本地化运行大模型测试
用笔记本试着跑了一下qwen3 8B模型，速度感人；deepseek-r1:8b-0528-qwen3-q8_0
速度大概 4.0 tokens/s左右，不插电掉到1.6 tokens/s左右
基本不可用，一个字一个字的蹦…
硬件
华为笔记本matebook 13s
处理器：Intel Core i5 11300H @ 3.10GHz 睿频4.0GHz
内存：LPDDR4 16G 3733 MH
软件
操作系统：Microsoft Windows 11 Home , Version 23H2,
软件：Ollama 0.9.2
运行 deepseek-r1:8b-0528-qwen3-q8_0
模型deepseek-r1:8b
速度大概 4.0 tokens/s左右，基本不可用，一个字一个字的蹦…
total duration: 42.9621588s
load duration: 18.0573ms
prompt eval count: 15 token(s)
prompt eval duration: 1.1304521s
prompt eval rate: 13.27 tokens/s
eval count: 173 token(s)
eval duration: 41.8129173s
eval rate: 4.14 tokens/s
插电（正常）
3.7-3.9GHz，60%-70%占用
total duration: 4m4.9069496s
load duration: 19.5485ms
prompt eval count: 52 token(s)
prompt eval duration: 2.653565s
prompt eval rate: 19.60 tokens/s
eval count: 983 token(s)
eval duration: 4m2.2065187s
eval rate: 4.06 tokens/s
低优先级，插电
3.7-3.9GHz，60%-70%占用
total duration: 2m34.7337782s
load duration: 21.9906ms
prompt eval count: 1430 token(s)
prompt eval duration: 50.8199411s
prompt eval rate: 28.14 tokens/s
eval count: 415 token(s)
eval duration: 1m43.8479937s
eval rate: 4.00 tokens/s
未插电，节能模式
1.3GHz，20%占用
total duration: 13m43.4436181s
load duration: 74.8943ms
prompt eval count: 685 token(s)
prompt eval duration: 2m1.6321035s
prompt eval rate: 5.63 tokens/s
eval count: 1149 token(s)
eval duration: 11m41.6676697s
eval rate: 1.64 tokens/s
模型 qwen3:4b-fp16
速度略快一点，大概4.6 tokens/s，基本不可用，一个字一个字的蹦…
插电，低优先级
3.7-3.9GHz，60%-70%占用
total duration: 4m24.0887835s
load duration: 19.7498ms
prompt eval count: 14 token(s)
prompt eval duration: 707.2566ms
prompt eval rate: 19.79 tokens/s
eval count: 1223 token(s)
eval duration: 4m23.3607047s
eval rate: 4.64 tokens/s
total duration: 3m52.060574s
load duration: 9.0139398s
prompt eval count: 784 token(s)
prompt eval duration: 28.3682621s
prompt eval rate: 27.64 tokens/s
eval count: 870 token(s)
eval duration: 3m14.6603767s
eval rate: 4.47 tokens/s
DMI Processor
manufacturer Intel(R) Corporation
model 11th Gen Intel(R) Core(TM) i5-11300H @ 3.10GHz
clock speed 3100.0 MHz
FSB speed 100.0 MHz
multiplier 31.0x
max clock speed 4400.0 MHz
Windows Version Microsoft Windows 11 Home China (x64), Version 23H2, Build 22631.5472
Windows Installation Date 9/4/2022
DirectX Version 12.0
Number of cores 4 (max 4)
Number of threads 8 (max 8)
Manufacturer GenuineIntel
Name Intel Core i5 11300H
Codename Tiger Lake-U
Specification 11th Gen Intel(R) Core(TM) i5-11300H @ 3.10GHz
Package (platform ID) Socket 1449 FCBGA (0x7)
CPUID 6.C.1
Extended CPUID 6.8C
Core Stepping B1
Technology 10 nm
TDP Limit 35.0 Watts
Tjmax 100.0 癈
Core Speed 1397.8 MHz
Multiplier x Bus Speed 14.0 x 99.8 MHz
Base frequency (cores) 99.8 MHz
Stock frequency 3100 MHz
Max frequency 4400 MHz
Instructions sets MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, EM64T, VT-x, AES, AVX, AVX2, AVX512 (DQ, BW, VL, CD, IFMA, VBMI, VBMI2, VNNI, BITALG, VPOPCNTDQ, VP2INTERSECT), FMA3, SHA
Microcode Revision 0x8A
L1 Data cache 4 x 48 KB (12-way, 64-byte line)
L1 Instruction cache 4 x 32 KB (8-way, 64-byte line)
L2 cache 4 x 1.25 MB (20-way, 64-byte line)
L3 cache 8 MB (8-way, 64-byte line)
Max CPUID level 0000001Bh
Max CPUID ext. level 80000008h
FID/VID Control yes
Turbo Mode supported, enabled
Max non-turbo ratio 31x
Max turbo ratio 44x
Max efficiency ratio 4x
Min operating ratio 4x
Speedshift Autonomous
O/C bins none
Power Max (PL1) 45.00 W
PL1 Time Window 28.00 s
Short Power Max (PL2) 64.00 W
Max Peak Power (PL4) 121.00 W
Ratio 1 core 44x
Ratio 2 cores 44x
Ratio 3 cores 40x
Ratio 4 cores 40x
Ratio 5 cores 40x
Ratio 6 cores 40x
Ratio 7 cores 40x
Ratio 8 cores 40x
TDP Level 35.0 W @ 31x
TDP Level 28.0 W @ 26x
DMI Physical Memory Array
location Motherboard
usage System Memory
correction None
max capacity 16 GB
max# of devices 8
DMI Memory Device
designation ChannelA-DIMM0
format Row of chips
type LPDDR4
total width 64 bits
data width 64 bits
size 2 GB
speed 3733 MHz
manufacturer Micron Technology
part number 53E1G32D2NP-046
serial number 00000000
voltage 0.600000
manufacturer id 0x2C00
product id 0x0