![[FS] [US-CA] AI Server Supermicro H12SSL-I - EPYC 4 x RTX 3090 AI](https://external-preview.redd.it/k3ED3dcqvyWbi8lXj__pKYp4WMteD2NKeq7Cm42ngrk.jpeg?width=1080&crop=smart&auto=webp&s=b64048fc2e5d85af69769d87eb9c2a76906683ca)
[FS] [US-CA] AI Server Supermicro H12SSL-I - EPYC 4 x RTX 3090 AI
Selling my AI LLM inference server
Specs:
CPU: AMD EPYC 7262 — 8 Core / 16 Thread | 128MB L3 Cache
Motherboard: Supermicro H12SSL-I — Server ATX | PCIe 4.0 | IPMI
RAM: 256GB DDR4 3200MHz ECC RDIMM — NEMIX 8x32GB ($2,990 new on Amazon)
Storage: Solidigm P41 Plus 2TB NVMe PCIe 4.0 M.2
PSU: Super Flower Leadex Platinum 1600W — 80+ Platinum | 10 Year Warranty
GPUs — 96GB Total VRAM:
NVIDIA RTX 3090 Founders Edition 24GB
EVGA RTX 3090 XC3 Ultra Gaming 24GB
ZOTAC RTX 3090 Trinity 24GB
ZOTAC RTX 3090 Trinity OC 24GB
I also have 4 GPU risers PCIe4 (~$200), a GPU open rack with multiple fans, and an ARCTIC Freezer 4U-SP5 (~$50).
The 3090 Founder Edition was originally used for gaming and then switched to LLM Inference.
The 3 other GPUs are from Newegg and are only used for AI Inference. No mining or 24/7 operations. The whole server was used sporadically, mostly to test out local LLMs like GPT-OSS or Qwen. Selling because I mostly use cloud models nowadays.
The FE and EVGA have some metal fins added to dissipate more heat, but the GPUs were never opened or repasted.
| Item | Price | Priority |
|---|---|---|
| Server+GPUs | $6000 | Preferred |
| 4 x RTX 3090 | $4000 | Preferred as a 4 x bundle |
| Server Only (CPU+Memory+Mobo+NVMe+PSU+Cooler) | $2500 | If GPUs have already been sold |
| NVIDIA RTX 3090 Founders Edition | $1200 | If I split the GPUs |
| EVGA RTX 3090 XC3 Ultra Gaming 24GB | $1100 | If I split the GPUs |
| ZOTAC RTX 3090 Trinity 24GB | $1000 | If I split the GPUs |
| ZOTAC RTX 3090 Trinity OC 24GB | $1050 | If I split the GPUs |
Prefer local only; I can meet anywhere within the San Francisco Bay Area. Cash, PayPal, or QuickBooks Payment (bank transfer with Invoice + extra fees)
I might be open to trading it for a DGX Spark + Cash difference (negotiable)
For local AI enthusiast:
If you are interested, I also included some benchmarks trying to make a summary of a 98-page book at a 128k context window
GPT-OSS:120B - Text 95 t/s, prompt 2340 t/s
qwen3.6:35b - Text 114 t/s, prompt 4325 t/s
gemma4:31b - Text 29 t/s, prompt 1174 t/s