Local LLM GPU List
57 GPUs for local LLMs tracked as of Oct 21, 2026, with VRAM, bandwidth and compute specs plus prices from eBay, Xianyu and Amazon.jp.
| Model | Brand | Type | Memory type | Ecosystem | ||||||
|---|---|---|---|---|---|---|---|---|---|---|
 H100 PCIe | NVIDIA | Data center | 80 | HBM2e | 2,000 | 756 | 1,513 | 350 | CUDA | Mar 22, 2022 |
| RTX PRO 6000 | NVIDIA | Workstation | 96 | GDDR7 ECC | 1,792 | 500 | 1,000 | 600 | CUDA | Mar 18, 2025 |
| GeForce RTX 5090 | NVIDIA | Consumer | 32 | GDDR7 | 1,792 | 419 | 838 | 575 | CUDA | Jan 30, 2025 |
| RTX PRO 6000D | NVIDIA | Data center | 84 | GDDR7 ECC | 1,398 | 385.2 | 770.5 | 600 | CUDA | Sep 1, 2025 |
| L40S | NVIDIA | Data center | 48 | GDDR6 | 864 | 362.1 | 733 | 350 | CUDA | Aug 8, 2023 |
| RTX 6000 Ada | NVIDIA | Workstation | 48 | GDDR6 ECC | 960 | 364.3 | 728.5 | 300 | CUDA | Sep 20, 2022 |
| GeForce RTX 4090 | NVIDIA | Consumer | 24 | GDDR6X | 1,008 | 330.3 | 660.6 | 450 | CUDA | Oct 12, 2022 |
| GeForce RTX 4090 48GB | NVIDIA | Consumer | 48 | GDDR6X | 1,008 | 330.3 | 660.6 | 450 | CUDA | Oct 12, 2022 |
| A100 40GB PCIe | NVIDIA | Data center | 40 | HBM2 | 1,555 | 312 | 624 | 250 | CUDA | Jun 22, 2020 |
| A100 80GB PCIe | NVIDIA | Data center | 80 | HBM2e | 1,935 | 312 | 624 | 300 | CUDA | Jun 28, 2021 |
| GeForce RTX 5090 D | NVIDIA | Consumer | 32 | GDDR7 | 1,792 | 296.9 | 593.8 | 575 | CUDA | Jan 30, 2025 |
| GeForce RTX 5090 D V2 | NVIDIA | Consumer | 24 | GDDR7 | 1,344 | 296.9 | 593.8 | 575 | CUDA | Aug 12, 2025 |
| GeForce RTX 4090D 48GB | NVIDIA | Consumer | 48 | GDDR6X | 1,008 | 294.3 | 588.5 | 425 | CUDA | Dec 28, 2023 |
| RTX PRO 5000 48GB | NVIDIA | Workstation | 48 | GDDR7 ECC | 1,344 | 258 | 516 | 300 | CUDA | Mar 18, 2025 |
| RTX PRO 5000 72GB | NVIDIA | Workstation | 72 | GDDR7 ECC | 1,344 | 258 | 516 | 300 | CUDA | Dec 18, 2025 |
| GeForce RTX 5080 | NVIDIA | Consumer | 16 | GDDR7 | 960 | 225.1 | 450.2 | 360 | CUDA | Jan 30, 2025 |
| RTX PRO 4500 | NVIDIA | Workstation | 32 | GDDR7 ECC | 896 | 210.9 | 421.8 | 200 | CUDA | Mar 18, 2025 |
| GeForce RTX 4080 SUPER 32GB | NVIDIA | Consumer | 32 | GDDR6X | 736 | 208.9 | 417.8 | 320 | CUDA | Jan 31, 2024 |
| CMP 170HX 40GB | NVIDIA | Data center | 40 | HBM2e | 1,560 | 202 | 404 | 250 | CUDA | Sep 1, 2021 |
| CMP 170HX 64GB | NVIDIA | Data center | 64 | HBM2e | 1,493 | 202 | 404 | 250 | CUDA | Jun 1, 2021 |
| GeForce RTX 4080 | NVIDIA | Consumer | 16 | GDDR6X | 717 | 195 | 389.9 | 320 | CUDA | Nov 16, 2022 |
| Radeon RX 9070 XT | AMD | Consumer | 16 | GDDR6 | 640 | 195 | 389 | 304 | ROCm | Mar 6, 2025 |
| Radeon AI PRO R9700 | AMD | Workstation | 32 | GDDR6 | 640 | 191 | 383 | 300 | ROCm | Jul 23, 2025 |
 Arc Pro B70 | Intel | Workstation | 32 | GDDR6 | 608 | 183.5 | 367 | 230 | oneAPI | Mar 25, 2026 |
| GeForce RTX 5070 Ti | NVIDIA | Consumer | 16 | GDDR7 | 896 | 175.8 | 351.5 | 300 | CUDA | Feb 20, 2025 |
| RTX PRO 4000 | NVIDIA | Workstation | 24 | GDDR7 ECC | 672 | 161.3 | 322.5 | 145 | CUDA | Mar 18, 2025 |
| RTX A6000 | NVIDIA | Workstation | 48 | GDDR6 ECC | 768 | 154.9 | 309.7 | 300 | CUDA | Oct 5, 2020 |
| GeForce RTX 3090 | NVIDIA | Consumer | 24 | GDDR6X | 936 | 142.3 | 284.7 | 350 | CUDA | Sep 24, 2020 |
