# Teamcenter AI Assistant Sizing and Value Calculator

> Source: https://netray.co/tools/teamcenter-ai-assistant-sizing-and-value-calculator
> Publisher: Netray (netray.co), AI and integration engineering for ERP and manufacturing systems.
> Section: Free Tools

## Short answer

Estimate GPU count, daily tokens, vector index size and payback for an on-prem AI assistant on Teamcenter from your user and usage numbers.

## Category

engineering-plm

## Erp

Siemens Teamcenter

## Audience

- PLM administrator
- IT architect
- CIO

## Tool

**Kind:** calculator

**Currency:** $

**Inputs:**

#### Named users

**Key:** users

**Type:** number

**Default:** 400

**Min:** 1

**Step:** 10

**Unit:** users

#### Daily active share

**Key:** active_pct

**Type:** slider

**Default:** 30

**Min:** 5

**Max:** 100

**Step:** 5

**Unit:** %

**Help:** Share of named users who ask at least one question on a given day.

#### Questions per active user per day

**Key:** q_per_user

**Type:** number

**Default:** 8

**Min:** 1

**Step:** 1

#### Prompt plus answer tokens per question

**Key:** tokens_per_q

**Type:** number

**Default:** 3500

**Min:** 200

**Step:** 100

**Help:** Retrieval context makes prompts long. 2,000 to 6,000 is common.

#### Documents to index

**Key:** docs

**Type:** number

**Default:** 250000

**Min:** 0

**Step:** 10000

#### Average pages per document

**Key:** pages_per_doc

**Type:** number

**Default:** 6

**Min:** 1

**Step:** 1

#### Minutes saved per question

**Key:** min_saved

**Type:** number

**Default:** 6

**Min:** 0

**Step:** 1

#### Loaded hourly cost

**Key:** hourly_cost

**Type:** number

**Default:** 75

**Min:** 0

**Step:** 5

**Unit:** $/h

#### GPU monthly cost

**Key:** gpu_cost

**Type:** number

**Default:** 2200

**Min:** 0

**Step:** 100

**Unit:** $/GPU/month

**Help:** Lease, depreciation or colocation equivalent per GPU.

#### Aggregate tokens per second per GPU

**Key:** gpu_tps

**Type:** number

**Default:** 1200

**Min:** 100

**Step:** 100

**Help:** Batched throughput for your model and GPU. Measure it, do not trust a brochure.

#### GPUs needed just to hold the model

**Key:** min_gpus

**Type:** number

**Default:** 2

**Min:** 1

**Step:** 1

#### One-off integration cost

**Key:** integ_cost

**Type:** number

**Default:** 60000

**Min:** 0

**Step:** 5000

**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:** Assumes 25% of daily questions in the busiest hour and about 20 seconds per answer.

#### GPU estimate

**Key:** gpus

**Expr:** max(min_gpus, ceil(peak_conc * 25 / gpu_tps))

**Format:** integer

**Primary:** true

**Description:** Assumes about 25 tokens per second per concurrent stream. The model footprint sets the floor.

#### Vector index size

**Key:** idx_gb

**Expr:** docs * pages_per_doc * 2 * 6000 / 1000000000

**Format:** decimal

**Description:** Assumes 2 chunks per page and about 6 KB per chunk including vector and metadata.

#### 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

**Description:** Zero means monthly value does not exceed GPU cost.

**Chart:**

#### Monthly value vs GPU cost by named users

**Type:** line

**X:** users

**From:** 50

**To:** 2000

**Steps:** 20

**Series:**

- value_month
- cost_month

**Disclaimer:** Planning estimate only. Hours saved depend on adoption and are the least certain input. Validate throughput with your own model before buying hardware.

## Content

**Intro:** This calculator estimates what a private AI assistant on Teamcenter needs in GPUs and index storage, and whether time saved covers the cost. Change the inputs and watch the chart.

**Sections:**

#### How to use it

**Body:** Start with named users and the share that is active on a typical day. Adjust questions per user to match what a pilot shows. Enter your real GPU cost and a measured tokens-per-second figure. The primary results are the GPU estimate and monthly value.

#### Formulas and assumptions

**Body:** Daily questions equal named users times active share times questions per user. Peak concurrency assumes a quarter of daily questions fall in the busiest hour and each takes about 20 seconds. GPUs equal the larger of the model floor and peak concurrency times 25 tokens per second divided by GPU throughput. Index size assumes two chunks per page at about 6 KB each.

**Bullets:**

- Hours saved = questions x minutes saved / 60 x 21 working days
- Value = hours x loaded hourly cost
- Payback = one-off cost / (value minus GPU cost)

#### Reading the result

**Body:** At small scale the model's memory need, not throughput, sets the GPU count, so the estimate stays flat as users grow and then steps up. The chart shows where value overtakes cost. If payback is long, test whether saved minutes per question are realistic before cutting scope.

#### Common mistakes

**Body:** Using vendor peak throughput instead of batched, measured throughput, and counting every named user as active. Also remember that retrieval makes prompts long, which raises tokens per question far above chat defaults.

**Cta:** If you want these numbers checked against your Teamcenter estate and a pilot plan, Netray can run that review. The calculator works without it.

**Faq:**

#### Why does the GPU count not rise smoothly?

**Answer:** A large model needs a minimum number of GPUs just to load. Until peak load exceeds that hardware's throughput, extra users add no GPUs, then capacity steps up in whole GPUs.

#### What tokens per second should I enter?

**Answer:** Use a measured aggregate figure for your model, quantisation and context length under concurrent load. Ranges from several hundred to a few thousand tokens per second per GPU are common.

#### Are the time savings reliable?

**Answer:** They are the weakest input. Take minutes saved from a pilot with timed tasks, then apply a haircut for adoption before presenting a business case.

#### Does it include power and staff?

**Answer:** Only if you fold them into the GPU monthly cost and the one-off cost. Add admin time, power, cooling and support to be realistic.

## Search terms

Teamcenter AI sizing, PLM AI ROI calculator, LLM GPU sizing calculator, on-prem AI payback, Teamcenter assistant cost, RAG index size estimate

---
Talk to an engineer: https://netray.co/contact?intent=erp_ai | Machine catalog: https://netray.co/agent/catalog.json