Models / Nemotron 3.5 Lightning 30B-A3B / VRAM
How much VRAM do you need for Nemotron 3.5 Lightning 30B-A3B?
Last updated:
Weight math from the model card, plus every measured RigBase result for Nemotron 3.5 Lightning 30B-A3B. Leave extra room for context (KV cache).
Weight size by quant
| Quant | Bits/weight | Weights only |
|---|---|---|
| Q8_0 | 8.50 | 31.9 GB |
| Q6_K | 6.56 | 24.6 GB |
| Q5_K_M | 5.70 | 21.4 GB |
| Q4_K_M | 4.89 | 18.4 GB |
| Q3_K_M | 4.00 | 15.0 GB |
| Q2_K | 3.16 | 11.8 GB |
Bits-per-weight for GGUF rows: llama.cpp quantize README. Parameter count: 30B total / 3B active MoE (hybrid Mamba), per DGX Spark benchmark table. We advise staying at Q4 or above for chat quality — see Is Q4 good enough?.
Measured speeds for Nemotron 3.5 Lightning 30B-A3B
| Hardware | Quant | Engine | tok/s |
|---|---|---|---|
| NVIDIA GeForce RTX 3090 24 GB · 24 GB | Q4_K_M (23.7 GiB, partial CPU offload) | llama.cpp CUDA b11804 | 101.0 |
| Nvidia GB10 (unified) · 128 GB | + DSpark speculative decoding (k=3) | vLLM 0.31.0, 128K ctx, 1 user | 94.4 ⚡ |
| Nvidia GB10 (unified) · 128 GB | no speculative decoding | vLLM 0.31.0, 128K ctx, 1 user | 70.3 |
| Apple M6 (12-core GPU, unified) · 32 GB | GGUF Q4_K_M (lmstudio-community) | LM Studio (llama.cpp) | 45.1 |
⚡ = speculative decoding (MTP, DFlash…). Full list: best home rig for Nemotron 3.5 Lightning 30B-A3B.
Related
All Nemotron 3.5 Lightning 30B-A3B 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 Nemotron 3.5 Lightning 30B-A3B?
Weights alone are about 18.4 GB at Q4_K_M (30B × 4.89 bits ÷ 8). Add memory for the KV cache (context) and the runtime. The smallest-memory measured result in RigBase is 24 GB (NVIDIA GeForce RTX 3090 24 GB) at 101.0 tok/s. Going below Q4 usually hurts quality for chat; see the Q4 guide.
What is the fastest measured speed for Nemotron 3.5 Lightning 30B-A3B?
NVIDIA GeForce RTX 3090 24 GB: 101.0 tok/s plain decode (Q4_K_M (23.7 GiB, partial CPU offload), llama.cpp CUDA b11804).
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.
Referral link: if you sign up with Google and add $10, Runpod gives you and RigBase bonus credit (terms). It never affects our data.