Best GPUs for Local AI — Full Comparison

Every GPU reviewed for local LLM inference, ranked by VRAM and tokens/sec speed on Llama 3.1 8B Q4_K_M — now with the memory bandwidth that figure is derived from.

Specs verified 2026-07-21 · Prices verified 2026-07-06

DeviceMemoryBandwidthLlama 3.1 8B Q4_K_MPrice$/GBTDPReleasedRuns with
Apple M3 Ultra512 GB unified819 GB/s43–90 tok/s$3,999$7.8160 W2025metal
Apple M2 Ultra192 GB unified800 GB/s43–89 tok/s$3,999$20.8360 W2023metal
Apple M1 Ultra128 GB unified800 GB/s43–89 tok/s$3,999$31.2460 W2022metal
Apple M4 Max128 GB unified546 GB/s31–64 tok/s$3,499$27.3435 W2024metal
Apple M5 Max128 GB unified614 GB/s34–71 tok/s$3,499$27.3435 W2026metal
Apple M2 Max96 GB unified400 GB/s24–49 tok/s$3,499$36.4540 W2023metal
NVIDIA RTX PRO 6000 Blackwell96 GB1,792 GB/s113–234 tok/s$8,565$89.22600 W2025cuda, vulkan
NVIDIA A100 80GB80 GB1,935 GB/s102–196 tok/s$15,000$187.50300 W2021cuda, vulkan
NVIDIA H100 80GB80 GB2,000 GB/s120–250 tok/s$25,000$312.50350 W2022cuda, vulkan
Apple M1 Max64 GB unified400 GB/s24–49 tok/s$3,499$54.6730 W2021metal
Apple M3 Max64 GB unified400 GB/s24–49 tok/s$2,499$39.0535 W2023metal
Apple M5 Pro64 GB unified307 GB/s19–39 tok/s$1,999$31.2330 W2026metal
NVIDIA L40S48 GB864 GB/s77–147 tok/s$7,499$156.23350 W2023cuda, vulkan
NVIDIA RTX 6000 Ada Generation48 GB960 GB/s83–159 tok/s$6,799$141.65300 W2022cuda, vulkan
Apple M3 Pro36 GB unified153 GB/s9.7–20 tok/s$1,999$55.5330 W2023metal
Apple M1 Pro32 GB unified200 GB/s13–26 tok/s$1,999$62.4730 W2021metal
Apple M2 Pro32 GB unified200 GB/s13–26 tok/s$1,999$62.4730 W2023metal
Apple M432 GB unified120 GB/s7.7–16 tok/s$999$31.2222 W2024metal
Apple M532 GB unified153 GB/s9.7–20 tok/s$1,599$49.9723 W2025metal
NVIDIA GeForce RTX 509032 GB1,792 GB/s113–234 tok/s$1,999$62.47575 W2025cuda, vulkan
Apple M224 GB unified100 GB/s6.4–13 tok/s$1,099$45.7920 W2022metal
Apple M324 GB unified100 GB/s6.4–13 tok/s$1,099$45.7922 W2023metal
Apple M4 Pro24 GB unified273 GB/s17–35 tok/s$1,999$83.2930 W2024metal
NVIDIA GeForce RTX 309024 GB936 GB/s60–115 tok/s$1,499$62.46350 W2020cuda, vulkan
NVIDIA GeForce RTX 409024 GB1,008 GB/s86–165 tok/s$1,599$66.63450 W2022cuda, vulkan
AMD Radeon RX 7900 XTX24 GB960 GB/s54–112 tok/s$999$41.63355 W2022rocm, vulkan
AMD Radeon RX 7900 XT20 GB800 GB/s47–98 tok/s$899$44.95315 W2022rocm, vulkan
Apple M116 GB unified68 GB/s4.4–9.2 tok/s$999$62.4420 W2020metal
