Used GPU prices by model
Live eBay prices for 51 GPU models, from 3,452 current
single-card listings as of October 3, 2026. These are asking prices on live eBay listings, not sold prices, shipping included. Only single-card fixed-price listings count, so lots, auction bids and for-parts listings can't skew them. The low end is the 10th-percentile ask: one in ten listings ask that much or less, so it's a price you can find near the top of a search sorted by price. The median is shown beside it. Asks under 50% of the median don't count toward the low end, since prices that far below the market are often scams or mislabeled listings.
| Model | Live listings |
Low end | Median ask | Low-end $/GB VRAM | GB/s | GB/s per $1k |
|---|
| NVIDIA |
|---|
| RTX 5080 16GB | 396 | $1,714 | $2,347 | $107 | 960 | 560 |
| RTX 3090 24GB | 328 | $1,550 | $1,880 | $64.58 | 936 | 604 |
| Tesla V100 16GB | 226 | $202 | $339 | $12.66 | 900 | 4,444 |
| Tesla V100 32GB | 211 | $720 | $850 | $22.50 | 900 | 1,250 |
| Tesla P100 16GB | 184 | $120 | $185 | $7.50 | 732 | 6,101 |
| Tesla P40 24GB | 147 | $290 | $360 | $12.08 | 346 | 1,193 |
| Quadro RTX 5000 16GB | 110 | $380 | $580 | $23.75 | 448 | 1,179 |
| RTX 4080 Super 16GB | 109 | $1,184 | $1,583 | $73.99 | 736 | 622 |
| RTX 4080 16GB | 107 | $1,150 | $1,545 | $71.87 | 717 | 623 |
| RTX A4000 16GB | 93 | $995 | $1,498 | $62.19 | 448 | 450 |
| A100 40GB | 89 | $4,273 | $6,387 | $107 | 1,555 | 364 |
| A2 16GB | 78 | $479 | $556 | $29.94 | 200 | 418 |
| Quadro RTX 6000 24GB | 78 | $970 | $1,641 | $40.42 | 672 | 693 |
| Quadro RTX 8000 48GB | 68 | $2,010 | $2,869 | $41.88 | 672 | 334 |
| RTX 4000 Ada 20GB | 63 | $1,600 | $1,986 | $80.00 | 360 | 225 |
| RTX 5090 D 32GB | 58 | $6,198 | $8,860 | $194 | 1,792 | 289 |
| RTX A6000 48GB | 56 | $4,700 | $5,922 | $97.92 | 768 | 163 |
| RTX 3090 Ti 24GB | 54 | $1,600 | $2,062 | $66.67 | 1,008 | 630 |
| A40 48GB | 53 | $4,249 | $6,021 | $88.53 | 696 | 164 |
| RTX A4500 20GB | 47 | $1,300 | $1,696 | $65.00 | 640 | 492 |
| RTX A5000 24GB | 46 | $2,400 | $3,500 | $100 | 768 | 320 |
| RTX 4090 D 24GB | 45 | $3,549 | $4,377 | $148 | 1,008 | 284 |
| A16 64GB | 43 | $2,385 | $3,649 | $37.27 | 800 | 335 |
| L4 24GB | 43 | $3,999 | $4,845 | $167 | 300 | 75 |
| A30 24GB | 41 | $2,459 | $3,955 | $102 | 933 | 379 |
| A100 80GB | 37 | $13,376 | $18,999 | $167 | 1,935 | 145 |
| L40S 48GB | 37 | $7,450 | $9,246 | $155 | 864 | 116 |
| RTX 5090 D 24GB | 29 | $5,031 | $5,573 | $210 | 1,344 | 267 |
| RTX 6000 Ada 48GB | 26 | $7,034 | $10,048 | $147 | 960 | 136 |
| RTX 4090 D 48GB | 25 | $4,680 | $5,121 | $97.50 | 1,008 | 215 |
| RTX 4060 Ti 16GB | 24 | $740 | $942 | $46.25 | 288 | 389 |
| RTX 4090 48GB | 23 | $5,155 | $5,450 | $107 | 1,008 | 196 |
