AMD · CDNA 3 · 192 GB · Qwen3.8 27B · Q4_K_M at 8K context: ~160 tok/s
AMD Instinct MI300X
192 GB of HBM3 at 5.3 TB/s on one module, the most memory of anything here, but it is an OAM module that only lives in an 8-way server. Listed because surplus modules do surface; budget a chassis around them.
Lowest price right now
$27,570.56
eBay · $27,549.95 + $20.61 shipping · $143 per GB · observed 2026-11-14 18:37 UTC
Buy the cheapest on eBay, $27,570.56
$27,549.95 + $20.61 shipping · Open box · rtone-32 995 · 100.0%
Price alert
One email per move, item plus shipping. Clear the price to hear about any drop. Manage alerts.
Price history
Daily low across every tracked vendor, last 6 days.
Where to buy
| Vendor | Price | Condition | Observed | |
|---|---|---|---|---|
| eBay cheapest active listing |
$27,570.56 $27,549.95 + $20.61 shipping |
new and used | 2026-11-14 18:37 UTC | Go to eBay |
On eBay now
4 listings passed our filters at the last check, 7 min ago. We check every 28 minutes. Cheapest delivered first. Seller reputation is feedback count and positive rate.
-
$27,570.56$27,549.95 + $20.61 shipping -
$31,999free shipping -
$34,999.99free shipping -
$40,299.99free shipping
What it runs
For each model, grouped by vendor with the newest first: the largest quant that fits with 8K of context and the estimated speed for one user, at 8K and at 32K of context. On a card, a model too big for its memory can still run with part of the weights in system RAM; we assume 64 GB, and the speed shown accounts for it. The model you picked on the ladder is highlighted. Estimates from the spec sheet, not measurements: see how we estimate.
| Model | Quant | Needs at 8K | Fit | Max context | Decode at 8K | At 32K | Prefill tok/s |
|---|---|---|---|---|---|---|---|
| DeepSeek | |||||||
| DeepSeek V4.1 Flash 552B-A16B Sep 2026 | IQ2_M | 189 GB | fits | 128K+ | ~180 | ~180 | ~10,000 |
| DeepSeek V4 Flash 284B-A13B Jul 2026 | Q4_K_M | 175 GB | fits | 128K+ | ~150 | ~140 | ~13,000 |
| DeepSeek V3.1 671B-A37B Aug 2025 | IQ2_M | 230 GB | runs with 39 GB in system RAM | ~17 | ~16 | ~2,200 | |
| Gemma | |||||||
| Gemma 4 12B May 2026 | Q8_0 | 14 GB | fits | 128K+ | ~210 | ~200 | ~14,000 |
| Gemma 4 31B Mar 2026 | Q8_0 | 35 GB | fits | 128K+ | ~80 | ~75 | ~5,300 |
| Gemma 4 26B-A4B Mar 2026 | Q8_0 | 28 GB | fits | 128K+ | ~200 | ~190 | ~43,000 |
| Gemma 3 27B Mar 2025 | Q8_0 | 31 GB | fits | 128K+ | ~89 | ~84 | ~6,000 |
| GLM | |||||||
| GLM-5.3 753B-A40B Aug 2026 | Q4_K_M | 462 GB | does not fit | ||||
| GLM-5.3 Flash 320B-A18B Aug 2026 | Q3_K_M | 159 GB | fits | 128K+ | ~140 | ~140 | ~9,100 |
| GLM-4.6 355B-A32B Sep 2025 | Q3_K_M | 179 GB | fits | 42K | ~95 | ~71 | ~5,100 |
| GLM-4.5 Air 106B-A12B Jul 2025 | Q8_0 | 115 GB | fits | 128K+ | ~110 | ~94 | ~14,000 |
| gpt-oss | |||||||
| gpt-oss-120b Aug 2025 | Q8_0 | 126 GB | fits | 128K+ | ~170 | ~150 | ~32,000 |
| gpt-oss-20b Aug 2025 | Q8_0 | 24 GB | fits | 128K+ | ~200 | ~180 | ~45,000 |
| Llama | |||||||
| Llama 4 Scout 109B-A17B Apr 2025 | Q8_0 | 119 GB | fits | 128K+ | ~92 | ~79 | ~9,600 |
| Llama 3.3 70B Dec 2024 | Q8_0 | 79 GB | fits | 128K+ | ~34 | ~31 | ~2,300 |
| Llama 3.1 8B Jul 2024 | Q8_0 | 11 GB | fits | 128K+ | ~280 | ~210 | ~20,000 |
| MiMo | |||||||
| MiMo V2.6 Pro 1.02T-A42B Sep 2026 | Q4_K_M | 624 GB | does not fit | ||||
| MiMo V2.6 Flash 309B-A15B Sep 2026 | Q4_K_M | 190 GB | fits | 70K | ~140 | ~140 | ~11,000 |
| Mistral | |||||||
| Mistral Medium 3.5 128B Mar 2026 | Q8_0 | 140 GB | fits | 128K+ | ~19 | ~18 | ~1,300 |
| Mistral Small 4 119B-A6.5B Mar 2026 | Q8_0 | 132 GB | fits | 106K | ~130 | ~75 | ~25,000 |
| Mistral Small 3.2 24B Jun 2025 | Q8_0 | 28 GB | fits | 128K+ | ~99 | ~86 | ~6,800 |
| Nemotron | |||||||
| Nemotron 3.5 Lightning 30B-A3B Aug 2026 | Q8_0 | 35 GB | fits | 128K+ | ~200 | ~200 | ~47,000 |
| Qwen | |||||||
| Qwen3.8 Flash-Next 125B-A6B Aug 2026 | Q8_0 | 134 GB | fits | 128K+ | ~170 | ~160 | ~27,000 |
| Qwen3.8 27B Aug 2026 (your pick) | Q8_0 | 31 GB | fits | 128K+ | ~91 | ~86 | ~6,100 |
| Qwen3.6 27B Apr 2026 | Q8_0 | 31 GB | fits | 128K+ | ~91 | ~86 | ~6,100 |
| Qwen3.6 35B-A3B Apr 2026 | Q8_0 | 39 GB | fits | 128K+ | ~210 | ~200 | ~54,000 |
| Qwen3.5 9B Feb 2026 | Q8_0 | 11 GB | fits | 128K+ | ~270 | ~250 | ~18,000 |
| Qwen3 Coder 480B-A35B Jul 2025 | IQ2_M | 167 GB | fits | 104K | ~110 | ~90 | ~4,700 |
| Qwen3 235B-A22B Jul 2025 | Q5_K_M | 170 GB | fits | 119K | ~100 | ~85 | ~7,400 |
| Qwen3 32B Apr 2025 | Q8_0 | 38 GB | fits | 128K+ | ~72 | ~61 | ~5,000 |
Specifications
- Memory
- 192 GB HBM3
- Memory bandwidth
- 5,300 GB/s
- Bus width
- 8192-bit
- Shader cores
- 19,456
- Tensor cores
- 1,216
- FP32
- 81.7 TFLOPS
- FP16 (dense)
- 1307 TFLOPS
- FP8 (dense)
- 2615 TFLOPS
- Power
- 750 W
- Form factor
- OAM module: needs an 8-way UBB server, not a desktop part
- Architecture
- CDNA 3
- Launched
- 2023
Source: manufacturer specifications.