Hardware requirements

What hardware do I need to run Olmo 3 7B Instruct?

7.3B 64K context Apache 2.0

At Q4_K_M the weights are 4.1 GB; with an 8K KV cache and runtime overhead that is 7.2 GB, so it wants a 10 GB card to run with headroom — in practice 10 GB VRAM: GeForce RTX 3080 and 1 other. 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 8 GB VRAM

GeForce RTX 3070 Ti — Q3_K_M, 6.4 GB of 7.0 GB, ~111 tok/s, up to 12K context.

Also: Arc A750, GeForce RTX 5060, GeForce RTX 3070, GeForce RTX 3060 Ti, GeForce RTX 4060 Ti +2 more
the smallest card that holds it at all, at Q3_K_M — the lowest quantisation worth running
Recommended 10 GB VRAM

GeForce RTX 3080 — Q4_K_M, 7.2 GB of 8.8 GB, ~112 tok/s, up to 20K context.

Also: Arc B570
runs the recommended quantisation with headroom at 8K context
High quality 16 GB VRAM

GeForce RTX 5080 — Q8_0, 10.3 GB of 14.4 GB, ~80 tok/s, up to 40K context.

Also: GeForce RTX 5070 Ti, GeForce RTX 4080 Super, GeForce RTX 4080, GeForce RTX 4070 Ti Super, Radeon RX 9070 XT +12 more
near-lossless Q8_0 with headroom
Long context 20 GB VRAM

Radeon RX 7900 XT — Q4_K_M, 14.2 GB of 18.4 GB, ~118 tok/s at 64K.

Also: RTX 4000 Ada
128K tokens (or the model’s maximum) at the recommended quantisation
Apple Silicon 16 GB unified

M1 · 16 GB — Q4_K_M, 7.2 GB of 10.7 GB, ~9.3 tok/s, up to 35K context.

smallest Mac that runs the recommended quantisation with headroom
CPU only 32 GB RAM

CPU only · DDR4 dual-channel — Q8_0, 10.3 GB of 25.6 GB, ~3.0 tok/s, up to 64K context.

no GPU — weights stream from system RAM

How much VRAM at each quantisation

QuantWeights+KV 8K+KV 32K+KV 128KTotal @ 8KNeedsQuality
F1613.6 GB2.50 GB5.50 GB9.5 GB 16.7 GB24 GB cardReference
Q8_07.2 GB2.50 GB5.50 GB9.5 GB 10.3 GB16 GB card−0.1% ppl
Q6_K5.6 GB2.50 GB5.50 GB9.5 GB 8.7 GB12 GB card−0.4% ppl
Q5_K_M4.8 GB2.50 GB5.50 GB9.5 GB 7.9 GB11 GB card−0.8% ppl
Q4_K_M4.1 GB2.50 GB5.50 GB9.5 GB 7.2 GB10 GB card−1.9% ppl
Q3_K_M3.3 GB2.50 GB5.50 GB9.5 GB 6.4 GB10 GB card−5.4% ppl
Q2_K2.8 GB2.50 GB5.50 GB9.5 GB 5.9 GB8 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 Q8_0 Q8_0 64K 264 yes Check
GeForce RTX 5080 16 GB Q8_0 Q8_0 64K 142 tight Check
GeForce RTX 5070 Ti 16 GB Q8_0 Q8_0 64K 132 tight Check
GeForce RTX 5070 12 GB Q6_K Q8_0 35K 99 no Check
GeForce RTX 5060 Ti 16 GB 16 GB Q8_0 Q8_0 64K 66 tight Check
GeForce RTX 5060 8 GB — Q3_K_M — — no Check
GeForce RTX 4090 24 GB Q8_0 Q8_0 64K 149 yes Check
GeForce RTX 4080 Super 16 GB Q8_0 Q8_0 64K 109 tight Check
GeForce RTX 4080 16 GB Q8_0 Q8_0 64K 106 tight Check
GeForce RTX 4070 Ti Super 16 GB Q8_0 Q8_0 64K 99 tight Check
GeForce RTX 4070 Ti 12 GB Q6_K Q8_0 35K 74 no Check
GeForce RTX 4070 Super 12 GB Q6_K Q8_0 35K 74 no Check
GeForce RTX 4070 12 GB Q6_K Q8_0 35K 74 no Check
GeForce RTX 4060 Ti 16 GB 16 GB Q8_0 Q8_0 64K 42 tight Check
GeForce RTX 4060 Ti 8 GB — Q3_K_M — — no Check
GeForce RTX 4060 8 GB — Q3_K_M — — no Check
GeForce RTX 3090 Ti 24 GB Q8_0 Q8_0 64K 149 yes Check
GeForce RTX 3090 24 GB Q8_0 Q8_0 64K 138 yes Check
GeForce RTX 3080 Ti 12 GB Q6_K Q8_0 35K 135 no Check
GeForce RTX 3080 12 GB 12 GB Q6_K Q8_0 35K 135 no Check
GeForce RTX 3080 10 GB Q4_K_M Q6_K 20K 112 no Check
GeForce RTX 3070 Ti 8 GB — Q3_K_M — — no Check
GeForce RTX 3070 8 GB — Q3_K_M — — no Check
GeForce RTX 3060 Ti 8 GB — Q3_K_M — — no Check
GeForce RTX 3060 12 GB 12 GB Q6_K Q8_0 35K 53 no Check
GeForce RTX 2080 Ti 11 GB Q5_K_M Q6_K 27K 91 no Check
GeForce RTX 2060 12 GB 12 GB Q6_K Q8_0 35K 50 no Check
GeForce GTX 1080 Ti 11 GB Q5_K_M Q6_K 27K 71 no Check

