Skip to content

Instantly share code, notes, and snippets.

@tvararu
Last active December 2, 2025 03:42
Show Gist options
  • Select an option

  • Save tvararu/aef4e2da6580ff965bd122bcbd4b8ff0 to your computer and use it in GitHub Desktop.

Select an option

Save tvararu/aef4e2da6580ff965bd122bcbd4b8ff0 to your computer and use it in GitHub Desktop.

Benchmarks

Results from testing my new Omarchy server. Open test bench, no case yet.

specs

Component Model
CPU AMD Ryzen 7 9800X3D
AIO Cooler Master 240 Core II
MBD Gigabyte B850I Aorus Pro
RAM Crucial Pro 128GB (2x64GB) DDR5-5600 CL46
SSD Samsung 2TB 990 EVO PLUS
GPU MSI RTX 5090 Ventus 3X OC
PSU Corsair SF1000
MON UPERFECT UColor O2 16" 2880x1800 OLED 120Hz

results

Benchmark Score Unit
geekbench6 3440 single
geekbench6 18769 multi
cyberpunk 2077 83 fps
wan 2.2 t2v 40 sec
sysbench cpu 49752 ops/s
sysbench memory 29600 MiB/s
gemma3:4b 302 tok/s
gemma3:27b 77 tok/s
qwen3-coder:30b 241 tok/s
gpt-oss:20b 255 tok/s

geekbench6

https://browser.geekbench.com/v6/cpu/15307954

  • CPU: AMD Ryzen 7 9800X3D (8c/16t @ 5.27 GHz)
  • Board: Gigabyte B850I AORUS PRO
  • RAM: 123 GB
  • Geekbench 6.5.0
Test Single Multi
Overall 3440 18769

stress test

5 min combined CPU + GPU stress (open test bench, no case)

stress-ng --cpu $(nproc) --timeout 5m & gpu_burn 300

tuned = CPU -35 Curve Optimizer, GPU 480W power limit (LACT)

Config CPU Temp CPU Power GPU Temp GPU Power
Stock 92°C 128W 75°C 572W
Tuned 72°C 100W 71°C 480W

cyberpunk 2077

2880x1800 (62.5% of 4K, 140% of 1440p)

Preset Run Avg Min Max
RT: Overdrive 1 83.17 73.77 92.87
RT: Overdrive 2 82.71 73.78 91.79
RT: Ultra 1 122.83 109.13 139.11
RT: Ultra 2 122.60 108.27 138.82

comfyui

video_wan2_2_14B_t2v.json (default)

Run Time
1 47.62s
2 39.80s
3 40.40s

sysbench

$ sysbench cpu --threads=$(nproc) run
CPU speed:
    events per second: 49752.26

General statistics:
    total time:                          10.0003s
    total number of events:              497564

Latency (ms):
         min:                                    0.16
         avg:                                    0.32
         max:                                   12.82
         95th percentile:                        0.32
$ sysbench memory --threads=$(nproc) run
Total operations: 104857600 (30310208.21 per second)

102400.00 MiB transferred (29599.81 MiB/sec)

General statistics:
    total time:                          3.4589s
    total number of events:              104857600

Latency (ms):
         min:                                    0.00
         avg:                                    0.00
         max:                                   12.68
         95th percentile:                        0.00

ollama

$ ollama run gemma3:4b --verbose "Write a haiku"
Green leaves softly sway,
Sunlight dances on the breeze,
Peaceful, quiet day.

total duration:       185.012485ms
load duration:        104.40441ms
prompt eval count:    13 token(s)
prompt eval duration: 3.990041ms
prompt eval rate:     3258.11 tokens/s
eval count:           21 token(s)
eval duration:        69.511282ms
eval rate:            302.11 tokens/s
$ ollama run gemma3:27b --verbose "Write a haiku"
Green leaves softly sway,
Sunlight dances on the breeze,
Summer's gentle kiss.

total duration:       418.834894ms
load duration:        105.766998ms
prompt eval count:    13 token(s)
prompt eval duration: 34.768012ms
prompt eval rate:     373.91 tokens/s
eval count:           21 token(s)
eval duration:        272.539731ms
eval rate:            77.05 tokens/s
$ ollama run qwen3-coder:30b --verbose "Write a haiku"
Cherry blossoms fall
Soft petals dance on warm breeze
Spring's gentle farewell

total duration:       128.89472ms
load duration:        41.735052ms
prompt eval count:    13 token(s)
prompt eval duration: 5.566987ms
prompt eval rate:     2335.19 tokens/s
eval count:           19 token(s)
eval duration:        78.643085ms
eval rate:            241.60 tokens/s
$ ollama run gpt-oss:20b --verbose "Write a haiku"
Thinking...
We need to write a haiku. Haiku is a 5-7-5 syllable Japanese poetic form, typically with nature. We'll produce a haiku. The user just says "Write a haiku". So we give a short poem. Let's produce a nature-themed haiku, 5-7-5. We'll do something like:

Autumn wind sighs,
Leaves dance, a silver river,
Night cradles the hush.

Let's count: "Autumn wind sighs," - 5 syllables? Let's count: Au-tumn (2), wind (1) =3, sighs (1) =4. Actually "Autumn" has 2 syllables, "wind" 1, "sighs" 1 = 4. We need 5. We can add "softly" maybe. Let's produce a proper one.

Better: "Autumn wind softly" = 2+1+2 =5. Good. Then "Leaves dance, a silver river" Count: Leaves (1), dance (1) =2, a (1)=3, sil-ver (2)=5, ri-ver (2)=7. Yes 7. Then "Night cradles the hush." Count: Night(1), cradles(1)=2, the(1)=3, hush(1)=4. Need 5. Add "soft" maybe. "Night cradles soft hush." Count: Night(1), cradles(1)=2, soft(1)=3, hush(1)=4. Still 4. Add "deep" maybe. "Night cradles deep soft hush." Count: Night(1), cradles(1)=2, deep(1)=3, soft(1)=4, hush(1)=5. Good.

Alternatively produce a classic haiku like "Frost on the pond / The fish lies dormant below / Silence fills the air." Let's check syllable counts: "Frost on the pond" Frost(1), on(1)=2, the(1)=3, pond(1)=4. Need 5. Add "quiet" maybe. "Frost on the quiet pond" 1+1+1+1+1 =5. Then second line: "The fish lies dormant below". The(1) fish(1)=2, lies(1)=3, dor-mant(2)=5, be-low(2)=7. Third: "Silence fills the air" Silence(2)=2, fills(1)=3, the(1)=4, air(1)=5. That works.

Let's produce a haiku. Provide 3 lines. Probably no explanation needed.

Ok produce.
...done thinking.

Frost on the quiet pond  
The fish lies dormant below—  
Silence fills the air.

total duration:       2.529069643s
load duration:        92.59669ms
prompt eval count:    72 token(s)
prompt eval duration: 4.553645ms
prompt eval rate:     15811.51 tokens/s
eval count:           595 token(s)
eval duration:        2.333744834s
eval rate:            254.96 tokens/s
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment