Skip to main contentThe Local Bench

Illustrative build · sample prices

Dual-GPU 48 GB

Two used 3090s split a 70B model by layers. Capacity doubles; single-stream bandwidth does not.

Multi-GPU tower
  • Data checked: 2026-10-12. Review interval shown below. Prices need review after 30 days; specifications and benchmark collections after 90 days. Dates describe the recorded check, not a live retailer quote.

    • Verified Checked against the primary source (for example the model’s config.json on Hugging Face).
    • Sourced Taken from a cited source; not yet independently checked.
    • Estimate Calculated with the published formulas; not a measurement.
    • Sample Placeholder data for layout; not a real figure.
    • Not yet measured No public, citable score found for this benchmark yet. We never fill the gap with a guess.

    Verified metadata is not a measured performance result. Only cited runs matching model, device, quant and recorded context can replace speed estimates. Estimates assume one stream and F16 KV cache, with a ±25% uncertainty band. A published run applies only to its recorded runtime and conditions; missing settings are unknown.

    Evidence labels · Freshness policy · Calculation methodology

$4,044
complete-system parts, USDSample
12.4tok/s
Llama 3.3 70B InstructEstimate
780W
at load; 110 W idle
23.0months
illustrative break-even against a $200/month plan

Includes the listed system parts. Excludes tax, shipping, peripherals, maintenance, financing and resale. Sample parts prices may differ from current hardware catalogue prices. Electricity assumes 780 W whole-system load for 4 h and 110 W idle for 20 h daily, 30 days at $0.15/kWh. Power is unmeasured.

Parts list

Sample parts and prices for layout.

Part
Item
Price, USD
GPU2× GeForce RTX 3090 24 GB (used)$2,400Sample
CPURyzen 9 7900$349Sample
MotherboardX670E ATX (x8/x8)$299Sample
Memory128 GB DDR5-5600 (4×32)$359Sample
Storage4 TB NVMe Gen4$229Sample
PSU1$249Sample
CaseFull tower$159Sample

No affiliate links yet. Prices will be timestamped per retailer once tracking starts.

Expected speed

Reference models, default quant, 8k context.

Model
Verdict
tok/s
Llama 3.3 70B InstructTight12.4Estimate
Qwen3 32BFits25.6Estimate
Gemma 3 27B ITFits31.8Estimate

Break-even card

Electricity $23.94/month (780 W × 4 h + 110 W idle × 20 h, $0.15/kWh).

Plan
Price / month
Months to break even
ChatGPT Plus$20never
Claude Pro$20never
Google AI Pro$19.99never
Claude Max (5×)$10053.2
ChatGPT Pro 100$10053.2
Google AI Ultra (5×)$99.9953.2
Claude Max (20×)$20023.0
ChatGPT Pro 200$20023.0
Google AI Ultra (20×)$199.9923.0
ChatGPT Pro 500 (top tier)$5008.5

Break-even = hardware ÷ (plan − electricity). It ignores capability: a local model is not a frontier model. Adjust the numbers.