Loose-Info.com
Last Update 2026/02/17
TOP - 各種テスト - LLM - ローカルLLMの実測値比較 Llama 3.2 [英語プロンプト]

低スペック寄りのPCでローカルLLMを動作させた際の記録です。
LLM以外の仮想マシンなどが起動され、多少負荷がかかった状態で実行しています。
ベンチマークなどでLLMの性能を評価する内容ではありません。

検証用PC

OS

Debian GNU/Linux 12 (bookworm)

CPU

Intel(R) Core(TM) i5-14400F

GPU

GeForce RTX 3060 12GB

メモリ

DDR4 PC4-25600 32GB × 4

SSD

crucial P310 CT1000P310SSD8-JP


構築環境 : Docker + Ollama (特別な設定などは無い状態)

検証用プロンプト

Could you please recommend some great places in the US to see beautiful scenery? Around 10 places in all four directions.

Llama 3.2 [英語プロンプト]

GPU無し
1b-instruct-q4_K_M(45.8TPS)   1b-instruct-q5_K_M(41.3TPS)   1b-instruct-q8_0(30.3TPS)   1b-instruct-fp16(16.9TPS)
3b-instruct-q4_K_M(18.8TPS)   3b-instruct-q5_K_M(16.5TPS)   3b-instruct-q8_0(11.9TPS)   3b-instruct-fp16(6.54TPS)  
GPU使用
1b-instruct-q4_K_M(303TPS)   1b-instruct-q5_K_M(278TPS)   1b-instruct-q8_0(210TPS)   1b-instruct-fp16(124TPS)
3b-instruct-q4_K_M(129TPS)   3b-instruct-q5_K_M(117TPS)   3b-instruct-q8_0(86.9TPS)   3b-instruct-fp16(49.5TPS)  

・TPS(tokens/s) は eval_count / eval_duration により算出
・モデルロード済みの検証は省略

llama3.2:1b-instruct-q4_K_M(GPU無し)

Model architecture llama parameters 1.2B context length 131072 embedding length 2048 quantization Q4_K_M 2026-02-16 total_duration(合計時間) : 11160466685 (11.160s) load_duration(モデルのロード時間) : 847985303 ( 0.848s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 304720629 ( 0.848s) eval_count(生成トークン数) : 446 eval_duration(生成時間) : 9743731034 ( 9.744s) real 0m11.169s user 0m0.027s sys 0m0.004s メモリ使用量(RSS) : 1034732 KB

llama3.2:1b-instruct-q5_K_M(GPU無し)

Model architecture llama parameters 1.2B context length 131072 embedding length 2048 quantization Q5_K_M 2026-02-16 total_duration(合計時間) : 10392110807 (10.392s) load_duration(モデルのロード時間) : 793192947 ( 0.793s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 315504314 ( 0.316s) eval_count(生成トークン数) : 373 eval_duration(生成時間) : 9024002528 ( 9.024s) real 0m10.403s user 0m0.025s sys 0m0.006s メモリ使用量(RSS) : 1143640 KB

llama3.2:1b-instruct-q8_0(GPU無し)

Model architecture llama parameters 1.2B context length 131072 embedding length 2048 quantization Q8_0 2026-02-16 total_duration(合計時間) : 14223443411 (14.223s) load_duration(モデルのロード時間) : 772590835 ( 0.773s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 189872136 ( 0.190s) eval_count(生成トークン数) : 394 eval_duration(生成時間) : 12997048967 (12.997s) real 0m14.234s user 0m0.025s sys 0m0.006s メモリ使用量(RSS) : 1533476 KB

llama3.2:1b-instruct-fp16(GPU無し)

Model architecture llama parameters 1.2B context length 131072 embedding length 2048 quantization F16 2026-02-16 total_duration(合計時間) : 22077110671 (22.077s) load_duration(モデルのロード時間) : 1048190912 ( 1.048s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 303637508 ( 0.304s) eval_count(生成トークン数) : 346 eval_duration(生成時間) : 20522973918 (20.523s) real 0m22.088s user 0m0.020s sys 0m0.012s メモリ使用量(RSS) : 2660904 KB

llama3.2:3b-instruct-q4_K_M(GPU無し)

Model architecture llama parameters 3.2B context length 131072 embedding length 3072 quantization Q4_K_M 2026-02-16 total_duration(合計時間) : 25071609188 (25.072s) load_duration(モデルのロード時間) : 1070302959 ( 1.070s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 669983267 ( 0.670s) eval_count(生成トークン数) : 432 eval_duration(生成時間) : 23030544016 (23.031s) real 0m25.082s user 0m0.024s sys 0m0.009s メモリ使用量(RSS) : 2544000 KB

llama3.2:3b-instruct-q5_K_M(GPU無し)

Model architecture llama parameters 3.2B context length 131072 embedding length 3072 quantization Q5_K_M 2026-02-16 total_duration(合計時間) : 30083680188 (30.084s) load_duration(モデルのロード時間) : 1068052723 ( 1.068s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 954498736 ( 0.954s) eval_count(生成トークン数) : 458 eval_duration(生成時間) : 27742392216 (27.742s) real 0m30.094s user 0m0.021s sys 0m0.011s メモリ使用量(RSS) : 2840260 KB

llama3.2:3b-instruct-q8_0(GPU無し)

Model architecture llama parameters 3.2B context length 131072 embedding length 3072 quantization Q8_0 2026-02-16 total_duration(合計時間) : 44373719247 (44.374s) load_duration(モデルのロード時間) : 1298096668 ( 1.298s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 528690690 ( 0.529s) eval_count(生成トークン数) : 501 eval_duration(生成時間) : 42198202232 (42.198s) real 0m44.385s user 0m0.025s sys 0m0.015s メモリ使用量(RSS) : 3918908 KB

llama3.2:3b-instruct-fp16(GPU無し)

Model architecture llama parameters 3.2B context length 131072 embedding length 3072 quantization F16 2026-02-16 total_duration(合計時間) : 71134149187 (71.134s) load_duration(モデルのロード時間) : 1528063840 ( 1.528s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 841500698 ( 0.842s) eval_count(生成トークン数) : 448 eval_duration(生成時間) : 68450339042 (68.450s) real 1m11.145s user 0m0.020s sys 0m0.018s メモリ使用量(RSS) : 6855184 KB

llama3.2:1b-instruct-q4_K_M(GPU使用)

Model architecture llama parameters 1.2B context length 131072 embedding length 2048 quantization Q4_K_M 2026-02-16 total_duration(合計時間) : 3061415092 (3.061s) load_duration(モデルのロード時間) : 1140996439 (1.141s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 12823968 (0.013s) eval_count(生成トークン数) : 481 eval_duration(生成時間) : 1587401058 (1.587s) real 0m3.073s user 0m0.032s sys 0m0.001s +---------------------------------------------------------------------------------------+ | NVIDIA-SMI 535.261.03 Driver Version: 535.261.03 CUDA Version: 12.2 | |-----------------------------------------+----------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+======================+======================| | 0 NVIDIA GeForce RTX 3060 On | 00000000:01:00.0 On | N/A | | 0% 38C P2 165W / 170W | 1605MiB / 12288MiB | 83% Default | | | | N/A | +-----------------------------------------+----------------------+----------------------+ +---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| | 0 N/A N/A 1242 G /usr/lib/xorg/Xorg 119MiB | | 0 N/A N/A 1899 G xfwm4 2MiB | | 0 N/A N/A 2423 G /usr/bin/x-www-browser 144MiB | | 0 N/A N/A 36875 C /usr/bin/ollama 1326MiB | +---------------------------------------------------------------------------------------+ メモリ使用量(RSS) : 639336 KB

llama3.2:1b-instruct-q5_K_M(GPU使用)

Model architecture llama parameters 1.2B context length 131072 embedding length 2048 quantization Q5_K_M 2026-02-16 total_duration(合計時間) : 3502409051 (3.502s) load_duration(モデルのロード時間) : 887488583 (0.887s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 13636906 (0.014s) eval_count(生成トークン数) : 608 eval_duration(生成時間) : 2183891028 (2.184s) real 0m3.513s user 0m0.030s sys 0m0.000s +---------------------------------------------------------------------------------------+ | NVIDIA-SMI 535.261.03 Driver Version: 535.261.03 CUDA Version: 12.2 | |-----------------------------------------+----------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+======================+======================| | 0 NVIDIA GeForce RTX 3060 On | 00000000:01:00.0 On | N/A | | 0% 41C P2 165W / 170W | 1703MiB / 12288MiB | 85% Default | | | | N/A | +-----------------------------------------+----------------------+----------------------+ +---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| | 0 N/A N/A 1242 G /usr/lib/xorg/Xorg 119MiB | | 0 N/A N/A 1899 G xfwm4 2MiB | | 0 N/A N/A 2423 G /usr/bin/x-www-browser 144MiB | | 0 N/A N/A 36946 C /usr/bin/ollama 1424MiB | +---------------------------------------------------------------------------------------+ メモリ使用量(RSS) : 641900 KB

llama3.2:1b-instruct-q8_0(GPU使用)