| A10 24GB | 22 | $3,385 | $4,669 | $141 | 600 | 177 |
| Tesla V100S 32GB | 21 | $4,899 | $4,899 | $153 | 1,134 | 231 |
| H100 80GB | 18 | $26,880 | $40,880 | $336 | 2,000 | 74 |
| RTX 4000 SFF Ada 20GB | 18 | $1,850 | $2,045 | $92.50 | 280 | 151 |
| RTX 5000 Ada 32GB | 18 | $3,507 | $4,691 | $110 | 576 | 164 |
| A10G 24GB | 15 | $2,902 | $3,420 | $121 | 600 | 207 |
| RTX 4080 Super 32GB | 14 | $2,400 | $2,569 | $75.00 | 736 | 307 |
| RTX A4000 8GB | 14 | $225 | $374 | $28.19 | — | — |
| Tesla V100 12GB | 13 | $382 | $480 | $31.83 | — | — |
| A100 64GB | 10 | $4,990 | $9,450 | $77.97 | — | — |
| Tesla P100 12GB | 10 | $86.92 | $110 | $7.24 | 549 | 6,316 |
| RTX A4500 16GB | 8 | $350 | $437 | $21.88 | — | — |
| AMD |
|---|
| Radeon VII 16GB | 68 | $260 | $357 | $16.25 | 1,024 | 3,938 |
| Instinct MI50 32GB | 60 | $610 | $738 | $19.06 | 1,024 | 1,679 |
| Radeon Pro W6800 32GB | 53 | $998 | $1,291 | $31.19 | 512 | 513 |
| Instinct MI50 16GB | 34 | $200 | $245 | $12.50 | 1,024 | 5,120 |
| Instinct MI210 64GB | 32 | $4,100 | $4,981 | $64.06 | 1,638 | 400 |
| Radeon Pro VII 16GB | 32 | $494 | $630 | $30.86 | 1,024 | 2,074 |
| Instinct MI100 32GB | 18 | $1,500 | $1,845 | $46.87 | 1,229 | 819 |
Most memory bandwidth per dollar
As of October 3, 2026, the used GPU with the most memory bandwidth per dollar is the NVIDIA Tesla P100 12GB: 549 GB/s for a $86.92 low-end ask, or 6,316 GB/s per $1,000.
| # | Model | GB/s |
Low-end ask | GB/s per $1k |
|---|
| 1 | NVIDIA Tesla P100 12GB | 549 | $86.92 | 6,316 |
| 2 | NVIDIA Tesla P100 16GB | 732 | $120 | 6,101 |
| 3 | AMD Instinct MI50 16GB | 1,024 | $200 | 5,120 |
| 4 | NVIDIA Tesla V100 16GB | 900 | $202 | 4,444 |
| 5 | AMD Radeon VII 16GB | 1,024 | $260 | 3,938 |
| 6 | AMD Radeon Pro VII 16GB | 1,024 | $494 | 2,074 |
| 7 | AMD Instinct MI50 32GB | 1,024 | $610 | 1,679 |
| 8 | NVIDIA Tesla V100 32GB | 900 | $720 | 1,250 |
| 9 | NVIDIA Tesla P40 24GB | 346 | $290 | 1,193 |
| 10 | NVIDIA Quadro RTX 5000 16GB | 448 | $380 | 1,179 |
| 11 | AMD Instinct MI100 32GB | 1,229 | $1,500 | 819 |
| 12 | NVIDIA Quadro RTX 6000 24GB | 672 | $970 | 693 |
Memory bandwidth is the vendor's published spec; the dollars are the low-end ask,
which one in ten listings match or beat.
It leaves out compute, VRAM size, power and cooling. SXM modules need an SXM board, and passively cooled datacenter cards need server airflow, so check the form-factor split on a card's page. Laptop parts have no single published figure and aren't ranked.
A model gets a page once it has 8 live single-card listings.
Each page breaks the price down by condition, shows the asking-price trend and lists the
cheapest live listings. To price a whole server, use the build sheet.
Also: CPU prices · Proven AI builds.