Why model weights are not enough
Running a model also needs KV cache, working buffers and runtime memory. A lower bound does not prove fit. Additional memory depends on the model, format and runtime.
Primary sourceAI hardware knowledge base
Memory, workloads and the limits of preliminary hardware planning. Each guide provides a short answer, an explanation and a next step.
- How much VRAM does an LLM need?
Estimate LLM memory from weights, quantization, KV cache, context and concurrency. An 8B example and the limits of a preliminary Monsder calculation.
- How to choose an AI server
Choose an AI server from workload requirements: model, load, GPU, memory, power and cooling. What a calculator estimates and what needs engineering validation.
- Which models fit your GPU?
Match an LLM to your GPU using available VRAM, quantization, context and concurrency. Why fitting in memory does not establish a validated deployment.
About Monsder
Monsder translates a task, model and workload into preliminary AI hardware requirements. Start with a model or find model candidates for an existing GPU. Engineering suitability comes first; price and availability are considered separately.
The calculator uses a deterministic calculation engine. The assistant can suggest workload parameters, which are applied only after user confirmation. Manual calculation also works without the assistant.
Monsder selects new equipment and remains neutral between manufacturers and suppliers. Price and availability are considered separately from engineering suitability. Automated estimates do not replace specialist review or a commercial proposal.
Sources and updates
These materials were prepared for the Monsder knowledge base. They explain preliminary estimation and link to primary documentation in each guide. Editorial dates describe article revisions; specification verification dates are shown in the catalogs.
Editorial revision: .
Additional reference material
Users and concurrent requests
Registered users are not concurrent requests. KV-cache estimation needs active sequences and combined prompt/output lengths. Performance needs a separate benchmark.
Primary sourceHow project versions work
A draft is editable. A calculation freezes its requirements, method and catalog. A request references that exact snapshot. Engineering revisions and commercial offers are stored separately.
Budget and independent selection
Budget is guidance, not a reason to degrade the required configuration. Suppliers and prices never enter the calculation kernel. Staff add services separately after confirming fulfillment.