Model architecture llama parameters 1.2B context length 131072 embedding length 2048 quantization Q8_0 2026-02-16 total_duration(合計時間) : 3267031012 (3.267s) load_duration(モデルのロード時間) : 873931484 (0.874s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 12435959 (0.012s) eval_count(生成トークン数) : 440 eval_duration(生成時間) : 2093887745 (2.094s) real 0m3.278s user 0m0.025s sys 0m0.005s +---------------------------------------------------------------------------------------+ | NVIDIA-SMI 535.261.03 Driver Version: 535.261.03 CUDA Version: 12.2 | |-----------------------------------------+----------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+======================+======================| | 0 NVIDIA GeForce RTX 3060 On | 00000000:01:00.0 On | N/A | | 0% 43C P2 150W / 170W | 2095MiB / 12288MiB | 89% Default | | | | N/A | +-----------------------------------------+----------------------+----------------------+ +---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| | 0 N/A N/A 1242 G /usr/lib/xorg/Xorg 119MiB | | 0 N/A N/A 1899 G xfwm4 2MiB | | 0 N/A N/A 2423 G /usr/bin/x-www-browser 144MiB | | 0 N/A N/A 37019 C /usr/bin/ollama 1816MiB | +---------------------------------------------------------------------------------------+ メモリ使用量(RSS) : 693328 KB

llama3.2:1b-instruct-fp16(GPU使用)

Model architecture llama parameters 1.2B context length 131072 embedding length 2048 quantization F16 2026-02-16 total_duration(合計時間) : 7787854514 (7.788s) load_duration(モデルのロード時間) : 1155829205 (1.156s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 53479837 (0.053s) eval_count(生成トークン数) : 754 eval_duration(生成時間) : 6086524249 (6.087s) real 0m7.799s user 0m0.023s sys 0m0.007s +---------------------------------------------------------------------------------------+ | NVIDIA-SMI 535.261.03 Driver Version: 535.261.03 CUDA Version: 12.2 | |-----------------------------------------+----------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+======================+======================| | 0 NVIDIA GeForce RTX 3060 On | 00000000:01:00.0 On | N/A | | 0% 46C P2 148W / 170W | 3289MiB / 12288MiB | 92% Default | | | | N/A | +-----------------------------------------+----------------------+----------------------+ +---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| | 0 N/A N/A 1242 G /usr/lib/xorg/Xorg 119MiB | | 0 N/A N/A 1899 G xfwm4 2MiB | | 0 N/A N/A 2423 G /usr/bin/x-www-browser 144MiB | | 0 N/A N/A 37090 C /usr/bin/ollama 3010MiB | +---------------------------------------------------------------------------------------+ メモリ使用量(RSS) : 1028096 KB

llama3.2:3b-instruct-q4_K_M(GPU使用)

Model architecture llama parameters 3.2B context length 131072 embedding length 3072 quantization Q4_K_M 2026-02-16 total_duration(合計時間) : 4505531008 (4.506s) load_duration(モデルのロード時間) : 1143182015 (1.143s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 22459426 (0.022s) eval_count(生成トークン数) : 398 eval_duration(生成時間) : 3074794246 (3.075s) real 0m4.517s user 0m0.020s sys 0m0.011s +---------------------------------------------------------------------------------------+ | NVIDIA-SMI 535.261.03 Driver Version: 535.261.03 CUDA Version: 12.2 | |-----------------------------------------+----------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+======================+======================| | 0 NVIDIA GeForce RTX 3060 On | 00000000:01:00.0 On | N/A | | 0% 43C P2 169W / 170W | 3113MiB / 12288MiB | 91% Default | | | | N/A | +-----------------------------------------+----------------------+----------------------+ +---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| | 0 N/A N/A 1242 G /usr/lib/xorg/Xorg 119MiB | | 0 N/A N/A 1899 G xfwm4 2MiB | | 0 N/A N/A 2423 G /usr/bin/x-www-browser 144MiB | | 0 N/A N/A 57066 C /usr/bin/ollama 2834MiB | +---------------------------------------------------------------------------------------+ メモリ使用量(RSS) : 752148 KB

llama3.2:3b-instruct-q5_K_M(GPU使用)

