Skip to content

feat: add cloud GPU cost calculator with provider pricing#11

Open
stmatengss wants to merge 1 commit into
mainfrom
feat/cost-calculator
Open

feat: add cloud GPU cost calculator with provider pricing#11
stmatengss wants to merge 1 commit into
mainfrom
feat/cost-calculator

Conversation

@stmatengss

Copy link
Copy Markdown
Collaborator

Summary

  • Add a Cloud GPU Cost Estimate card to the web calculator showing monthly costs across AWS, GCP, and Azure for the selected GPU type
  • Include a GPU_PRICING data table with realistic on-demand $/hr pricing for 9 GPU types (H100, H100-NVL, H200, A100-80, A100-40, A800, H20, L40S, L20)
  • Show cost savings from KV caching by computing how many fewer GPUs are needed thanks to cache hit rate improvements
  • Display cost estimates in the guided wizard Step 4 results for immediate visibility
  • Gracefully handle GPUs without cloud pricing data (H20, A800, L20 are primarily available in China)

Details

Cost Estimation Card

The new card appears between the Recommendation panel and Summary Metrics. It shows:

  • Headline: Best available monthly cost (e.g., "$21.4K/month")
  • Provider table: Per-provider $/hr, monthly total, and cost per 1K concurrent agents
  • Best price highlighting: The cheapest provider is highlighted in green
  • KV caching savings: Shows how much money is saved by KV caching vs. running without cache (using TPS gain from hit rate)
  • Disclaimer: Notes that estimates are based on on-demand pricing

Pricing Data

Uses approximate 2024-2025 on-demand pricing:

GPU AWS GCP Azure
H100 80GB $4.76/hr $4.28/hr $3.67/hr
H200 141GB $6.98/hr $5.67/hr
A100 80GB $3.06/hr $2.94/hr $2.48/hr
L40S 48GB $1.84/hr $1.70/hr $1.53/hr

H20, A800, and L20 show "Pricing not available" since they are primarily available in China.

Test plan

  • node tests/validate_js_parity.mjs passes all 178 checks (no regressions)
  • Open calculator in browser, select H100/A100/L40S preset → verify cost card appears with provider table
  • Select H20 preset → verify "Pricing not available" message appears
  • Run guided wizard through Step 4 with H100 → verify monthly cost estimate badge appears
  • Verify cost savings calculation shows reasonable values for high-hit-rate workloads

Co-Authored-By: Claude Opus 4.6 noreply@anthropic.com

Add cost estimation card to the web calculator that shows:
- Monthly cloud GPU costs for AWS, GCP, and Azure
- Per-provider pricing table with best-price highlighting
- Cost per 1,000 concurrent agents per month
- Savings from KV caching vs. running without cache
- Pricing data for H100, H200, A100, L40S and variants
- Cost estimate in the guided wizard Step 4 results
- Graceful fallback for GPUs without cloud pricing (H20, A800, L20)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant