More
Submit build

Models / Ripple 65B Preview / VRAM

How much VRAM do you need for Ripple 65B Preview?

Last updated:

Weight math from the model card, plus every measured RigBase result for Ripple 65B Preview. Leave extra room for context (KV cache).

Quick answer: at Q4_K_M the weights are about 39.8 GB (65B params × 4.89 bits ÷ 8, llama.cpp quantize table). the smallest-memory build with a measured result here is RTX 3090 (24 GB) at 25.9 tok/s. fastest plain decode: 25.9 tok/s on RTX 3090. Open the VRAM calculator →

Weight size by quant

QuantBits/weightWeights only
Q8_08.5069.1 GB
Q6_K6.5653.3 GB
Q5_K_M5.7046.3 GB
Q4_K_M4.8939.8 GB
Q3_K_M4.0032.5 GB
Q2_K3.1625.7 GB

Bits-per-weight for GGUF rows: llama.cpp quantize README. Parameter count: Ricursive model card: dense 65B, 80 layers (64 sliding-window + 16 global), Apac. We advise staying at Q4 or above for chat quality — see Is Q4 good enough?.

Measured speeds for Ripple 65B Preview

HardwareQuantEnginetok/s
RTX 3090 · 24 GBAD-IQ2_XXS (AtomicChat GGUF, 20.0 GB)llama.cpp b12167 llama-server (CUDA)25.9
Apple M3 Ultra (unified, Mac Studio) · 96 GBQ4_K_M (self-quantized, 38.8 GB)llama.cpp (Metal)15.0
RTX A6000 48 GB · 48 GBQ4_K_M (AtomicChat GGUF, 38.8 GB)llama.cpp b12167 llama-server (CUDA)14.4
4x Intel Arc Pro B70 32 GB · 128 GBQ4_K_M (AtomicChat GGUF, 38.8 GB)llama.cpp b12314 (SYCL), layer split acr13.0

⚡ = speculative decoding (MTP, DFlash…). Full list: best home rig for Ripple 65B Preview.

Related

All Ripple 65B Preview builds VRAM + KV calculator By VRAM size GGUF vs AWQ vs FP8 vs NVFP4 How much system RAM?

FAQ

How much VRAM do you need for Ripple 65B Preview?

Weights alone are about 39.8 GB at Q4_K_M (65B × 4.89 bits ÷ 8). Add memory for the KV cache (context) and the runtime. The smallest-memory measured result in RigBase is 24 GB (RTX 3090) at 25.9 tok/s. Going below Q4 usually hurts quality for chat; see the Q4 guide.

What is the fastest measured speed for Ripple 65B Preview?

RTX 3090: 25.9 tok/s plain decode (AD-IQ2_XXS (AtomicChat GGUF, 20.0 GB), llama.cpp b12167 llama-server (CUDA)).

Does system RAM replace VRAM?

Only for overflow. When the whole model fits in GPU or unified memory, system RAM barely matters for decode speed. See the system-RAM guide.

Not enough VRAM? Rent before you buy. Run Ripple 65B Preview on a rented 24–96 GB cloud GPU by the hour and see the real speed before spending €1,000+. Try it on Runpod
Referral link: if you sign up with Google and add $10, Runpod gives you and RigBase bonus credit (terms). It never affects our data.