Builds / RTX 3090
RTX 3090 24 GB: Ripple-65B Preview 2-bit AD-IQ2_XXS on llama.cpp at 23.6-30.9 tok/s (16K ctx, one wrong test in three coding answers)
Last updated:
Source: inferencekey.com ↗📏 How measured: reported by the original source; RigBase re-checked the numbers at that source
- GPU(s)
- RTX 3090
Amazon · eBay (used) · Newegg - VRAM / unified memory (GB)
- 24
- CPU
- not stated
- Motherboard
- not stated
- RAM
- not stated
- PSU
- not stated
- Other
- InferenceKey Cloud host, RTX 3090 24 GB; llama.cpp b12167 prebuilt CUDA 13.4 tarball, -ngl 99 -fa on --jinja, about 130 W while generating
- Reported total cost
- not stated
- Power
- about 130 W GPU while generating
Measured speed (generation tokens/sec)
| Model | Quant | Engine | tok/s | Prompt processing (tok/s) | Details |
|---|---|---|---|---|---|
| Ripple 65B Preview | AD-IQ2_XXS (AtomicChat GGUF, 20.0 GB) | llama.cpp b12167 llama-server (CUDA) | 25.9 | not stated | context: 16K ctx, q8_0 KV, one request at a time📏 SDK example final answer at 25.9 tok/s; 23.6-30.9 tok/s across the tests, 22.5 at a 12,562-token needle; peak VRAM 20,901 MiB |
Owner-run test by InferenceKey (8 Nov 2026). Quality caveat from the same test: 1 of 3 coding answers had a wrong test (function correct), and AtomicChat's card lists the 2-bit file at 80.8% same-top-token vs BF16 (93.0% for Q4_K_M). The agent bug-fix task passed. Compare with the RTX A6000 Q4_K_M build from the same article (14.3-16.3 tok/s).
Source: https://inferencekey.com/research/ripple-65b-dense-coder-one-card
Blank fields = not stated in the source. We never estimate build data; GPU street prices above are separately sourced and dated.
Rent before you buy: try this model on a cloud GPU for a few dollars first: Runpod (referral link; sign up with Google to get the new-user bonus)
Quick answers: Can RTX 3090 run Ripple 65B Preview?