Hardware requirements

What hardware do I need to run Nemotron 3.5 Lightning 30B-A3B?

31.6B MoE · 3.2B active 256K context OpenMDW-1.1 New this month

At Q4_K_M the weights are 17.8 GB; with an 8K KV cache and runtime overhead that is 18.4 GB, so it wants a 24 GB card to run with headroom — in practice 24 GB VRAM: GeForce RTX 4090 and 8 others. Every figure below is computed from the model’s published shape against each device’s usable memory; tokens per second are estimates. The method.

Minimum practical 20 GB VRAM

Radeon RX 7900 XT — Q3_K_M, 15.0 GB of 18.4 GB, ~128 tok/s, up to 256K context.

Also: RTX 4000 Ada
the smallest card that holds it at all, at Q3_K_M — the lowest quantisation worth running
Recommended 24 GB VRAM

GeForce RTX 4090 — Q4_K_M, 18.4 GB of 22.4 GB, ~130 tok/s, up to 256K context.

Also: Radeon RX 7900 XTX, GeForce RTX 3090 Ti, GeForce RTX 3090, GeForce RTX 5090 Laptop, RTX A5000 +3 more
runs the recommended quantisation with headroom at 8K context
High quality 40 GB VRAM

A100 40 GB — Q8_0, 31.9 GB of 38.4 GB, ~114 tok/s, up to 256K context.

near-lossless Q8_0 with headroom
Long context 32 GB VRAM

GeForce RTX 5090 — Q4_K_M, 19.1 GB of 30.4 GB, ~232 tok/s at 128K.

Also: RTX 5000 Ada, Radeon Pro W7800
128K tokens (or the model’s maximum) at the recommended quantisation
Apple Silicon 32 GB unified

M1 Pro · 32 GB — Q4_K_M, 18.4 GB of 24.0 GB, ~29 tok/s, up to 256K context.

Also: M2 Pro · 32 GB, M4 · 32 GB
smallest Mac that runs the recommended quantisation with headroom
CPU only 32 GB RAM

CPU only · DDR4 dual-channel — Q6_K, 24.8 GB of 25.6 GB, ~3.9 tok/s, up to 148K context.

no GPU — weights stream from system RAM

How much VRAM at each quantisation

QuantWeights+KV 8K+KV 32K+KV 128KTotal @ 8KNeedsQuality
Q8_031.3 GB0.05 GB0.19 GB0.8 GB 31.9 GB40 GB card−0.1% ppl
Q6_K24.1 GB0.05 GB0.19 GB0.8 GB 24.8 GB32 GB card−0.4% ppl
Q5_K_M20.9 GB0.05 GB0.19 GB0.8 GB 21.5 GB32 GB card−0.8% ppl
Q4_K_M17.8 GB0.05 GB0.19 GB0.8 GB 18.4 GB24 GB card−1.9% ppl
Q3_K_M14.4 GB0.05 GB0.19 GB0.8 GB 15.0 GB20 GB card−5.4% ppl
Q2_K12.3 GB0.05 GB0.19 GB0.8 GB 13.0 GB20 GB card−15% ppl

Totals add 0.6 GB runtime overhead and an f16 KV cache; "needs" is the smallest discrete card class whose usable memory (after display and driver reserve) leaves 15% headroom. A q8_0 cache roughly halves the KV columns.

Every device, checked

System RAM for offload:
163264128256

NVIDIA · GeForce

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
GeForce RTX 5090 32 GB Q6_K Q6_K 256K 232 yes Check
GeForce RTX 5080 16 GB — offload — ~93 no Check
GeForce RTX 5070 Ti 16 GB — offload — ~89 no Check
GeForce RTX 5070 12 GB — offload — ~26 no Check
GeForce RTX 5060 Ti 16 GB 16 GB — offload — ~56 no Check
GeForce RTX 5060 8 GB — offload — ~16 no Check
GeForce RTX 4090 24 GB Q4_K_M Q5_K_M 256K 130 tight Check
GeForce RTX 4080 Super 16 GB — offload — ~79 no Check
GeForce RTX 4080 16 GB — offload — ~77 no Check
GeForce RTX 4070 Ti Super 16 GB — offload — ~74 no Check
GeForce RTX 4070 Ti 12 GB — offload — ~25 no Check
GeForce RTX 4070 Super 12 GB — offload — ~25 no Check
GeForce RTX 4070 12 GB — offload — ~25 no Check
GeForce RTX 4060 Ti 16 GB 16 GB — offload — ~39 no Check
GeForce RTX 4060 Ti 8 GB — offload — ~15 no Check
GeForce RTX 4060 8 GB — offload — ~15 no Check
GeForce RTX 3090 Ti 24 GB Q4_K_M Q5_K_M 256K 130 tight Check
GeForce RTX 3090 24 GB Q4_K_M Q5_K_M 256K 121 tight Check
GeForce RTX 3080 Ti 12 GB — offload — ~27 no Check
GeForce RTX 3080 12 GB 12 GB — offload — ~27 no Check
GeForce RTX 3080 10 GB — offload — ~20 no Check
GeForce RTX 3070 Ti 8 GB — offload — ~16 no Check
GeForce RTX 3070 8 GB — offload — ~16 no Check
GeForce RTX 3060 Ti 8 GB — offload — ~16 no Check
GeForce RTX 3060 12 GB 12 GB — offload — ~23 no Check
GeForce RTX 2080 Ti 11 GB — offload — ~22 no Check
GeForce RTX 2060 12 GB 12 GB — offload — ~22 no Check
GeForce GTX 1080 Ti 11 GB — offload — ~21 no Check