NVIDIA · GeForce laptop

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
GeForce RTX 5090 Laptop 24 GB Q8_0 Q8_0 64K 132 yes Check
GeForce RTX 4090 Laptop 16 GB Q8_0 Q8_0 64K 85 tight Check
GeForce RTX 4080 Laptop 12 GB Q6_K Q8_0 35K 64 no Check
GeForce RTX 4070 Laptop 8 GB — Q3_K_M — — 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 64K 142 yes Check
RTX 5000 Ada 32 GB Q8_0 Q8_0 64K 85 yes Check
RTX 4000 Ada 20 GB Q8_0 Q8_0 64K 53 yes Check
RTX 2000 Ada 16 GB Q8_0 Q8_0 64K 33 tight Check
RTX A6000 48 GB Q8_0 Q8_0 64K 113 yes Check
RTX A5000 24 GB Q8_0 Q8_0 64K 113 yes Check
RTX A4000 16 GB Q8_0 Q8_0 64K 66 tight Check

NVIDIA · Data centre

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
L40S 48 GB Q8_0 Q8_0 64K 127 yes Check
L4 24 GB Q8_0 Q8_0 64K 44 yes Check
A10 24 GB Q8_0 Q8_0 64K 88 yes Check
A100 40 GB 40 GB Q8_0 Q8_0 64K 229 yes Check
A100 80 GB 80 GB Q8_0 Q8_0 64K 301 yes Check
H100 PCIe 80 GB Q8_0 Q8_0 64K 295 yes Check
H100 SXM 80 GB Q8_0 Q8_0 64K 494 yes Check
H200 SXM 141 GB Q8_0 Q8_0 64K 708 yes Check
Tesla P40 24 GB Q8_0 Q8_0 64K 51 yes Check

AMD · Radeon

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
Radeon RX 9070 XT 16 GB Q8_0 Q8_0 64K 95 tight Check
Radeon RX 9070 16 GB Q8_0 Q8_0 64K 95 tight Check
Radeon RX 7900 XTX 24 GB Q8_0 Q8_0 64K 142 yes Check
Radeon RX 7900 XT 20 GB Q8_0 Q8_0 64K 118 yes Check
Radeon RX 7900 GRE 16 GB Q8_0 Q8_0 64K 85 tight Check
Radeon RX 7800 XT 16 GB Q8_0 Q8_0 64K 92 tight Check
Radeon RX 7700 XT 12 GB Q6_K Q8_0 35K 64 no Check
Radeon RX 7600 XT 16 GB Q8_0 Q8_0 64K 42 tight Check
Radeon RX 6900 XT 16 GB Q8_0 Q8_0 64K 76 tight Check
Radeon RX 6800 XT 16 GB Q8_0 Q8_0 64K 76 tight Check
Radeon RX 6700 XT 12 GB Q6_K Q8_0 35K 57 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 64K 127 yes Check
Radeon Pro W7800 32 GB Q8_0 Q8_0 64K 85 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 64K 242 yes Check
Instinct MI300X 192 GB Q8_0 Q8_0 64K 782 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 64K 35 yes Check
Ryzen AI Max+ 395 · 64 GB 64 GB Q8_0 Q8_0 64K 35 yes Check

Apple · Apple Silicon

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
M1 · 16 GB 16 GB Q6_K Q8_0 35K 9.3 no Check
M1 Pro · 32 GB 32 GB Q8_0 Q8_0 64K 27 yes Check
M1 Max · 64 GB 64 GB Q8_0 Q8_0 64K 54 yes Check
M1 Ultra · 128 GB 128 GB Q8_0 Q8_0 64K 109 yes Check
M2 · 24 GB 24 GB Q8_0 Q8_0 64K 14 yes Check
M2 Pro · 32 GB 32 GB Q8_0 Q8_0 64K 27 yes Check
M2 Max · 96 GB 96 GB Q8_0 Q8_0 64K 54 yes Check
M2 Ultra · 192 GB 192 GB Q8_0 Q8_0 64K 109 yes Check
M3 · 24 GB 24 GB Q8_0 Q8_0 64K 14 yes Check
M3 Pro · 36 GB 36 GB Q8_0 Q8_0 64K 20 yes Check
M3 Max · 64 GB 64 GB Q8_0 Q8_0 64K 54 yes Check
M3 Max · 128 GB 128 GB Q8_0 Q8_0 64K 54 yes Check
M3 Ultra · 256 GB 256 GB Q8_0 Q8_0 64K 109 yes Check
M3 Ultra · 512 GB 512 GB Q8_0 Q8_0 64K 109 yes Check
M4 · 32 GB 32 GB Q8_0 Q8_0 64K 16 yes Check
M4 Pro · 48 GB 48 GB Q8_0 Q8_0 64K 37 yes Check
M4 Pro · 64 GB 64 GB Q8_0 Q8_0 64K 37 yes Check
M4 Max · 64 GB 64 GB Q8_0 Q8_0 64K 56 yes Check
M4 Max · 128 GB 128 GB Q8_0 Q8_0 64K 74 yes Check

Intel · Arc

DeviceMemoryBest quant with headroomFits at allMax context @ Q4_K_MTok/s est.128K
Arc B580 12 GB Q6_K Q8_0 35K 67 no Check
Arc B570 10 GB Q4_K_M Q6_K 20K 56 no Check
Arc A770 16 GB 16 GB Q8_0 Q8_0 64K 83 tight Check
Arc A750 8 GB — Q3_K_M — — 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 Q8_0 Q8_0 64K 5.2 yes Check
CPU only · DDR5 dual-channel 64 GB Q8_0 Q8_0 64K 9.2 yes Check
CPU only · DDR5 8-channel server 256 GB Q8_0 Q8_0 64K 31 yes Check