Home
Services
Local AI Deployment Test Results Image & Video NeoCloud Results Offline Private Security System Digital Transformation Professional Photography
Team
Bruce Experience Education Contact

NeoCloud Test Results

The same kinds of engines and models we run on our own workstations, measured on rented GPUs from neocloud providers, starting with Google Colab. Each neocloud provider gets its own section as we test it, and every notebook is published on GitHub.Engines and models measured on neocloud GPUs, starting with Google Colab.

Google Colab

Our open notebooks run language and image models on Google Colab Pro GPU runtimes. Each one installs a single engine, downloads its model once to Google Drive and serves it through one authenticated OpenAI-compatible API. Measured on real Colab runtimes in November 2026, one request at a time.Our open notebooks on Google Colab Pro GPU runtimes, measured in November 2026.

Decode is a 512-token reply at temperature 0. Prefill speed and time to first token use a fresh prompt of about 3,500 tokens on a loaded model (no prefix-cache reuse); notebooks not yet rerun with it read “next test”. Image and video speed is per output after the first.

L4 · 24 GB (RTX 4070 Ti class)

NVIDIA L4, a low-profile single-slot data-centre GPU with a gold finned heatsink
NVIDIA L4 · 24 GB GDDR6 · 72 W, single slot Image: NVIDIA
Language models
Model notebookEnginePrefill t/sTime to first tokenDecode t/s
nemotron-3.5-bolt-74b-iq2llama.cpp—next test—next test38
qwen3.8-27b-w4a16-sglangSGLang—next test—next test16
flash-next-gyro-sStrata—next test—next test68
qwen3.8-27b-exl3-2bpwTensorFold—next test—next test83
Image and video models
Model notebookEngineOutputTime per output
flux3-klein-comfyComfyUI1024×1024 image8 s
minimax-h3-l4-comfyComfyUI832×480 5 s video clip124 s
qwen-image-2.1-turbo-comfyComfyUI1024×1024 image8 s
qwen-image-2.1-nf4DiffusersImage37 s
sdxl-baseDiffusersImage17 s
sdxl-turboDiffusersImage3 s

A100 · 80 GB (High-RAM)

NVIDIA A100 GPU module with its chip and HBM memory stacks
NVIDIA A100 · 80 GB HBM2e Image: NVIDIA
Language models
Model notebookEnginePrefill t/sTime to first tokenDecode t/s
nemotron-3.5-bolt-74b-iq2llama.cpp—next test—next test79
sage-a1.5-q4llama.cpp3,5581.00 s134
qwen3.8-27b-w4a16-sglangSGLang2,0181.76 s74
flash-next-strata-iq3-sStrata—next test—next test167
qwen3.8-27b-exl3-2bpwTensorFold—next test—next test144
Image and video models
Model notebookEngineOutputTime per output
flux3-klein-comfyComfyUIImage3 s
minimax-h3-comfyComfyUI832×480 5 s video clip35 s
qwen-image-2.1-bf16-comfyComfyUI1024×1024 image9 s
qwen-image-2.1-turbo-bf16-comfyComfyUI1024×1024 image4 s
wan22-5b-comfyComfyUI832×480 5 s video clip61 s

G4 · RTX PRO 6000 Blackwell, 96 GB

NVIDIA RTX PRO 6000 Blackwell Server Edition, a dual-slot passive GPU
NVIDIA RTX PRO 6000 Blackwell · 96 GB GDDR7 Image: NVIDIA
Language models
Model notebookEnginePrefill t/sTime to first tokenDecode t/s
basalt-flash-next-nvfp4-dq6kBasalt—next test—next test273
hyperqwen-qwen3.8-27b-dflash2HyperQwen—next test—next test~315
glm-maya-sMaya—next test—next test77
qwen3.8-27b-w4a16-sglangSGLang—next test—next test77

The remaining A100 and G4 notebooks are still being tested.

Google Colab notebooks on GitHub →

Ready to get started?

Tell us about your environment, we’ll design a deployment that keeps your data yours.

Get in Touch