GPU price index
104.3
+1.9%Base 100 · 41 cards, equal weight
Lowest price per GB
$8.2 / GB16 GB900 GB/s
Lowest price per 100 GB/s
$14.5 / 100GB/s16 GB900 GB/s
Lowest price per TOPS
$1.8 / TOPS22 GB616 GB/s
Latest release
770BQ4 weights 467.3 GB
Biggest 24h drops
- 1Mac Studio M4 Max 128GB$4,788-6.7%
- 2GeForce RTX 5090 D V2$3,554-4.4%
- 3GeForce RTX 5070$918-4.3%
- 4Instinct MI50 32GB$405-3.7%
- 5Radeon RX 7800 XT$461-3.1%
Biggest 24h rises
- 1GeForce RTX 5070 Ti$1,472+21.0%
- 2Radeon PRO W7800$1,844+19.9%
- 3Radeon RX 9070 XT$907+17.3%
- 4RTX PRO 6000 Blackwell$18,422+12.6%
- 5Tesla V100 16GB$131+10.0%
Most listings
- 1DGX Spark 128GB$4,83231 listings
- 2RTX PRO 4000 Blackwell$2,32718 listings
- 3RTX PRO 5000 Blackwell 72GB$9,66517 listings
- 4RTX PRO 6000D$10,11117 listings
- 5Radeon RX 7900 XTX$86217 listings
Latest releases: minimum VRAM
- 1
Hy4 preview770B18 GB - 2
GLM 5.3744B41 GB - 3GLM 5.3 Flash320B30 GB
- 4
Qwen3.8 Flash Next176B4 GB - 5
DeepSeek V4 Pro1,600B17 GB
Local LLM GPU Ranking
Xianyu
| # | Model | 30d trend | Ecosystem | Buy | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 |  RTX PRO 6000 Blackwell | $18,422Using previous price6 listings | +12.6% | 96 | 1,792 | 500 | 1,000 | 600 | $192 | $1,028 | $18.4 | CUDA | Check prices | |
| 2 | GeForce RTX 5090 | $6,513Using previous price13 listings | +1.9% | 32 | 1,792 | 419 | 838 | 575 | $204 | $363 | $7.8 | CUDA | Check prices | |
| 3 | RTX PRO 6000D | $10,111Using previous price17 listings | -1.6% | 84 | 1,568 | 385.2 | 770.5 | 600 | $120 | $645 | $13.1 | CUDA | Check prices | |
| 4 | GeForce RTX 4090 | $3,494Using previous price16 listings | +2.6% | 24 | 1,008 | 330.3 | 660.6 | 450 | $146 | $347 | $5.3 | CUDA | Check prices | |
| 5 | GeForce RTX 4090 48GB | $3,955Using previous price3 listings | 0.0% | 48 | 1,008 | 330.3 | 660.6 | 450 | $82 | $392 | $6 | CUDA | Check prices | |
| 6 | A100 40GB PCIe | $4,602Using previous price4 listings | 0.0% | 40 | 1,555 | 312 | 624 | 250 | $115 | $296 | $7.4 | CUDA | Check prices | |
| 7 | GeForce RTX 5090 D | $4,892Using previous price13 listings | -2.9% | 32 | 1,792 | 296.9 | 593.8 | 575 | $153 | $273 | $8.2 | CUDA | Check prices | |
| 8 | GeForce RTX 5090 D V2 | $3,554Using previous price14 listings | -4.4% | 24 | 1,344 | 296.9 | 593.8 | 575 | $148 | $264 | $6 | CUDA | Check prices | |
| 9 | RTX PRO 5000 Blackwell 48GB | $6,959Using previous price15 listings | -0.4% | 48 | 1,344 | 258 | 516 | 300 | $145 | $518 | $13.5 | CUDA | Check prices | |
| 10 | RTX PRO 5000 Blackwell 72GB | $9,665Using previous price17 listings | +5.2% | 72 | 1,344 | 258 | 516 | 300 | $134 | $719 | $18.7 | CUDA | Check prices | |
| 11 | GeForce RTX 5080 | $1,851Using previous price12 listings | +0.8% | 16 | 960 | 225.1 | 450.2 | 360 | $116 | $193 | $4.11 | CUDA | Check prices | |
| 12 | RTX PRO 4500 Blackwell | $3,844Using previous price16 listings | -0.2% | 32 | 896 | 202.1 | 404.3 | 200 | $120 | $429 | $9.5 | CUDA | Check prices | |
| 13 | CMP 170HX 40GB | $1,680Using previous price4 listings | +4.6%18h | 40 | 1,560 | 202 | 404 | 250 | $42 | $108 | $4.16 | CUDA | Check prices | |
| 14 | CMP 170HX 64GB | $2,089Using previous price8 listings | 0.0% | 64 | 1,493 | 202 | 404 | 250 | $32.6 | $140 | $5.2 | CUDA | Check prices | |
| 15 | GeForce RTX 4080 | $1,152Using previous price9 listings | -0.2% | 16 | 717 | 195 | 389.9 | 320 | $72 | $161 | $2.96 | CUDA | Check prices | |
| 16 |  Radeon RX 9070 XT | $907Using previous price12 listings | +17.3% | 16 | 640 | 195 | 389 | 304 | $57 | $142 | $2.33 | ROCm | Check prices | |
| 17 | Radeon AI PRO R9700 | $1,643Using previous price8 listings | 0.0% | 32 | 640 | 191 | 383 | 300 | $51 | $257 | $4.29 | ROCm | Check prices | |
| 18 |  Arc Pro B70 | $1,465Using previous price6 listings | +1.5% | 32 | 608 | 183.5 | 367 | 230 | $45.8 | $241 | $3.99 | oneAPI | Check prices | |
