GPUs / Apple

Apple M3 Pro

Memory
LPDDR5
Bus
192-bit
Published peak
150 GB/s
Capacity
18 GB

Measured

Read
128.2GB/s
Write
120.5GB/s
Copy
124.4GB/s
Random
19.9GB/s
Latency
364ns
Sustain
123.7GB/s

Medians of 1 verified run from 1 operator. Read reaches 85 % of the published peak.

Local AI speed

ModelParameters4-bit, tokens/s8-bit, tokens/s
Llama 3.1 8B8.03B≤ 32≤ 16
Qwen 2.5 14B14.7B≤ 17≤ 8.7
gpt-oss-20b21B, 3.6B active≤ 71Does not fit
Mistral Small 24B24B≤ 11Does not fit
Qwen 2.5 32B32.5B≤ 7.9Does not fit
Llama 3.3 70B70.6BDoes not fitDoes not fit
gpt-oss-120b117B, 5.1B activeDoes not fitDoes not fit
Upper bounds: median read ÷ bytes read per token. Mixture-of-experts models read only their active experts. Fit checks weights against the 18 GB configuration; the KV cache needs room on top.

Top operators

#OperatorScoreReadDate
1OP-1731D554.7128.2 GB/s2026-09-27

Full Apple M3 Pro leaderboard