AI/ML●●Solid
Can Europe train a frontier AI model on the compute it owns?
Sourced model with 52 tests showing federated compute beats waiting for grid power.
Big BrainBold Bet
smashini
1432954d ago
Back to browse
GitHub Repository
Reproducible microbenchmark for modeling domain crossing energy in heterogeneous compute systems.
0 starsPython
CrossingBench – Modeling when data movement dominates compute energy
by JessyMorissette·Feb 17, 2026·3 points·6 comments
AI Analysis
●●SolidBig BrainNiche Gem
Rigorous energy model for domain crossings in chiplets and analog CIM systems.
Strengths
- •Novel elasticity metric (ε) quantifies crossing dominance—not obvious framing
- •Comprehensive cost model with sourced 7nm parameters, DOI-published methodology
- •Pure Python, zero dependencies, reproducible sweep CLI designed for tweaking assumptions
Weaknesses
- •Niche audience—only relevant for architects optimizing chiplet designs or analog systems
- •Early stage: working paper not peer-reviewed, small GitHub activity suggests pre-launch
- •Limited empirical validation—relies on published parameters, not new measurements
Category
Target Audience
Computer architects, systems researchers, heterogeneous computing engineers
Post Description
I built a small reproducible microbenchmark exploring when system energy becomes dominated by boundary crossings rather than intra-domain compute.
The model decomposes total energy into: C = C_intra + Σ V_b · c_b
Includes:
CLI sweeps
Elasticity metric (ε) as a dominance indicator
CSV outputs
Working draft paper
*DOI
Looking for critique, counter-examples, or prior related work I may have missed.
Similar Projects
AI/ML●Mid
Real-time violent action detection using computer vision
Standard YOLO + LSTM pipeline for violence detection, nothing novel beyond the specific use case.
Bold Bet
iraton
213mo ago
Education●Mid
I built a small repertoir of different computing systems
Curated list of computing substrates lacks depth beyond basic taxonomy tags.
Niche GemCozy
tugdual
401mo ago
Open Source●●●Banger
opub, donated compute for open-source
Donors fund compute credits so maintainers can use AI tools without burning personal cash.
Solve My ProblemDark HorseCozy
goodroot
2029d ago
AI/ML●●Solid
Reproducible open-source STT API benchmarks with full methodology
Fixes WER scores by normalizing '$50' and 'fifty dollars' as equivalent.
Solve My ProblemCozy
jilijeanlouis
112mo ago
AI/ML●●Solid
Structured Python control over AI computer use agents
Accessibility tree beats screenshot tokens, per-step model control is genuinely clever.
Big BrainWizardry
aadyachinubhai
142mo ago