QuickSight Dashboard

Amazon QuickSight provides visual dashboards for GPU chargeback data. This page covers creating Athena tables from your S3 reports and building QuickSight visualizations.

Configure QuickSight S3 Access

1. Go to QuickSight Console → Manage QuickSight → Security & permissions
2. Click Manage under "QuickSight access to AWS services"
3. Select S3 and add your GPU accounting bucket

Create Athena Tables

Run in the Athena Query Editor. These tables match the Hive-style S3 paths from S3 Upload.

:::warning Each Athena Table Needs Its Own S3 Path Each table must point to a unique S3 sub-path matching the report=<type> prefix used by the upload script. If all tables share the same LOCATION, Athena will try to read every CSV into every table, causing schema mismatches. :::

CREATE DATABASE IF NOT EXISTS gpu_chargeback;

-- Account Utilization (GPU hours by team & user)
CREATE EXTERNAL TABLE IF NOT EXISTS gpu_chargeback.account_utilization (
    account STRING,
    login STRING,
    proper STRING,
    used DOUBLE
)
PARTITIONED BY (period STRING, year STRING, month STRING, day STRING)
ROW FORMAT DELIMITED FIELDS TERMINATED BY '|'
STORED AS TEXTFILE
LOCATION 's3://<YOUR_GPU_ACCOUNTING_BUCKET>/slurm-reports/report=account_utilization/'
TBLPROPERTIES ('skip.header.line.count'='0');

-- Top Users
CREATE EXTERNAL TABLE IF NOT EXISTS gpu_chargeback.top_users (
    account STRING,
    login STRING,
    used DOUBLE
)
PARTITIONED BY (period STRING, year STRING, month STRING, day STRING)
ROW FORMAT DELIMITED FIELDS TERMINATED BY '|'
STORED AS TEXTFILE
LOCATION 's3://<YOUR_GPU_ACCOUNTING_BUCKET>/slurm-reports/report=top_users/';

-- Jobs Detailed (includes project-id in comment)
CREATE EXTERNAL TABLE IF NOT EXISTS gpu_chargeback.jobs_detailed (
    jobid STRING,
    jobname STRING,
    user STRING,
    account STRING,
    partition STRING,
    state STRING,
    elapsed STRING,
    alloctres STRING,
    comment STRING
)
PARTITIONED BY (period STRING, year STRING, month STRING, day STRING)
ROW FORMAT DELIMITED FIELDS TERMINATED BY '|'
STORED AS TEXTFILE
LOCATION 's3://<YOUR_GPU_ACCOUNTING_BUCKET>/slurm-reports/report=jobs_detailed/';

-- GPU Count per Job
CREATE EXTERNAL TABLE IF NOT EXISTS gpu_chargeback.gpu_count_per_job (
    jobid STRING,
    jobname STRING,
    user STRING,
    account STRING,
    allocgres STRING,
    elapsed STRING,
    state STRING,
    start_time STRING,
    end_time STRING
)
PARTITIONED BY (period STRING, year STRING, month STRING, day STRING)
ROW FORMAT DELIMITED FIELDS TERMINATED BY '|'
STORED AS TEXTFILE
LOCATION 's3://<YOUR_GPU_ACCOUNTING_BUCKET>/slurm-reports/report=gpu_count_per_job/';

-- GPU Count Summary
CREATE EXTERNAL TABLE IF NOT EXISTS gpu_chargeback.gpu_count_summary (
    account STRING,
    user STRING,
    total_jobs INT,
    total_gpus_allocated INT,
    avg_gpus_per_job DOUBLE,
    max_gpus INT
)
PARTITIONED BY (period STRING, year STRING, month STRING, day STRING)
ROW FORMAT DELIMITED FIELDS TERMINATED BY '|'
STORED AS TEXTFILE
LOCATION 's3://<YOUR_GPU_ACCOUNTING_BUCKET>/slurm-reports/report=gpu_count_summary/';

-- Load all partitions
MSCK REPAIR TABLE gpu_chargeback.account_utilization;
MSCK REPAIR TABLE gpu_chargeback.top_users;
MSCK REPAIR TABLE gpu_chargeback.jobs_detailed;
MSCK REPAIR TABLE gpu_chargeback.gpu_count_per_job;
MSCK REPAIR TABLE gpu_chargeback.gpu_count_summary;

:::note Partition Discovery Run MSCK REPAIR TABLE after each new upload to discover new partitions. Or use a Glue Crawler to auto-detect them. :::

Create QuickSight Datasets

1. QuickSight Console → Datasets → New Dataset → Athena
2. Create datasets:

Dataset Name	Athena Source	Import Mode
gpu_account_utilization	gpu_chargeback.account_utilization	SPICE
gpu_user_cost	gpu_chargeback.user_cost	SPICE
gpu_team_cost	gpu_chargeback.team_cost	SPICE
gpu_project_cost	gpu_chargeback.project_cost	SPICE

3. Schedule SPICE refresh daily (after report upload cron)

:::info Cost Views The user_cost, team_cost, and project_cost views are created in Cost Allocation from CUR. Create them before building cost-based datasets. :::

Build GPU Chargeback Dashboard

Option A: Standalone Dashboard

1. Create a new Analysis from your GPU datasets
2. Build sheets with the visuals below
3. Publish as a dashboard

Option B: Add Tab to CID/CUDOS

1. Open CID Dashboard → Save As → Create Analysis (backup first)
2. In Analysis → click Add dataset → add your GPU cost datasets
3. Click + Add sheet → name it "GPU Chargeback"
4. Build visuals below
5. Publish updated dashboard

:::warning CID Updates CID template updates may overwrite custom sheets. Always save a backup Analysis copy before updating CID. :::

Recommended Visuals

GPU Usage Sheet

Visual	Type	Configuration
GPU Hours by Team	Bar Chart	X: account, Value: SUM(used)
Top Users	Table	Columns: login, account, used
Cluster Utilization	KPI	Value: SUM(allocated) / SUM(total)
Usage Trend	Line Chart	X: month, Value: SUM(used), Color: account

GPU Cost Sheet

Visual	Type	Configuration
Team Cost Summary	Bar Chart	X: team, Value: SUM(total_cost_usd)
Top Users by Cost	Table	Columns: username, team, cost_usd, gpu_hours
Cost by Project	Donut Chart	Group: project_id, Value: SUM(cost_usd)
Monthly Cost Trend	Line Chart	X: month, Value: SUM(total_cost_usd), Color: team
KPI: Total GPU Cost	KPI	Value: SUM(cost_usd)
KPI: Total GPU Hours	KPI	Value: SUM(gpu_hours)
KPI: Cost per GPU-Hour	KPI	Value: AVG(gpu_hourly_cost)

Modify Dashboard

Action	Steps
Add visualization	Edit dashboard → Add visual → Select chart type
Add date filter	Edit → Add filter → Select date field
Add calculated field	Edit dataset → Add calculated field → Enter formula
Share dashboard	Dashboard → Share → Users/Groups → Set permissions
Schedule email reports	Dashboard → Schedule → Set frequency → Add recipients
Refresh data	Datasets → Select dataset → Refresh now