MMOOZZGG.com

Market case · IT companies · Russia

LLM VRAM requirements in practice: two models in 22.4 GB of 24 GB with a 65,536-token context

1 min read1 source

The team built a local AI agent in a closed environment with Ollama and an agent layer, on a rented GPU server with one NVIDIA A30 card of 24 GB. The models were gpt-oss 20b (13 GB) for text and tools and qwen3-vl 8b (6.1 GB) for images; with the KV cache for a 65,536-token context, they used 22.4 GB.

On this page · 5 sections
Resulttwo models fit into 22.4 GB of 24 GB of video memory with a 65,536-token context
SourceDigital Platforms, case no. 4084, 2 September 2026
A similar task for usAI employee
IndustryIT companies

What was done and what came out

  • The agent edits texts in set personas, runs a night queue of research tasks started by a scheduler every 10 minutes, searches the web and works with files inside its working folder only.
  • It is reliable for a chain of about 6 tool calls in a row; the timeout for a cold model load had to go above 300 seconds; fixing twelve errors took one evening.
  • It serves one user at a time.
  • Published on 21 August 2026.

What to take for yourself

Exact memory math from the case: model weights plus the KV cache must fit into video memory with a margin. Our memory calculator uses the same formula.

AI agents on a local model for your team: a pilot in 14 days. A pilot AI agent on a local model in 14 days from $1,499 with the metric in the contract, on your hardware or on rented GPUs. An AI department of 3–5 agents in 4–6 weeks from $4,999. You buy the hardware to our specification.

Sources and fact-checking

The facts and numbers on this page come from the publication. Data type: project catalogue entry. The company is not our client: the page shows what a smaller company can repeat from this experience. Checked by the editors on 30 September 2026.

  1. Digital Platforms, case no. 4084 (platforms.su)· 2 September 2026

Found a mistake? Tell us

Similar cases

Prices and measurements in this section were checked against the sources at the bottom of each page. We build AI agents on cloud and local models and do not sell hardware. Calculations are estimates; legal conclusions on data protection come from your lawyer.

Contact us