GPUs / NVIDIA

RTX 3090

Memory
GDDR6X
Bus
384-bit
Published peak
936 GB/s
Capacity
24 GB

Measured

Read
831.1GB/s
Write
809.8GB/s
Copy
812.4GB/s
Random
69.7GB/s
Latency
270ns
Sustain
812.4GB/s

Medians of 1 verified run from 1 operator. Read reaches 89 % of the published peak.

Local AI speed

ModelParameters4-bit, tokens/s8-bit, tokens/s
Llama 3.1 8B8.03B≤ 207≤ 103
Qwen 2.5 14B14.7B≤ 113≤ 57
gpt-oss-20b21B, 3.6B active≤ 462≤ 231
Mistral Small 24B24B≤ 69≤ 35
Qwen 2.5 32B32.5B≤ 51Does 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 24 GB configuration; the KV cache needs room on top.

Top operators

#OperatorScoreReadDate
1OP-33B96291.8831.1 GB/s2026-09-27

Full RTX 3090 leaderboard