01
GPU card
6,288
112.5
48 GB
ModelDock / model benchmark
Qwen3.8 27B local hardware benchmarks
Compare Qwen3.8 27B reading / prefill and single-stream writing / decode performance across source-backed hardware and clearly labelled projections. Model profile: 27B.
ModelQwen3.8 27B27B · Dataset as of 2026-10-15
Hardware × model dataset
Published Qwen3.8 27B hardware performance
02
GPU card
3,571.4
111.8
32 GB
03
GPU card
7,756
109.3
96 GB
04
GPU card
2,238.2
97
32 GB
05
Mac Studio M5 Ultra 256 GBEstimate
GPU card
~1,075
~91
256 GB
06
Mac Studio M5 Ultra 96 GBEstimate
GPU card
~1,075
~91
96 GB
07
GPU card
1,458.3
69.8
24 GB
08
GPU card
723
65.8
32 GB
09
GPU card
2,692.3
65.3
24 GB
10
GPU card
745
59.2
128 GB
11
Mac Studio M5 Max 128 GBEstimate
GPU card
~745
~59.2
128 GB
12
Mac Studio M5 Max 64 GBEstimate
GPU card
~745
~59.2
64 GB
13
GPU card
705
53.2
24 GB
14
GPU card
1,555.6
50.8
16 GB
15
GPU card
877.8
47.2
24 GB
16
GPU card
424.8
46.3
32 GB
17
GPU card
462.5
44.3
512 GB
18
GPU card
1,225.5
44
128 GB
19
GPU card
1,750
43.6
16 GB
20
GPU card
282.1
38.3
64 GB
21
GPU card
1,482.8
36.8
64 GB
22
GPU card
563.8
35.2
32 GB
23
GPU card
925.9
28.3
16 GB
24
GPU card
779.5
27.4
16 GB
25
GPU card
300
24.2
128 GB
26
GPU card
209.4
14
48 GB
Reading / prefill
Prefill · tok/s
Writing / decode
Decode · single stream · tok/s
Evidence
Rows use the same hardware × model projection as the live Benchmark and Ranking interfaces. Estimates remain visibly marked.