| DGX Spark 128GB | NVIDIA | Unified memory | 128 | LPDDR5X · unified memory | 273 | 125 | 250 | 240 | CUDA | Oct 15, 2025 |
| GeForce RTX 5070 | NVIDIA | Consumer | 12 | GDDR7 | 672 | 123.5 | 247 | 250 | CUDA | Mar 5, 2025 |
| GeForce RTX 2080 Ti 22GB | NVIDIA | Consumer | 22 | GDDR6 | 616 | 107.6 | 215.2 | 250 | CUDA | Sep 20, 2018 |
| Arc Pro B60 | Intel | Workstation | 24 | GDDR6 | 456 | 98.5 | 197 | 200 | oneAPI | Aug 1, 2025 |
| Arc Pro B65 | Intel | Workstation | 32 | GDDR6 | 608 | 98.5 | 197 | 200 | oneAPI | Mar 25, 2026 |
| GeForce RTX 5060 Ti 16GB | NVIDIA | Consumer | 16 | GDDR7 | 448 | 94.9 | 189.8 | 180 | CUDA | Apr 16, 2025 |
| Instinct MI210 | AMD | Data center | 64 | HBM2e | 1,638 | 181 | 181 | 300 | ROCm | Mar 22, 2022 |
| GeForce RTX 4060 Ti 16GB | NVIDIA | Consumer | 16 | GDDR6 | 288 | 88.3 | 176.5 | 165 | CUDA | Jul 18, 2023 |
| Tesla T10 | NVIDIA | Data center | 16 | GDDR6 | 403 | 80 | 160 | 150 | CUDA | Feb 4, 2020 |
| Radeon PRO W7900 | AMD | Workstation | 48 | GDDR6 ECC | 864 | 123 | 123 | 295 | ROCm | Apr 13, 2023 |
| Radeon RX 7900 XTX | AMD | Consumer | 24 | GDDR6 | 960 | 123 | 123 | 355 | ROCm | Dec 13, 2022 |
| Radeon RX 7900 XT | AMD | Consumer | 20 | GDDR6 | 800 | 103 | 103 | 315 | ROCm | Dec 13, 2022 |
| GeForce RTX 3060 12GB | NVIDIA | Consumer | 12 | GDDR6 | 360 | 51 | 101.9 | 170 | CUDA | Feb 25, 2021 |
| Radeon RX 6950 XT | AMD | Consumer | 16 | GDDR6 | 576 | 47.3 | 94.6 | 335 | ROCm | May 10, 2022 |
| Instinct MI100 | AMD | Data center | 32 | HBM2 | 1,229 | 92.3 | 92.3 | 300 | ROCm | Nov 16, 2020 |
| Radeon PRO W7800 | AMD | Workstation | 32 | GDDR6 ECC | 576 | 90.4 | 90.4 | 260 | ROCm | Apr 13, 2023 |
| Radeon RX 7800 XT | AMD | Consumer | 16 | GDDR6 | 624 | 74.7 | 74.7 | 263 | ROCm | Sep 6, 2023 |
| Ryzen AI Max+ 395 128GB | AMD | Unified memory | 128 | LPDDR5X · unified memory | 256 | 59.4 | 59.4 | 120 | ROCm | Mar 1, 2025 |
| Tesla V100 16GB | NVIDIA | Data center | 16 | HBM2 | 900 | 112 | 56.1 | 250 | CUDA | Jun 21, 2017 |
| Tesla V100 32GB | NVIDIA | Data center | 32 | HBM2 | 900 | 112 | 56.1 | 250 | CUDA | Mar 27, 2018 |
| Instinct MI50 32GB | AMD | Data center | 32 | HBM2 | 1,024 | 26.5 | 53 | 300 | ROCm | Nov 18, 2018 |
| Tesla P40 | NVIDIA | Data center | 24 | GDDR5 | 347 | — | 47 | 250 | CUDA | Sep 13, 2016 |
| Mac Studio M3 Ultra 512GB | Apple | Unified memory | 512 | LPDDR5-6400 · unified memory | 819 | — | — | 480 | Metal | Mar 12, 2025 |
| Mac Studio M4 Max 128GB | Apple | Unified memory | 128 | LPDDR5X-8533 · unified memory | 546 | — | — | 480 | Metal | Mar 12, 2025 |
| Mac Studio M5 Max 128GB | Apple | Unified memory | 128 | LPDDR5X-9600 · unified memory | 614 | — | — | 480 | Metal | Aug 25, 2026 |
| Mac Studio M5 Ultra 96GB | Apple | Unified memory | 96 | LPDDR5X-9600 · unified memory | 1,229 | — | — | 480 | Metal | Aug 25, 2026 |
| MacBook Pro M4 Max 128GB | Apple | Unified memory | 128 | LPDDR5X-8533 · unified memory | 546 | — | — | 140 | Metal | Oct 30, 2024 |
| MacBook Pro M5 Max 128GB | Apple | Unified memory | 128 | LPDDR5X-9600 · unified memory | 614 | — | — | 140 | Metal | Mar 4, 2026 |
| RTX PRO 5500 | NVIDIA | Workstation | 84 | GDDR7 ECC | 1,398 | — | — | 600 | CUDA | Sep 14, 2026 |
No GPUs match “”.