Tarunyaa Sivakumar
Stream of consciousness — that's what this is. I think too much, so I started writing some of it down.
- GPU Chip Design Engineer at NVIDIA
- UPenn EE '25
- Singapore → Philadelphia → Bay Area
Projects
Find where smart people disagree
Decentralized compute infrastructure for AI training
Designed and taped out a hardware learning chip at Penn
Electrokinetic propulsion for autonomous micro-robots
Built a custom RTX 5090 workstation
Notes
“The person who says he knows what he thinks but cannot express it usually does not know what he thinks.”
— Mortimer Adler, How to Read a Book
Archive
Older writing from high school — a few thoughts that could either brighten up your day or trigger an existential crisis :)
23.10.20
Milkomeda
16.08.20
What is Real?
04.06.20
Superposition
04.06.20
Brave the Future
24.04.20
Change as a Constant
06.01.20
Our Summer: எங்கள் கோடை
17.08.19
Print the Future
17.07.19
Are We Alone?
17.12.18
Perception
Writing
Projects
Find where smart people disagree
Decentralized compute infrastructure for AI training
Designed and taped out a hardware learning chip at Penn
Electrokinetic propulsion for autonomous micro-robots
Built a custom RTX 5090 workstation
Values
01
Build what matters
Technology is only meaningful when it serves real problems. Work on things that push the boundary.
02
Stay curious, stay humble
The more you learn, the more you realize how much you don't know. Approach every problem with beginner's eyes.
03
Do the hard thing
Growth lives on the other side of discomfort. Whether it's a tapeout, a startup pivot, or an honest conversation — lean in.
04
Think in systems
Everything is connected. A great chip designer thinks about the software. Zoom out before you zoom in.
05
Write it down
Writing is thinking made visible. If you can't explain it simply, you don't understand it well enough.
06
People over everything
The best work happens with people you respect and trust. Nobody builds anything great alone.
Learnings
A running glossary of terms I've come across and wanted to pin down.
Ablation study
In ML, removing parts of a system one at a time to measure how much each contributes.
Red herring
A misleading clue or detail that diverts attention from the real issue.
Sharpe ratio
Risk-adjusted return: how much excess return you earn per unit of volatility.
Server production
The pace of building servers and bringing them online; a supply bottleneck in the AI compute buildout.
ASPs (average selling prices)
The average price a product actually sells for; a key driver of revenue and margin trends.
RPO / RPOs (remaining performance obligations)
Contracted future revenue a company has booked but not yet recognized; effectively backlog.
Midcurve
Markets slang for the mediocre, middle-of-the-distribution take (the "midwit"); also the belly of a yield curve.
SBC (stock-based compensation)
Paying employees in equity instead of cash; a non-cash expense that still dilutes shareholders.
Buybacks
A company repurchasing its own shares, returning capital and lifting per-share metrics.
Hawks and doves
In monetary policy, hawks favor higher interest rates to curb inflation while doves favor lower rates to support growth and employment. (Also used in foreign policy: hawks lean toward force, doves toward diplomacy.)
Conflation
Combining or blurring two distinct things into one, often mistakenly.
Ontology
A set of concepts and categories in a domain, with their properties and the relationships between them.
Backstop
Guaranteeing a deal with your own capital in case other investors don't show up.
Terminal value
The value of a business beyond the explicit forecast period, assuming it grows steadily forever.
Elo score
A rating system (from chess) that ranks relative skill from head-to-head results.
Conway's Law
Products mirror the communication structure of the organizations that build them.
Price's Law
Roughly the square root of a workforce produces 50% of the output.
Amdahl's Law
A program's maximum speedup is capped by its sequential part, the bottleneck.
Strong scaling
Fix the total problem size and add processors; measures how much faster a fixed workload runs (the Amdahl's Law regime).
Weak scaling
Grow the problem size in step with the processor count so work per processor stays fixed; measures whether bigger problems stay tractable (the Gustafson's Law regime).
Law of large numbers
As a sample grows, its average converges on the true expected value.
Timelines I've put together to keep scale and sequence straight.
A history of everything
4.54 billion years ago
Earth forms
3.5 billion years ago
First simple life appears
541 million years ago
Complex life rapidly expands in the oceans
252-201 million years ago
Triassic period
First dinosaurs appear.
201-145 million years ago
Jurassic period
Giant dinosaurs dominate.
145-66 million years ago
Cretaceous period
T. rex and Triceratops live.
66 million years ago
Dinosaur extinction
Asteroid impact.
7 million years ago
Earliest human ancestors appear
300,000 years ago
First modern humans (Homo sapiens)
70,000-50,000 years ago
Humans migrate across the world
12,000-10,000 BCE
Neolithic Revolution
Farming and permanent villages begin.
3500-3000 BCE
First civilizations
Mesopotamia and Sumerian cities emerge.
2630-2500 BCE
Great Pyramids of Giza built
776 BCE
First Olympic Games in Greece
27 BCE
Roman Empire begins
69-30 BCE
Cleopatra VII
Last active ruler of Ptolemaic Egypt. She lived closer in time to the Moon landing than to the building of the pyramids.
Human milestones
300,000 years ago
Homo sapiens
12,000-10,000 years ago
Agriculture
~5,300 years ago
Writing
~2,000 years ago
Christianity
~1,400 years ago
Islam
About 98% of human history happened before writing existed, so almost everything before ~3000 BCE is prehistoric.
The human line
7 million years ago
First human ancestors
2 million years ago
Early Homo species
300,000 years ago
Homo sapiens appear
40,000 years ago
Most other human species gone
Today
Only Homo sapiens remain
From Rome to the Industrial Revolution
Collapse of Rome
Breakdown of centralized authority
Rise of feudalism and manorial agriculture
Slow recovery of trade and towns
Black Death weakens the feudal labor system
Growth of wage labor and markets
Rise of stronger states and merchant capitalism
Atlantic trade and colonial expansion
Capital accumulation and commercial growth
England gains wealth, markets, labor, and resources
Industrial Revolution
Each step removes a constraint the last one created. Rome's collapse fragments power into feudalism; the Black Death's labor shortage breaks serfdom into wage labor and markets; Atlantic trade concentrates capital in stronger states; and once England has capital, labor, markets, and resources in one place, it tips into industrialization.