NV-Link Connector available.
This is the FTW3 ULTRA — EVGA's flagship 3090 — with the official EVGA HYBRID Kit (400-HY-1988-B1) professionally installed. That's not a third-party block, not a custom loop — it's EVGA's own factory-engineered AIO conversion. Same kit that ships on the 24G-P5-3988-KR hybrid models, just applied to a 3987-KR donor card. The result is a quieter, cooler, smaller-footprint 3090 that solves the GDDR6X memory junction temp problem stock 3090s are infamous for.
The card:
- NVIDIA GeForce RTX 3090, 24GB GDDR6X (384-bit)
- 10,496 CUDA cores, 2nd-gen RT Cores, 3rd-gen Tensor Cores
- Non-LHR — 24G-P5-3987-KR launch silicon, unrestricted CUDA
- PCIe Gen 4
- 1x HDMI 2.1, 3x DisplayPort 1.4a
- HDCP 2.3, NVLink (SLI-ready)
- 3x 8-pin power (or 6+2-pin), 750W+ PSU recommended
- Adjustable ARGB LED, metal backplate, EVGA Precision X1 support
- Max digital resolution: 7680 × 4320
- UPC: 843368067137
The HYBRID Kit (installed):
- All-in-one closed-loop liquid cooler — sealed, leak-tested, no maintenance
- 240mm radiator with dual 120mm fans
- Copper base heatsink covering GPU + VRAM
- Sleeved tubing, integrated wiring
- Fan curves controllable via EVGA Precision X1
- Drops card footprint from 2.75-slot to ~2-slot
Condition: Pulled from a working Gaming PC / WorkStation. Tested, cleaned, repasted at install. AIO is sealed and self-contained — no top-up required. No coil whine, no artifacts. Original card box included in photos.
24GB of unrestricted VRAM is the whole reason 3090s still command the prices they do — 4K maxed, 8K-capable in supported titles, and the only consumer card from this era with enough memory to run serious local LLMs (Llama 70B at decent quants, full SDXL pipelines, ComfyUI without VRAM gymnastics). With the HYBRID conversion, you skip the thermal throttling that haunts stock air-cooled 3090s during sustained AI workloads. EVGA is dead. NVIDIA isn't making this kind of memory accessible at consumer prices anymore. This is the card a lot of people wish they had bought.
