Skip to main content
29 GPUs ranked Table built CSV (CC BY 4.0)

GPU Value for Local LLMs: tok/s per $1,000 (October 2026)

Of 29 GPUs with a cited 7-9B Q4 median and an Amazon price checked in the last 24 hours, the NVIDIA GeForce RTX 2060 SUPER gives the most local-LLM speed per dollar: 61.1 tok/s (median of 7 cited runs, per KnightLi llama.cpp GPU Benchmark Scoreboard, Ian L. Paterson blog, llama.cpp GitHub) at $299.97 checked 2026-10-20 16:14 UTC, or 203.8 tok/s per $1,000.

By SpecPicks Editorial · recomputed every 30 minutes from the latest price reads

As an Amazon Associate, SpecPicks earns from qualifying purchases.

The table: tok/s per $1,000, highest first

Speed is the median of the cited single-user runs on 7-9B dense models at 4-bit (Q4) under one runtime, the same figure each card’s benchmark page leads with. Price is the LLM GPU Price Index figure for every card the index tracks (the second-cheapest matching Amazon listing, so the two tables print the same price on the same day), and for other cards the cheapest new Amazon listing the card’s benchmark page prices. Either way the listing was read in the last 24 hours and sits between 0.5× and 1.6× the card’s launch MSRP, the index’s rule: above that it is a marketplace markup, not the card’s price. The other 77 GPUs with 7-9B runs on file lack a median (3+ comparable runs from 2+ publishers), a VRAM figure on file, a launch MSRP on file or a price that fresh inside that band, so they are left out rather than estimated. A card catalogued under two names is ranked once. Sort the CSV by any column.

#GPUVRAMMedian tok/s (7-9B Q4)Cited runsNew price (Amazon)tok/s per $1,000Buy
1 NVIDIA GeForce RTX 2060 SUPER 8 GB 61.1 llama.cpp 7 runs, 4 publishers
KnightLi llama.cpp GPU Benchmark Scoreboard, Ian L. Paterson blog		 $299.97
checked 		203.8 View Current Price
2 AMD Radeon RX 6800 16 GB 89.8 llama.cpp 4 runs, 2 publishers
llama.cpp GitHub, llama.cpp GitHub Discussion #10879		 $489.97
checked 		183.2 View Current Price
3 AMD Radeon RX 9070 16 GB 99.4 llama.cpp 5 runs, 4 publishers
KnightLi GPU Benchmark Scoreboard, llama.cpp GitHub Discussion		 $699.99
checked 		142.0 View Current Price
4 NVIDIA GeForce RTX 2070 8 GB 64.6 llama.cpp 5 runs, 4 publishers
knightli.com, llama.cpp GitHub Discussion #15013		 $469.00
checked 		137.8 View Current Price
5 NVIDIA GeForce RTX 5060 Ti 16GB 16 GB 71.0 ollama 5 runs, 4 publishers
RunAIHome, ComputingForGeeks		 $519.00
checked 		136.8 View Current Price
6 NVIDIA GeForce RTX 2080 Ti 11 GB 108 llama.cpp 5 runs, 4 publishers
GigaChad LLC, KnightLI		 $798.00
checked 		134.7 View Current Price
7 AMD Radeon RX 7900 XT 20 GB 116 llama.cpp 4 runs, 3 publishers
llama.cpp GitHub Discussion #15021, llama.cpp GitHub		 $869.99
checked 		133.5 View Current Price
8 NVIDIA GeForce RTX 3060 12 GB 12 GB 55.2 llama.cpp 21 runs, 9 publishers
llama.cpp GitHub Discussion #10879, llama.cpp GitHub		 $419.99
checked 		131.4 View Current Price
9 NVIDIA GeForce RTX 3080 10 GB 98.7 llama.cpp 11 runs, 6 publishers
llama.cpp GitHub Discussion #15013, knightli.com		 $759.00
checked 		130.0 View Current Price
10 AMD Radeon RX 9060 XT 8GB 8 GB 58.9 llama.cpp 6 runs, 5 publishers
llama.cpp GitHub, RunAIHome		 $459.99
checked 		128.0 View Current Price
11 AMD Radeon RX 9070 XT 16 GB 109 llama.cpp 6 runs, 4 publishers
llama.cpp GitHub, RunAIHome		 $869.99
checked 		125.1 View Current Price
12 NVIDIA GeForce RTX 3070 8 GB 59.6 llama.cpp 11 runs, 5 publishers
llama.cpp GitHub Discussions, GPU-Benchmarks-on-LLM-Inference		 $480.72
checked 		124.0 View Current Price
13 NVIDIA GeForce RTX 5060 Ti 8GB 8 GB 59.0 ollama 8 runs, 6 publishers
RunAIHome, computingforgeeks.com		 $491.99
checked 		120.0 View Current Price
14 AMD Radeon RX 6800 XT 16 GB 89.8 llama.cpp 6 runs, 4 publishers
KnightLi llama.cpp GPU Benchmark Scoreboard, llama.cpp GitHub Discussion #10879		 $798.00
checked 		112.5 View Current Price
15 NVIDIA GeForce RTX 3050 8 GB 28.4 ollama 6 runs, 2 publishers
LocalScore.ai, llmrun.dev		 $259.97
checked 		109.1 View Current Price
16 NVIDIA GeForce RTX 3070 Ti 8 GB 80.6 llama.cpp 6 runs, 4 publishers
GPU Battle, LocalScore.ai		 $740.00
checked 		108.9 View Current Price
17 NVIDIA GeForce RTX 5070 12 GB 93.3 llama.cpp 4 runs, 4 publishers
llama.cpp GitHub, knightli.com		 $856.99
checked 		108.9 View Current Price
18 NVIDIA GeForce RTX 3080 Ti 12 GB 89.7 llama.cpp 7 runs, 4 publishers
KnightLi llama.cpp GPU Scoreboard, Hardware Corner		 $899.99
checked 		99.7 View Current Price
19 NVIDIA GeForce GTX 1660 SUPER 6 GB 15.7 llama.cpp 3 runs, 2 publishers
llama.cpp GitHub, LocalScore.ai		 $194.99
checked 		80.5 View Current Price
20 NVIDIA GeForce RTX 4070 SUPER 12 GB 65.6 llama.cpp 8 runs, 5 publishers
llama.cpp GitHub Discussions, MyAIHardware		 $849.97
checked 		77.2 View Current Price
21 AMD Radeon RX 6600 XT 8 GB 34.0 llama.cpp 5 runs, 4 publishers
knightli.com, llama.cpp GitHub		 $499.00
checked 		68.1 View Current Price
22 NVIDIA GeForce RTX 4080 16 GB 102 llama.cpp 9 runs, 7 publishers
knightli.com, llama.cpp GitHub		 $1501.95
checked 		67.9 View Current Price
23 NVIDIA GeForce RTX 4070 Ti 12 GB 78.0 llama.cpp 9 runs, 4 publishers
Markaicode, GPU-Benchmarks-on-LLM-Inference		 $1149.00
checked 		67.9 View Current Price
24 NVIDIA GeForce RTX 4070 12 GB 52.1 llama.cpp 7 runs, 5 publishers
knightli.com llama.cpp GPU Benchmark Scoreboard, Local LLM Benchmarks		 $849.00
checked 		61.4 View Current Price
25 NVIDIA GeForce RTX 3090 24 GB 93.8 llama.cpp 6 runs, 5 publishers
Hardware Corner, GitHub: XiongjieDai/GPU-Benchmarks-on-LLM-Inference		 $1729.99
checked 		54.2 View Current Price
26 NVIDIA RTX A4000 16GB 16 GB 52.7 ollama 10 runs, 3 publishers
Trooper.AI GPU benchmarks, DatabaseMart		 $1044.99
checked 		50.5 View Current Price
27 NVIDIA GeForce RTX 3090 Ti 24 GB 110 llama.cpp 7 runs, 5 publishers
llama.cpp GitHub Discussions, Hardware Corner		 $2189.97
checked 		50.2 View Current Price
28 NVIDIA RTX A5000 24GB 24 GB 136 llama.cpp 3 runs, 2 publishers
KnightLi llama.cpp GPU Benchmark Scoreboard, llama.cpp GitHub		 $3149.00
checked 		43.1 View Current Price
29 AMD Radeon Pro W7900 48GB 48 GB 108 llama.cpp 4 runs, 2 publishers
llama.cpp GitHub, llama.cpp GitHub Discussions		 $3839.00
checked 		28.1 View Current Price

*Prices are tracked snapshots of Amazon listings, each read at the time printed under it, and are not live quotes: price may vary.

How to read it

Tok/s per $1,000 rewards cheap speed on small models. It says nothing about capacity: a 24 GB card that ranks below a 16 GB card still runs models the 16 GB card cannot hold. Check model-by-model fit on Will it run?, and price per GB of VRAM, new and used, on the LLM GPU Price Index. Runtimes differ by card (llama.cpp, Ollama, LM Studio), so treat close ranks as ties. Each row links the benchmark page that lists every run behind its median.

This table is editorial synthesis of publicly reported benchmarks and tracked retail prices. No first-party benchmarking is reported.