AMD Instinct MI300X
Live eBay prices, cheapest first. The best one right now is $27,550 (open box).
Not a PCIe card. OAM module: needs an OAM baseboard/server; not practical in a regular motherboard.
Specs
- VRAM
- 192 GB HBM3
- Bandwidth
- 5,300 GB/s
- FP32
- 163.4 TFLOPS
- FP16
- 1307 TFLOPS
- Power
- 750 W
- Connector
- OAM (PCIe 5.0)
- Released
- 2023
- Launch price
- $15,000 (est.)
- Llama 3.1 8B at Q4 (est.)
- 382–656 tokens/s
Price right now
Live on eBayFrom the live listings below; shipping included.
Listings
Live on eBay| Listing | Condition | Price | Checked |
|---|---|---|---|
| Open box | $27,550−14%shipping not listed | ||
| Open box | $29,900or best offer−7%free shipping | ||
| Used | $31,999or best offerfree shipping | ||
| Refurbished | $35,000or best offerfree shipping | ||
| Used | $40,300or 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- #1 of 1 GPUs with 192 GB or more for the money: $2,698 per GeForce RTX 3090 of performance.
- Software: ROCm (on AMD's support list).
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 MI300X needed | Costs now | Speed (est.) |
|---|---|---|---|
| Llama 3.1 8B | 1 card | $27,550 | 382–656 tokens/s |
| Qwen3 14B | 1 card | $27,550 | 207–354 tokens/s |
| gpt-oss 20B | 1 card | $27,550 | 510–874 tokens/s |
| Gemma 3 27B | 1 card | $27,550 | 112–191 tokens/s |
| Qwen3 32B | 1 card | $27,550 | 93–160 tokens/s |
| Llama 3.3 70B | 1 card | $27,550 | 43–74 tokens/s |
| gpt-oss 120B | 1 card | $27,550 | 360–617 tokens/s |
| Qwen3 235B-A22B | 1 card | $27,550 | 83–143 tokens/s |
| DeepSeek R1 / V3 671B | 3 cards | $89,449 | 50–85 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 MI300X needed | Costs now | Per question (est.) |
|---|---|---|---|
| Julia-10.144B | 1 card | $27,550 | 1.0–2.1 ms |
| Laya0.421B | 1 card | $27,550 | 1.7–3.5 ms |
| Kev-4B4B | 1 card | $27,550 | 4.0–8.5 ms |
| lev4B | 1 card | $27,550 | 12–26 ms |
| OpenJev27B | 1 card | $27,550 | 14–31 ms |
| Clef27B | 1 card | $27,550 | 14–31 ms |
| Perplexity Decider v1.227B | 1 card | $27,550 | 14–31 ms |
| mammoth-127B | 1 card | $27,550 | 14–31 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 MI300X. These open the shop's own search; we don't show their prices here.




