TEST #4198 RESULTS
09/22/2026 - 7:36 PM
ACCELERATOR
MODEL
106
tokens/s
generation
341
ms
time to first token
3786
tokens/s
prompt
1056
LocalScore
HOW YOU STACK UP
Explore All ResultsMeta Llama 3.1 8B Instruct - Q4_K - Medium
SYSTEM
CPU
AMD Ryzen Threadripper PRO 3975WX 32-Cores (znver2)
RAM
377.6GB
OS
Linux
Kernel Release
6.12.107+deb13-amd64
Architecture
x86_64
Version
Cosmopolitan 3.9.7 MODE=x86_64; #1 SMP PREEMPT_DYNAMIC Debian 6.12.107-1 (2026-08-29)
RUNTIME
Name
llamafile
Version
0.9.2
Commit Hash
a30b324
DETAILED RESULTS
TEST NAME
PROMPT
GENERATION
TTFT
pp1024+tg16
4495
tokens/s
109
tokens/s
237
ms
pp4096+tg256
3620
tokens/s
95.9
tokens/s
1.14
sec
pp2048+tg256
4182
tokens/s
104
tokens/s
499
ms
pp2048+tg768
4177
tokens/s
104
tokens/s
500
ms
pp1024+tg1024
4497
tokens/s
108
tokens/s
237
ms
pp1280+tg3072
4339
tokens/s
101
tokens/s
304
ms
pp384+tg1152
4589
tokens/s
109
tokens/s
93
ms
pp64+tg1024
3242
tokens/s
112
tokens/s
28
ms
pp16+tg1536
936
tokens/s
110
tokens/s
25
ms