| 19 | GeForce RTX 5070 Ti | $1,472Using previous price3 listings | +21.0% | 16 | 896 | 175.8 | 351.5 | 300 | $92 | $164 | $4.19 | CUDA | Check prices | |
| 20 | RTX PRO 4000 Blackwell | $2,327Using previous price18 listings | -0.9% | 24 | 672 | 161.3 | 322.5 | 145 | $97 | $346 | $7.2 | CUDA | Check prices | |
| 21 | GeForce RTX 3090 | $1,204Using previous price15 listings | +5.2% | 24 | 936 | 142.3 | 284.7 | 350 | $50 | $129 | $4.23 | CUDA | Check prices | |
| 22 | DGX Spark 128GB | $4,832Using previous price31 listings | 0.0% | 128 | 273 | 125 | 250 | 240 | $37.8 | $1,770 | $19.3 | CUDA | Check prices | |
| 23 | GeForce RTX 5070 | $918Using previous price10 listings | -4.3% | 12 | 672 | 123.5 | 247 | 250 | $77 | $137 | $3.72 | CUDA | Check prices | |
| 24 | GeForce RTX 2080 Ti 22GB | $387Using previous price3 listings | — | 22 | 616 | — | 215.2 | 250 | $17.6 | $63 | $1.8 | CUDA | Check prices | |
| 25 | Arc Pro B60 | $784Using previous price10 listings | -0.4% | 24 | 456 | 98.5 | 197 | 200 | $32.7 | $172 | $3.98 | oneAPI | Check prices | |
| 26 | GeForce RTX 5060 Ti 16GB | $697Using previous price4 listings | +1.9% | 16 | 448 | 94.9 | 189.8 | 180 | $43.6 | $156 | $3.67 | CUDA | Check prices | |
| 27 | Instinct MI210 | $2,751Using previous price12 listings | 0.0% | 64 | 1,638 | 181 | 181 | 300 | $43 | $168 | $15.2 | ROCm | Check prices | |
| 28 | Radeon PRO W7900 | $3,271Using previous price5 listings | 0.0% | 48 | 864 | 123 | 123 | 295 | $68 | $379 | $26.6 | ROCm | Check prices | |
| 29 | Radeon RX 7900 XTX | $862Using previous price17 listings | +3.6% | 24 | 960 | 123 | 123 | 355 | $35.9 | $90 | $7 | ROCm | Check prices | |
| 30 | Radeon RX 7900 XT | $639Using previous price16 listings | +4.1% | 20 | 800 | 103 | 103 | 315 | $32 | $80 | $6.2 | ROCm | Check prices | |
| 31 | Radeon RX 6950 XT | $446Using previous price13 listings | 0.0% | 16 | 576 | 47.3 | 94.6 | 335 | $27.9 | $77 | $4.71 | ROCm | Check prices | |
| 32 | Radeon PRO W7800 | $1,844Using previous price3 listings | +19.9% | 32 | 576 | 90.4 | 90.4 | 260 | $58 | $320 | $20.4 | ROCm | Check prices | |
| 33 | Radeon RX 7800 XT | $461Using previous price11 listings | -3.1% | 16 | 624 | 74.7 | 74.7 | 263 | $28.8 | $74 | $6.2 | ROCm | Check prices | |
| 34 | Instinct MI50 32GB | $405Using previous price4 listings | -3.7% | 32 | 1,024 | 26.5 | 53 | 300 | $12.7 | $39.6 | $7.6 | ROCm | Check prices | |
| 35 | Mac Studio M3 Ultra 512GB | $19,315Using previous price6 listings | 0.0% | 512 | 819 | — | — | 480 | $37.7 | $2,358 | — | Metal | Check prices | |
| 36 | Mac Studio M4 Max 128GB | $4,788Using previous price13 listings | -6.7% | 128 | 546 | — | — | 480 | $37.4 | $877 | — | Metal | Check prices | |
| 37 | Mac Studio M5 Max 128GB | $5,881Using previous price6 listings | -0.8% | 128 | 614 | — | — | — | $45.9 | $958 | — | Metal | Check prices | |
| 38 | MacBook Pro M4 Max 128GB | $5,264Using previous price4 listings | 0.0% | 128 | 546 | — | — | 140 | $41.1 | $964 | — | Metal | Check prices | |
| 39 | MacBook Pro M5 Max 128GB | $8,148Using previous price7 listings | -0.2% | 128 | 614 | — | — | 140 | $64 | $1,327 | — | Metal | Check prices | |
| 40 | Tesla V100 16GB | $131Using previous price13 listings | +10.0% | 16 | 900 | 112 | — | 250 | $8.2 | $14.5 | — | CUDA | Check prices | |
| 41 | Tesla V100 32GB | $476Using previous price11 listings | 0.0% | 32 | 900 | 112 | — | 250 | $14.9 | $53 | — | CUDA | Check prices |
Choose hardware for local LLM deployment
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. A model fitting in VRAM does not guarantee a particular speed.
Compare model VRAM requirementsWhat runs on 16GB of VRAM?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.
Local LLM hardware and deployment guidesRead 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. INT8 throughput alone is not an LLM benchmark.
Price sources and calculation methodsPrices are median asking prices, not sold prices · Methodology