NVIDIA GeForce RTX 4060 Ti 16GB16 GB288 GB/s31–59 tok/s$499$31.19165 W2023cuda, vulkan
NVIDIA GeForce RTX 4070 Ti Super16 GB672 GB/s63–121 tok/s$799$49.94285 W2024cuda, vulkan
NVIDIA GeForce RTX 408016 GB717 GB/s66–127 tok/s$1,199$74.94320 W2022cuda, vulkan
NVIDIA GeForce RTX 4080 Super16 GB736 GB/s68–130 tok/s$999$62.44320 W2024cuda, vulkan
NVIDIA GeForce RTX 5060 Ti 16GB16 GB448 GB/s40–83 tok/s$429$26.81180 W2025cuda, vulkan
NVIDIA GeForce RTX 5070 Ti16 GB896 GB/s70–145 tok/s$749$46.81300 W2025cuda, vulkan
NVIDIA GeForce RTX 508016 GB960 GB/s74–153 tok/s$999$62.44360 W2025cuda, vulkan
AMD Radeon RX 7800 XT16 GB624 GB/s39–81 tok/s$499$31.19263 W2023rocm, vulkan
AMD Radeon RX 9060 XT 16GB16 GB320 GB/s24–49 tok/s$349$21.81160 W2025rocm, vulkan
AMD Radeon RX 907016 GB640 GB/s42–87 tok/s$549$34.31220 W2025rocm, vulkan
AMD Radeon RX 9070 XT16 GB717 GB/s46–95 tok/s$599$37.44220 W2025rocm, vulkan
Intel Arc B58012 GB456 GB/s28–57 tok/s$249$20.75190 W2024sycl, openvino, vulkan
NVIDIA GeForce RTX 3060 (12GB)12 GB360 GB/s26–50 tok/s$329$27.42170 W2021cuda, vulkan
NVIDIA GeForce RTX 407012 GB504 GB/s50–96 tok/s$599$49.92200 W2023cuda, vulkan
NVIDIA GeForce RTX 4070 Super12 GB504 GB/s50–96 tok/s$599$49.92220 W2024cuda, vulkan
NVIDIA GeForce RTX 4070 Ti12 GB504 GB/s50–96 tok/s$799$66.58285 W2023cuda, vulkan
NVIDIA GeForce RTX 507012 GB672 GB/s56–116 tok/s$549$45.75250 W2025cuda, vulkan
Intel Arc B57010 GB380 GB/s24–49 tok/s$219$21.90150 W2025sycl, openvino, vulkan
NVIDIA GeForce RTX 3080 (10GB)10 GB760 GB/s50–97 tok/s$699$69.90320 W2020cuda, vulkan
NVIDIA GeForce RTX 30708 GB448 GB/s32–61 tok/s$499$62.38220 W2020cuda, vulkan
NVIDIA GeForce RTX 3070 Ti8 GB608 GB/s42–80 tok/s$599$74.88290 W2021cuda, vulkan
NVIDIA GeForce RTX 40608 GB272 GB/s29–56 tok/s$299$37.38115 W2023cuda, vulkan
NVIDIA GeForce RTX 50608 GB448 GB/s40–83 tok/s$299$37.38145 W2025cuda, vulkan
NVIDIA GeForce RTX 5060 Ti 8GB8 GB448 GB/s40–83 tok/s$379$47.38180 W2025cuda, vulkan
AMD Radeon RX 9060 XT 8GB8 GB320 GB/s24–49 tok/s$299$37.38150 W2025rocm, vulkan

Quick Buying Guide

Under $300RTX 3060 12GB — 12 GB VRAM, runs all 7–8B models
Under $500RTX 4060 Ti 16GB — best budget 16 GB card
$500–700RTX 4070 Super or AMD RX 9070 XT
$1,000–1,600RTX 4090 24GB — runs 70B models, fastest single card
Mac laptopM3 Pro 36GB or M4 Pro — silent, power-efficient
Mac desktopM4 Max 128GB — runs any quantized model

Looking wider? Every device we track · Laptops · AI stations and mini-PCs