NVIDIA · GeForce laptop

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
GeForce RTX 5090 Laptop 24 GB Q4_K_M Q5_K_M 256K 116 tight Check
GeForce RTX 4090 Laptop 16 GB — offload — ~67 no Check
GeForce RTX 4080 Laptop 12 GB — offload — ~24 no Check
GeForce RTX 4070 Laptop 8 GB — offload — ~14 no Check

NVIDIA · Workstation

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
RTX 6000 Ada 48 GB Q8_0 Q8_0 256K 124 yes Check
RTX 5000 Ada 32 GB Q6_K Q6_K 256K 75 yes Check
RTX 4000 Ada 20 GB Q3_K_M Q3_K_M — — no Check
RTX 2000 Ada 16 GB — offload — ~32 no Check
RTX A6000 48 GB Q8_0 Q8_0 256K 99 yes Check
RTX A5000 24 GB Q4_K_M Q5_K_M 256K 99 tight Check
RTX A4000 16 GB — offload — ~56 no Check

NVIDIA · Data centre

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
L40S 48 GB Q8_0 Q8_0 256K 112 yes Check
L4 24 GB Q4_K_M Q5_K_M 256K 39 tight Check
A10 24 GB Q4_K_M Q5_K_M 256K 78 tight Check
A100 40 GB 40 GB Q8_0 Q8_0 256K 201 yes Check
A100 80 GB 80 GB Q8_0 Q8_0 256K 264 yes Check
H100 PCIe 80 GB Q8_0 Q8_0 256K 259 yes Check
H100 SXM 80 GB Q8_0 Q8_0 256K 433 yes Check
H200 SXM 141 GB Q8_0 Q8_0 256K 621 yes Check
Tesla P40 24 GB Q4_K_M Q5_K_M 256K 45 tight Check

AMD · Radeon

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
Radeon RX 9070 XT 16 GB — offload — ~72 no Check
Radeon RX 9070 16 GB — offload — ~72 no Check
Radeon RX 7900 XTX 24 GB Q4_K_M Q5_K_M 256K 124 tight Check
Radeon RX 7900 XT 20 GB Q3_K_M Q3_K_M — — no Check
Radeon RX 7900 GRE 16 GB — offload — ~67 no Check
Radeon RX 7800 XT 16 GB — offload — ~71 no Check
Radeon RX 7700 XT 12 GB — offload — ~24 no Check
Radeon RX 7600 XT 16 GB — offload — ~39 no Check
Radeon RX 6900 XT 16 GB — offload — ~62 no Check
Radeon RX 6800 XT 16 GB — offload — ~62 no Check
Radeon RX 6700 XT 12 GB — offload — ~23 no Check

AMD · Radeon Pro

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
Radeon Pro W7900 48 GB Q8_0 Q8_0 256K 112 yes Check
Radeon Pro W7800 32 GB Q6_K Q6_K 256K 75 yes Check

AMD · Instinct

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
Instinct MI210 64 GB Q8_0 Q8_0 256K 212 yes Check
Instinct MI300X 192 GB Q8_0 Q8_0 256K 686 yes Check

AMD · Ryzen AI Max

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
Ryzen AI Max+ 395 · 128 GB 128 GB Q8_0 Q8_0 256K 37 yes Check
Ryzen AI Max+ 395 · 64 GB 64 GB Q8_0 Q8_0 256K 37 yes Check

Apple · Apple Silicon

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
M1 · 16 GB 16 GB — no — — no Check
M1 Pro · 32 GB 32 GB Q4_K_M Q5_K_M 256K 29 yes Check
M1 Max · 64 GB 64 GB Q8_0 Q8_0 256K 58 yes Check
M1 Ultra · 128 GB 128 GB Q8_0 Q8_0 256K 116 yes Check
M2 · 24 GB 24 GB Q3_K_M Q3_K_M — — no Check
M2 Pro · 32 GB 32 GB Q4_K_M Q5_K_M 256K 29 yes Check
M2 Max · 96 GB 96 GB Q8_0 Q8_0 256K 58 yes Check
M2 Ultra · 192 GB 192 GB Q8_0 Q8_0 256K 116 yes Check
M3 · 24 GB 24 GB Q3_K_M Q3_K_M — — no Check
M3 Pro · 36 GB 36 GB Q5_K_M Q6_K 256K 22 yes Check
M3 Max · 64 GB 64 GB Q8_0 Q8_0 256K 58 yes Check
M3 Max · 128 GB 128 GB Q8_0 Q8_0 256K 58 yes Check
M3 Ultra · 256 GB 256 GB Q8_0 Q8_0 256K 116 yes Check
M3 Ultra · 512 GB 512 GB Q8_0 Q8_0 256K 116 yes Check
M4 · 32 GB 32 GB Q4_K_M Q5_K_M 256K 17 yes Check
M4 Pro · 48 GB 48 GB Q6_K Q8_0 256K 40 yes Check
M4 Pro · 64 GB 64 GB Q8_0 Q8_0 256K 40 yes Check
M4 Max · 64 GB 64 GB Q8_0 Q8_0 256K 59 yes Check
M4 Max · 128 GB 128 GB Q8_0 Q8_0 256K 79 yes Check

Intel · Arc

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
Arc B580 12 GB — offload — ~24 no Check
Arc B570 10 GB — offload — ~18 no Check
Arc A770 16 GB 16 GB — offload — ~66 no Check
Arc A750 8 GB — offload — ~16 no Check

CPU · CPU + system RAM

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
CPU only · DDR4 dual-channel 32 GB Q5_K_M Q6_K 256K 5.3 yes Check
CPU only · DDR5 dual-channel 64 GB Q8_0 Q8_0 256K 9.3 yes Check
CPU only · DDR5 8-channel server 256 GB Q8_0 Q8_0 256K 32 yes Check