Model architecture llama parameters 3.2B context length 131072 embedding length 3072 quantization Q5_K_M 2026-02-16 total_duration(合計時間) : 4820130999 (4.820s) load_duration(モデルのロード時間) : 1132322462 (1.132s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 24964855 (0.025s) eval_count(生成トークン数) : 396 eval_duration(生成時間) : 3389919420 (3.390s) real 0m4.825s user 0m0.015s sys 0m0.004s +---------------------------------------------------------------------------------------+ | NVIDIA-SMI 535.261.03 Driver Version: 535.261.03 CUDA Version: 12.2 | |-----------------------------------------+----------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+======================+======================| | 0 NVIDIA GeForce RTX 3060 On | 00000000:01:00.0 On | N/A | | 0% 46C P2 168W / 170W | 3401MiB / 12288MiB | 92% Default | | | | N/A | +-----------------------------------------+----------------------+----------------------+ +---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| | 0 N/A N/A 1242 G /usr/lib/xorg/Xorg 119MiB | | 0 N/A N/A 1899 G xfwm4 2MiB | | 0 N/A N/A 2423 G /usr/bin/x-www-browser 144MiB | | 0 N/A N/A 57137 C /usr/bin/ollama 3122MiB | +---------------------------------------------------------------------------------------+ メモリ使用量(RSS) : 748272 KB

llama3.2:3b-instruct-q8_0(GPU使用)

Model architecture llama parameters 3.2B context length 131072 embedding length 3072 quantization Q8_0 2026-02-16 total_duration(合計時間) : 9131036373 (9.131s) load_duration(モデルのロード時間) : 1138349756 (1.138s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 25013479 (0.025s) eval_count(生成トークン数) : 655 eval_duration(生成時間) : 7539457693 (7.539s) real 0m9.141s user 0m0.020s sys 0m0.009s +---------------------------------------------------------------------------------------+ | NVIDIA-SMI 535.261.03 Driver Version: 535.261.03 CUDA Version: 12.2 | |-----------------------------------------+----------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+======================+======================| | 0 NVIDIA GeForce RTX 3060 On | 00000000:01:00.0 On | N/A | | 0% 49C P2 165W / 170W | 4455MiB / 12288MiB | 95% Default | | | | N/A | +-----------------------------------------+----------------------+----------------------+ +---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| | 0 N/A N/A 1242 G /usr/lib/xorg/Xorg 125MiB | | 0 N/A N/A 1899 G xfwm4 2MiB | | 0 N/A N/A 2423 G /usr/bin/x-www-browser 144MiB | | 0 N/A N/A 57204 C /usr/bin/ollama 4170MiB | +---------------------------------------------------------------------------------------+ メモリ使用量(RSS) : 839420 KB

llama3.2:3b-instruct-fp16(GPU使用)

Model architecture llama parameters 3.2B context length 131072 embedding length 3072 quantization F16 2026-02-16 total_duration(合計時間) : 9865138571 (9.865s) load_duration(モデルのロード時間) : 1399273409 (1.399s) prompt_eval_count(評価されたプロンプトのトークン数) : 49 prompt_eval_duration(プロンプトの評価時間) : 36981312 (0.037s) eval_count(生成トークン数) : 404 eval_duration(生成時間) : 8160622041 (8.161s) real 0m9.876s user 0m0.029s sys 0m0.004s +---------------------------------------------------------------------------------------+ | NVIDIA-SMI 535.261.03 Driver Version: 535.261.03 CUDA Version: 12.2 | |-----------------------------------------+----------------------+----------------------+ | GPU Name Persistence-M | Bus-Id Disp.A | Volatile Uncorr. ECC | | Fan Temp Perf Pwr:Usage/Cap | Memory-Usage | GPU-Util Compute M. | | | | MIG M. | |=========================================+======================+======================| | 0 NVIDIA GeForce RTX 3060 On | 00000000:01:00.0 On | N/A | | 0% 53C P2 156W / 170W | 7395MiB / 12288MiB | 96% Default | | | | N/A | +-----------------------------------------+----------------------+----------------------+ +---------------------------------------------------------------------------------------+ | Processes: | | GPU GI CI PID Type Process name GPU Memory | | ID ID Usage | |=======================================================================================| | 0 N/A N/A 1242 G /usr/lib/xorg/Xorg 119MiB | | 0 N/A N/A 1899 G xfwm4 2MiB | | 0 N/A N/A 2423 G /usr/bin/x-www-browser 144MiB | | 0 N/A N/A 57274 C /usr/bin/ollama 7116MiB | +---------------------------------------------------------------------------------------+ メモリ使用量(RSS) : 1294568 KB