AMD Instinct MI250X
Live eBay prices, cheapest first. The best one right now is $3,632 (used).
Not a PCIe card. OAM module: needs an OAM baseboard/server; not practical in a regular motherboard. No OAM-to-PCIe adapter runs it yet (November 2026): hobbyists get MI250 / MI250X modules to power up on one, but not past the driver's start-up.
Specs
- VRAM
- 128 GB HBM2e
- Bandwidth
- 3,277 GB/s
- FP32
- 47.9 TFLOPS
- FP16
- 383 TFLOPS
- Power
- 560 W
- Connector
- OAM (PCIe 4.0)
- Released
- 2021
- Llama 3.1 8B at Q4 (est.)
- 236–405 tokens/s
Price right now
Live on eBayFrom the live listings below; shipping included.
Listings
Live on eBay| Listing | Condition | Price | Checked |
|---|---|---|---|
| Used | $3,632or best offerfree shipping | ||
| Used | $6,000or best offerfree shipping |
Auctions ending soon
No auctions ending soon right now.
Prices include shipping and were read from eBay at the time shown. They can change: check the listing before you buy.
Owner reviews
No owner reviews yet. Own one? Tell others how it runs.
From signed-in owners, checked before they're published. PerGB doesn't pay for or edit reviews.
How it compares
Live on eBay- Software: ROCm (on AMD's support list).
Faster, for the least
- AMD Instinct MI300X: $27,550, 1021% of the GeForce RTX 3090, 192 GB
- NVIDIA H200 NVL: $32,300, 777% of the GeForce RTX 3090, 141 GB
- NVIDIA H100 NVL: $32,600, 700% of the GeForce RTX 3090, 94 GB
From the live listings of every GPU, shipping included; performance against the GeForce RTX 3090 (FP16 or FP32 compute and memory bandwidth). Updated with every build.
Local AI: what it runs
| Model (Q4, 8K context) | Instinct MI250X needed | Costs now | Speed (est.) |
|---|---|---|---|
| Llama 3.1 8B | 1 card | $3,632 | 236–405 tokens/s |
| Qwen3 14B | 1 card | $3,632 | 128–219 tokens/s |
| gpt-oss 20B | 1 card | $3,632 | 315–541 tokens/s |
| Gemma 3 27B | 1 card | $3,632 | 69–118 tokens/s |
| Qwen3 32B | 1 card | $3,632 | 58–99 tokens/s |
| Llama 3.3 70B | 1 card | $3,632 | 27–46 tokens/s |
| gpt-oss 120B | 1 card | $3,632 | 223–382 tokens/s |
| Qwen3 235B-A22B | 2 cards | $9,632 | 52–88 tokens/s |
| DeepSeek R1 / V3 671B | 4 cards | not enough listings | 31–53 tokens/s |
Cost: that many of the cheapest live listings. Speed: an estimate from the memory bandwidth for one person chatting. Try other models, sizes and contexts.
Decision models
| Model (Q4, 8K context) | Instinct MI250X needed | Costs now | Per question (est.) |
|---|---|---|---|
| Julia-10.144B | 1 card | $3,632 | 2.3–5.0 ms |
| Laya0.421B | 1 card | $3,632 | 3.9–8.3 ms |
| Kev-4B4B | 1 card | $3,632 | 9.3–20 ms |
| lev4B | 1 card | $3,632 | 28–60 ms |
| OpenJev27B | 1 card | $3,632 | 33–72 ms |
| Clef27B | 1 card | $3,632 | 33–72 ms |
| Perplexity Decider v1.227B | 1 card | $3,632 | 33–72 ms |
| mammoth-127B | 1 card | $3,632 | 33–72 ms |
Decision models score the options you give them instead of writing text, so they're timed per question. The time is scaled from each model's reported speed on an RTX PRO 6000 by this card's compute and memory bandwidth, within about half either way. About the models.
Compare elsewhere
Search other shops for AMD Instinct MI250X. These open the shop's own search; we don't show their prices here.

