Decentralization and the Concentration Effect
Integrating computing power into existing buildings can preserve land and make use of waste heat. Careful site selection and operation are essential.
Expertise and insights
Practical guidance for running AI on your own infrastructure.
Apertus v1.5 70B on a single NVIDIA RTX PRO 6000 Blackwell
Apertus v1.5 70B ships as approximately 145 GB (135 GiB) of BF16 weights that officially need two H200 datacenter GPUs. We show how to run it on a single NVIDIA RTX PRO 6000 Blackwell with 96 GB VRAM, with vision input, native tool calling, and a 192k context profile (229k measured maximum). Using three-tier mixed-precision quantization (NVFP4 + FP8) it decodes 55% faster than FP8 while using 32% less weight storage, with 99% of FP8 MMLU quality.
Which AI Server Is Right for My SME?
Four on-premise AI server configurations compared: tokens per second, concurrent users, and what KV cache and context windows mean for everyday office work.
Why On-Premise AI Is Almost Always Safer
Free models, exposed credentials, and pseudonymization that does not hold up: why local AI is the only truly consistent answer to the security question.
API Standards for Large Language Models (LLMs)
Practical guide to LLM API standards: OpenAI-compatible Chat Completion API for on-premise models (GPT-OSS, Llama, Qwen, DeepSeek), LiteLLM as unified gateway, control parameters, multimodality, structured outputs, tool calling and MCP. Everything developers need to know for enterprise LLM integration.
Precise AI Capacity Planning: SLA-Based Sizing with Engset's Formula
How proven telecommunications mathematics supports precise AI capacity planning. A technical white paper on SLA-based sizing with Engset's formula for on-premise AI infrastructure.
AI Costs by Office Role: Token Consumption Patterns Across Your Team
Calculate the real AI costs for your team. From Executive Assistants to Software Engineers: Understand what drives token consumption by office role and why developers are your biggest budget item.
Transparently Calculate On-Premises AI Costs
Our interactive cost calculator helps you accurately determine hardware requirements and costs for running your own Large Language Models, including a comparison of AMD and NVIDIA solutions.
Dataset: LLM Token Usage in Everyday Office Tasks
Comprehensive dataset of token consumption across 63 real-world AI tasks, including standard and reasoning models, with multimodal inputs.
On-Premise AI Solutions for every SME
On-premise AI for SMEs: Practical guide to models, hardware, and operations. Concrete recommendations on servers, data protection, and stability.
No matching articles found.
Try another search term.