24h drops
CMP 170HX 40GB$2,499-14.8%- A100 80GB PCIe$18,999-13.1%
- Radeon PRO W7800$1,799-10.0%
Arc Pro B70$2,027-5.9%- GeForce RTX 4080$1,300-3.7%
24h rises
- MacBook Pro M4 Max 128GB$5,999+10.1%
- Mac Studio M4 Max 128GB$6,000+7.2%
- CMP 170HX 64GB$4,000+6.7%
- Radeon RX 9070 XT$1,150+4.5%
- GeForce RTX 5080$2,449+3.4%
Most listings
- GeForce RTX 5090$7,500127 listings
- GeForce RTX 3060 12GB$449116 listings
- GeForce RTX 3090$1,850113 listings
- GeForce RTX 5080$2,449104 listings
- GeForce RTX 5070$90084 listings
Latest release
Kimi K3.1 Pro2,800BQ41,508.7 GB
Mistral Large 41,000BQ4604.8 GB
DeepSeek V4.1 Flash763BQ4444.7 GB
Muse Spark Open816BQ4505.5 GB
Hy4 preview770BQ4467.3 GB
Local LLM GPU Ranking
eBay
| Model | 30d trend | Ecosystem | |||||||
|---|---|---|---|---|---|---|---|---|---|
| H100 PCIe | No price in this market | — | 80 | 2,000 | 756 | 1,513 | 350 | CUDA | |
| RTX PRO 6000 | $20,00020 listings | 0.0% | 96 | 1,792 | 500 | 1,000 | 600 | CUDA | |
| GeForce RTX 5090 | $7,500127 listings | 0.0% | 32 | 1,792 | 419 | 838 | 575 | CUDA | |
| RTX PRO 6000D | No price in this market | — | 84 | 1,398 | 385.2 | 770.5 | 600 | CUDA | |
| L40S | $8,99917 listings | 0.0% | 48 | 864 | 362.1 | 733 | 350 | CUDA | |
| RTX 6000 Ada | $7,8396 listings | -1.9% | 48 | 960 | 364.3 | 728.5 | 300 | CUDA | |
| GeForce RTX 4090 | $3,97366 listings | -0.7% | 24 | 1,008 | 330.3 | 660.6 | 450 | CUDA | |
| GeForce RTX 4090 48GB | No price in this market | — | 48 | 1,008 | 330.3 | 660.6 | 450 | CUDA | |
| A100 40GB PCIe | $7,00017 listings | 0.0% | 40 | 1,555 | 312 | 624 | 250 | CUDA | |
| A100 80GB PCIe | $18,99911 listings | -13.1% | 80 | 1,935 | 312 | 624 | 300 | CUDA | |
| GeForce RTX 5090 D | No price in this market | — | 32 | 1,792 | 296.9 | 593.8 | 575 | CUDA | |
| GeForce RTX 5090 D V2 | No price in this market | — | 24 | 1,344 | 296.9 | 593.8 | 575 | CUDA | |
| GeForce RTX 4090D 48GB | No price in this market | — | 48 | 1,008 | 294.3 | 588.5 | 425 | CUDA | |
| RTX PRO 5000 48GB | $7,64916 listings | +0.6% | 48 | 1,344 | 258 | 516 | 300 | CUDA | |
| RTX PRO 5000 72GB | $10,2503 listings | 0.0% | 72 | 1,344 | 258 | 516 | 300 | CUDA | |
| GeForce RTX 5080 | $2,449104 listings | +3.4% | 16 | 960 | 225.1 | 450.2 | 360 | CUDA | |
| RTX PRO 4500 | $4,80017 listings | 0.0% | 32 | 896 | 210.9 | 421.8 | 200 | CUDA | |
| GeForce RTX 4080 SUPER 32GB | No price in this market | — | 32 | 736 | 208.9 | 417.8 | 320 | CUDA | |
| CMP 170HX 40GB | $2,4993 listings | -14.8% | 40 | 1,560 | 202 | 404 | 250 | CUDA | |
| CMP 170HX 64GB | $4,0007 listings | +6.7% | 64 | 1,493 | 202 | 404 | 250 | CUDA | |
| GeForce RTX 4080 | $1,30049 listings | -3.7% | 16 | 717 | 195 | 389.9 | 320 | CUDA | |
| Radeon RX 9070 XT | $1,15059 listings | +4.5% | 16 | 640 | 195 | 389 | 304 | ROCm | |
| Radeon AI PRO R9700 | $1,9503 listings | -3.5% | 32 | 640 | 191 | 383 | 300 | ROCm | |
| Arc Pro B70 | $2,0278 listings | -5.9% | 32 | 608 | 183.5 | 367 | 230 | oneAPI | |
| GeForce RTX 5070 Ti | $1,49939 listings | 0.0% | 16 | 896 | 175.8 | 351.5 | 300 | CUDA | |
| RTX PRO 4000 | $2,99710 listings | 0.0% | 24 | 672 | 161.3 | 322.5 | 145 | CUDA | |
| RTX A6000 | $5,0878 listings | +1.8% | 48 | 768 | 154.9 | 309.7 | 300 | CUDA | |
| GeForce RTX 3090 | $1,850113 listings | +2.8% | 24 | 936 | 142.3 | 284.7 | 350 | CUDA | |
| DGX Spark 128GB | $8,00022 listings | +0.6% | 128 | 273 | 125 | 250 | 240 | CUDA | |
| GeForce RTX 5070 | $90084 listings | 0.0% | 12 | 672 | 123.5 | 247 | 250 | CUDA | |
| GeForce RTX 2080 Ti 22GB | $5513 listings | — | 22 | 616 | 107.6 | 215.2 | 250 | CUDA | |
| Arc Pro B60 | $1,0736 listings | -2.5% | 24 | 456 | 98.5 | 197 | 200 | oneAPI | |
| Arc Pro B65 | No price in this market | — | 32 | 608 | 98.5 | 197 | 200 | oneAPI | |
| GeForce RTX 5060 Ti 16GB | $85054 listings | +0.7% | 16 | 448 | 94.9 | 189.8 | 180 | CUDA | |
