   "bandwidthTBs": 0.153,
   "bf16Tflops": 8.75,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "mac-m5-16",
   "name": "Mac M5 16 GB · ~11 GB GPU",
   "group": "Unified memory",
   "memoryGB": 10.7,
   "bandwidthTBs": 0.153,
   "bf16Tflops": 8.75,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "mac-m4-max-128",
   "name": "Mac M4 Max 128 GB · ~96 GB GPU",
   "group": "Unified memory",
   "memoryGB": 96,
   "bandwidthTBs": 0.546,
   "bf16Tflops": 18,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "mac-m4-pro-64",
   "name": "Mac M4 Pro 64 GB · ~48 GB GPU",
   "group": "Unified memory",
   "memoryGB": 48,
   "bandwidthTBs": 0.273,
   "bf16Tflops": 9,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "mac-m3-ultra-512",
   "name": "Mac Studio M3 Ultra 512 GB · ~384 GB GPU",
   "group": "Unified memory",
   "memoryGB": 384,
   "bandwidthTBs": 0.819,
   "bf16Tflops": 28,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "mac-m2-ultra-192",
   "name": "Mac Studio M2 Ultra 192 GB · ~144 GB GPU",
   "group": "Unified memory",
   "memoryGB": 144,
   "bandwidthTBs": 0.8,
   "bf16Tflops": 27,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "mac-m1-max-64",
   "name": "Mac M1 Max 64 GB · ~48 GB GPU",
   "group": "Unified memory",
   "memoryGB": 48,
   "bandwidthTBs": 0.4,
   "bf16Tflops": 10,
   "fp8": false,
   "fp4": false
  },
  {
   "id": "dgx-spark",
   "name": "DGX Spark (GB10) · 128 GB",
   "group": "Unified memory",
   "memoryGB": 128,
   "bandwidthTBs": 0.273,
   "bf16Tflops": 100,
   "fp8": true,
   "fp4": true
  },
  {
   "id": "strix-halo",
   "name": "Ryzen AI Max+ 395 · 96 GB GPU",
   "group": "Unified memory",
   "memoryGB": 96,
   "bandwidthTBs": 0.256,
   "bf16Tflops": 59,
   "fp8": false,
   "fp4": false
  }
 ]
}
