TokenAssemble

Browse · 36 tracked

GPUs for local LLMs

We track 36 GPUs for local LLM inference — 26 NVIDIA, 8 AMD, and 2 Intel — with sourced VRAM, bandwidth, and price, each graded against every model we cover. “Runs up to” is the largest model that loads cleanly (grade B or better) at a quality quant with llama.cpp at 8K context — see how the math works.

NVIDIA

GPUVRAMBandwidthMSRPRuns up to
NVIDIA RTX 6000 Ada Generation48 GB960 GB/s$6,799Qwen3.6-35B-A3B
NVIDIA RTX A600048 GB768 GB/s$4,649Qwen3.6-35B-A3B
NVIDIA GeForce RTX 509032 GB1790 GB/s$1,999Qwen3.6-35B-A3B
NVIDIA RTX 5000 Ada Generation32 GB576 GB/s$4,000Qwen3.6-35B-A3B
NVIDIA GeForce RTX 3090 Ti24 GB1010 GB/s$1,999Qwen3-30B-A3B
NVIDIA GeForce RTX 409024 GB1010 GB/s$1,599Qwen3-30B-A3B
NVIDIA GeForce RTX 309024 GB936.2 GB/s$1,499Qwen3-30B-A3B
NVIDIA GeForce RTX 508016 GB960 GB/s$999gpt-oss-20b
NVIDIA GeForce RTX 5070 Ti16 GB896 GB/s$749gpt-oss-20b
NVIDIA GeForce RTX 4080 SUPER16 GB736.3 GB/s$999gpt-oss-20b
NVIDIA GeForce RTX 408016 GB716.8 GB/s$1,199gpt-oss-20b
NVIDIA GeForce RTX 4070 Ti SUPER16 GB672.3 GB/s$799gpt-oss-20b
NVIDIA GeForce RTX 5060 Ti 16 GB16 GB448 GB/s$429gpt-oss-20b
NVIDIA GeForce RTX 4060 Ti 16 GB16 GB288 GB/s$499gpt-oss-20b
NVIDIA GeForce RTX 3080 12 GB12 GB912.4 GB/s$799Mistral-Nemo-Instruct-2407
NVIDIA GeForce RTX 3080 Ti12 GB912.4 GB/s$1,199Mistral-Nemo-Instruct-2407
NVIDIA GeForce RTX 507012 GB672 GB/s$549Mistral-Nemo-Instruct-2407
NVIDIA GeForce RTX 407012 GB504.2 GB/s$599Mistral-Nemo-Instruct-2407
NVIDIA GeForce RTX 4070 SUPER12 GB504.2 GB/s$599Mistral-Nemo-Instruct-2407
NVIDIA GeForce RTX 4070 Ti12 GB504.2 GB/s$799Mistral-Nemo-Instruct-2407
NVIDIA GeForce RTX 3060 12 GB12 GB360 GB/s$329Mistral-Nemo-Instruct-2407
NVIDIA GeForce RTX 3080 10 GB10 GB760.3 GB/s$699Qwen3.5-9B
NVIDIA GeForce RTX 3070 Ti8 GB608.3 GB/s$599Qwen3-8B
NVIDIA GeForce RTX 30708 GB448 GB/s$499Qwen3-8B
NVIDIA GeForce RTX 4060 Ti 8 GB8 GB288 GB/s$399Qwen3-8B
NVIDIA GeForce RTX 40608 GB272 GB/s$299Qwen3-8B

AMDbeta tiers

GPUVRAMBandwidthMSRPRuns up to
AMD Radeon PRO W790048 GB864 GB/s$3,999Qwen3.6-35B-A3B
AMD Radeon RX 7900 XTX24 GB960 GB/s$999Qwen3-30B-A3B
AMD Radeon RX 7900 XT20 GB800 GB/s$899Mistral-Small-24B-Instruct-2501
AMD Radeon RX 907016 GB640 GB/s$549gpt-oss-20b
AMD Radeon RX 9070 XT16 GB640 GB/s$599gpt-oss-20b
AMD Radeon RX 7800 XT16 GB624 GB/s$499gpt-oss-20b
AMD Radeon RX 680016 GB512 GB/s$579gpt-oss-20b
AMD Radeon RX 6900 XT16 GB512 GB/s$999gpt-oss-20b

Intelbeta tiers

GPUVRAMBandwidthMSRPRuns up to
Intel Arc A770 16 GB16 GB560 GB/s$349gpt-oss-20b
Intel Arc B58012 GB456 GB/s$249Mistral-Nemo-Instruct-2407

AMD and Intel fit verdicts are physics and hold; their speed tiers are beta until real measurements calibrate them. Specs stamped and sourced on each card's page.