| Instinct MI210 | $5,04411 listings | 0.0% | 64 | 1,638 | 181 | 181 | 300 | ROCm | |
| GeForce RTX 4060 Ti 16GB | $72512 listings | -0.7% | 16 | 288 | 88.3 | 176.5 | 165 | CUDA | |
| Tesla T10 | $4955 listings | 0.0% | 16 | 403 | 80 | 160 | 150 | CUDA | |
| Radeon PRO W7900 | $3,5993 listings | 0.0% | 48 | 864 | 123 | 123 | 295 | ROCm | |
| Radeon RX 7900 XTX | $1,39922 listings | -0.1% | 24 | 960 | 123 | 123 | 355 | ROCm | |
| Radeon RX 7900 XT | $87019 listings | +2.4% | 20 | 800 | 103 | 103 | 315 | ROCm | |
| GeForce RTX 3060 12GB | $449116 listings | -0.2% | 12 | 360 | 51 | 101.9 | 170 | CUDA | |
| Radeon RX 6950 XT | $69016 listings | +1.5% | 16 | 576 | 47.3 | 94.6 | 335 | ROCm | |
| Instinct MI100 | $1,5037 listings | +0.2% | 32 | 1,229 | 92.3 | 92.3 | 300 | ROCm | |
| Radeon PRO W7800 | $1,7993 listings | -10.0% | 32 | 576 | 90.4 | 90.4 | 260 | ROCm | |
| Radeon RX 7800 XT | $65026 listings | 0.0% | 16 | 624 | 74.7 | 74.7 | 263 | ROCm | |
| Ryzen AI Max+ 395 128GB | $3,87522 listings | -0.6% | 128 | 256 | 59.4 | 59.4 | 120 | ROCm | |
| Tesla V100 16GB | $41023 listings | -2.4% | 16 | 900 | 112 | 56.1 | 250 | CUDA | |
| Tesla V100 32GB | $1,02112 listings | 0.0% | 32 | 900 | 112 | 56.1 | 250 | CUDA | |
| Instinct MI50 32GB | $7344 listings | 0.0% | 32 | 1,024 | 26.5 | 53 | 300 | ROCm | |
| Tesla P40 | $39016 listings | +1.6% | 24 | 347 | — | 47 | 250 | CUDA | |
| Mac Studio M3 Ultra 512GB | $22,99913 listings | 0.0% | 512 | 819 | — | — | 480 | Metal | |
| Mac Studio M4 Max 128GB | $6,0007 listings | +7.2% | 128 | 546 | — | — | 480 | Metal | |
| Mac Studio M5 Max 128GB | $7,1174 listings | +2.4% | 128 | 614 | — | — | 480 | Metal | |
| Mac Studio M5 Ultra 96GB | $7,4007 listings | 0.0% | 96 | 1,229 | — | — | 480 | Metal | |
| MacBook Pro M4 Max 128GB | $5,99916 listings | +10.1% | 128 | 546 | — | — | 140 | Metal | |
| MacBook Pro M5 Max 128GB | $7,20027 listings | +2.9% | 128 | 614 | — | — | 140 | Metal | |
| RTX PRO 5500 | No price in this market | — | 84 | 1,398 | — | — | 600 | CUDA |
No GPUs match “”.
Log in to add GPUs to your watchlist and see it on any device. Log in
No GPUs in your watchlist yet. Click the ☆ before a model name to add one.
| Spec | Price per stick | Price per GB | Max channels | Max bandwidth | Max memory |
|---|---|---|---|---|---|
| 16 GB | $225130 listings | $14.1 | 2 | 96 GB/s | 64 GB4 sticks × 16 GB |
| 32 GB | $49991 listings | $15.6 | 2 | 96 GB/s | 128 GB4 sticks × 32 GB |
| 32 GB | $1,38834 listings | $43.4 | 24 | 1,075.2 GB/s | 768 GB24 sticks × 32 GB |
| 64 GB | $2,66717 listings | $41.7 | 24 | 1,075.2 GB/s | 1,536 GB24 sticks × 64 GB |
How to choose a GPU for local LLMs
Find the right GPU for local LLMs. Compare VRAM, GPU prices and model memory requirements to choose hardware that fits your models and budget.
What is VRAM, and how much do you need?
VRAM is the fast memory on a graphics card. Local LLM inference needs space for model weights, the context (KV) cache and runtime buffers. Quantization can reduce weight memory, while longer prompts and more simultaneous users need additional memory. Once a model fits, generation speed comes down mostly to memory bandwidth.
Plan a local deployment
Choose a model and quantization, set the context length, and check both GPU memory and system RAM. CPU or expert offloading can make a large model fit with less VRAM, but it depends on RAM bandwidth and can be much slower. Check CUDA, ROCm or Metal support before buying hardware.
Read the rankings and price data
Compare VRAM capacity first, then bandwidth, power and software support. Our prices are median asking prices, not completed sales; sample counts, observation dates and stale-price labels help you judge their usefulness. Generation speeds on each GPU page are calibrated against public llama.cpp benchmarks.
