# Autodesk Vault AI Assistant Sizing and Value Calculator > Source: https://netray.co/tools/autodesk-vault-ai-assistant-sizing-and-value-calculator > Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems. > Section: Free Tools ## Short answer Size GPUs and index storage and test payback for a private AI assistant on an on-prem Autodesk Vault. ## Category engineering-plm ## Erp Autodesk Vault Professional ## Audience - CAD administrator - IT manager - CIO ## Tool **Kind:** calculator **Currency:** $ **Inputs:** #### Named users **Key:** users **Type:** number **Default:** 90 **Min:** 1 **Step:** 5 **Unit:** users #### Daily active share **Key:** active_pct **Type:** slider **Default:** 35 **Min:** 5 **Max:** 100 **Step:** 5 **Unit:** % #### Questions per active user per day **Key:** q_per_user **Type:** number **Default:** 7 **Min:** 1 **Step:** 1 #### Prompt plus answer tokens per question **Key:** tokens_per_q **Type:** number **Default:** 2800 **Min:** 200 **Step:** 100 #### Documents to index **Key:** docs **Type:** number **Default:** 120000 **Min:** 0 **Step:** 10000 #### Average pages per document **Key:** pages_per_doc **Type:** number **Default:** 4 **Min:** 1 **Step:** 1 #### Minutes saved per question **Key:** min_saved **Type:** number **Default:** 5 **Min:** 0 **Step:** 1 #### Loaded hourly cost **Key:** hourly_cost **Type:** number **Default:** 60 **Min:** 0 **Step:** 5 **Unit:** $/h #### GPU monthly cost **Key:** gpu_cost **Type:** number **Default:** 1100 **Min:** 0 **Step:** 50 **Unit:** $/GPU/month #### Aggregate tokens per second per GPU **Key:** gpu_tps **Type:** number **Default:** 700 **Min:** 50 **Step:** 50 #### GPUs needed just to hold the model **Key:** min_gpus **Type:** number **Default:** 1 **Min:** 1 **Step:** 1 #### One-off integration cost **Key:** integ_cost **Type:** number **Default:** 30000 **Min:** 0 **Step:** 2500 **Unit:** $ **Outputs:** #### Daily active users **Key:** active_users **Expr:** users * active_pct / 100 **Format:** integer #### Questions per day **Key:** q_day **Expr:** active_users * q_per_user **Format:** integer #### Tokens per day **Key:** tokens_day **Expr:** q_day * tokens_per_q **Format:** number #### Peak concurrent requests **Key:** peak_conc **Expr:** q_day * 0.25 / 3600 * 20 **Format:** decimal **Description:** 25% of daily questions in the busiest hour, about 20 seconds per answer. #### GPU estimate **Key:** gpus **Expr:** max(min_gpus, ceil(peak_conc * 25 / gpu_tps)) **Format:** integer **Primary:** true #### Vector index size **Key:** idx_gb **Expr:** docs * pages_per_doc * 2 * 6000 / 1000000000 **Format:** decimal **Description:** 2 chunks per page and about 6 KB per chunk. #### Hours saved per month **Key:** hours_month **Expr:** q_day * min_saved / 60 * 21 **Format:** hours #### Monthly value **Key:** value_month **Expr:** hours_month * hourly_cost **Format:** currency **Primary:** true #### Monthly GPU cost **Key:** cost_month **Expr:** gpus * gpu_cost **Format:** currency #### Simple payback **Key:** payback_months **Expr:** if(gt(value_month - cost_month, 0), integ_cost / (value_month - cost_month), 0) **Format:** months **Chart:** #### Tokens per day by named users **Type:** line **X:** users **From:** 20 **To:** 1000 **Steps:** 25 **Series:** - tokens_day **Disclaimer:** Planning estimate. Confirm throughput on your hardware and minutes saved in a pilot. ## Content **Intro:** This calculator sizes a private AI assistant on an on-prem Autodesk Vault: GPUs, daily tokens, index storage and payback. It suits teams from a few dozen to a few hundred users. **Sections:** #### How to use it **Body:** Enter named users, daily active share and questions per user from pilot logs if you have them. Set tokens per question to match your retrieval depth. Enter GPU cost as a monthly equivalent. The chart shows how daily token volume grows with users. #### Formulas and assumptions **Body:** Questions per day equal users times active share times questions per user. Tokens per day multiply that by tokens per question. Peak concurrency assumes 25 percent of questions in the busiest hour at about 20 seconds each. GPU count is the larger of the model floor and peak streams times 25 tokens per second divided by GPU throughput. #### Reading the result **Body:** Daily tokens help you size throughput, but peak concurrency drives GPUs. The index estimate counts only text-bearing documents, so enter PDFs and Office files, not every CAD file. Payback is shown as zero when value does not exceed GPU cost. #### Common mistakes **Body:** Counting every Vault file as a document, using average rather than busy-hour load, and ignoring Job Processor and embedding capacity during the initial index build, which can be heavy for a large vault. **Cta:** Netray can compare these results with a measured pilot on your Vault. The calculator is usable on its own. **Faq:** #### How long does the first index build take? **Answer:** It depends on file counts, parsing speed and hardware. Large vaults can take hours to days. Plan the build outside working hours and keep it separate from the live assistant's GPU. #### Should the index include old revisions? **Answer:** Usually only released and latest working versions. Old revisions add size and can mislead answers unless chunks are labelled with revision and state. #### What if the result shows one GPU? **Answer:** That is common. Consider failover separately, since a single GPU is a single point of failure for the assistant, though not for Vault itself. #### Can I trust the monthly value? **Answer:** Only as much as the minutes-saved input. Measure it with timed tasks and apply a realistic adoption rate. ## Search terms Autodesk Vault AI sizing, Vault assistant ROI, on-prem LLM GPU estimate, AI payback calculator engineering, RAG storage estimate, private AI cost calculator --- Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json