Trading Hardware & Quant-AI Glossary
Building owned trading infrastructure means wading through jargon — tick data, walk-forward, kernel bypass, VRAM, kill switch — that mostly hides simple ideas. This is our plain-English glossary of the terms that actually matter when you spec a trading rig, a backtesting server or the software that runs on them, with no hype and no fake precision. Each definition is one or two sentences, and where a term deserves a deeper answer we link to the guide that explains it.
A note before the definitions: TIS builds the hardware and software you own — not financial advice, signals, or guaranteed performance. Trading involves substantial risk of loss.
Market data: what feeds the rig
The terms that describe the data flowing into your charts, backtests and dashboards. Own this layer and your watchlist never leaves the building — more in the market-data feed server guide.
- Tick data
- The most granular market data: every individual trade and quote with a timestamp, before any aggregation into bars. It is the raw material a serious research stack stores and replays, and it adds up fast — which is why a tick-storage server is its own build.
- Bar / OHLCV
- Data aggregated over a time window into Open, High, Low, Close and Volume. It is what most charts and many backtests use, because it is far lighter to store and reason about than raw ticks.
- Point-in-time data
- Data stored so each query returns only what was actually known at that timestamp. It is essential to avoid look-ahead bias — get the storage layer right and your backtests stay honest before you write a single strategy.
- Time-series database
- A database optimized for timestamped data — kdb+, TimescaleDB, InfluxDB — with fast range queries and compression. Choosing and tuning one is the heart of a market-data feed server.
Backtesting and machine learning
The terms that decide whether a backtest is telling you the truth — and the failure modes that make most of them lie. The honest treatment of these is what separates a usable research box from an expensive way to fool yourself.
- Backtesting
- Running a strategy against historical data to estimate how it would have performed. Results never guarantee future outcomes — a backtest measures the past, not your account. The hardware that makes it fast is a GPU-accelerated research box.
- Walk-forward analysis
- Repeatedly training on a past window, freezing the parameters, then testing on the next unseen window. It mimics real deployment to reduce overfitting, and it is far more honest than a single train/test split.
- Overfitting
- When a model learns noise specific to the training data, so it looks great in-sample but fails on new data. It is the single most common reason a backtest that dazzles dies in live trading.
- Look-ahead bias
- Accidentally using information in a backtest that would not have been available at that point in time, which falsely inflates results. Point-in-time data storage is the defense against it at the data layer.
- Survivorship bias
- Testing only on assets that still exist today and ignoring the delisted or failed ones, which overstates historical performance. The companies that went to zero are exactly the ones a naive backtest forgets.
- Regime change
- When market behavior shifts — volatility, correlations, liquidity — so a model trained on the old regime stops working. No amount of hardware fixes a strategy that the market has simply moved past.
- Slippage
- The difference between the expected fill price and the actual fill. It is a real cost that a backtest must model, or it lies — ignore slippage and a marginal strategy looks profitable that never was.
- GPU dataframe (RAPIDS cuDF)
- A pandas-like library that runs dataframe operations on the GPU, speeding up data exploration and vectorized analytics. It is the workload where a research GPU actually earns its keep — see GPU-accelerated backtesting.
- Sentiment analysis
- Using NLP or LLMs to score news, filings or earnings calls as positive or negative. Run the model locally and the data stays in your building — a triage tool for research, not a buy/sell signal.
Hardware and latency
The terms that decide how a trading box performs — at the desk and on the wire. Most of these only matter past a point, and we will tell you honestly where that point is — see the low-latency trading PC guide.
- Single-thread performance
- How fast one CPU core runs. Charting and order-entry platforms often lean on it more than on core count, which is why a fast single thread beats a big core count for a discretionary desk.
- Core pinning / isolation
- Assigning a process to specific CPU cores and keeping the OS off them, for more deterministic, lower-jitter timing. It is a real tuning lever for an active algo box — and overkill for a swing trader.
- Huge pages
- Larger memory pages — 2MB or 1GB instead of the usual 4KB — that reduce address-translation overhead and improve latency consistency. A finishing touch on a tuned execution box, not a headline feature.
- VRAM
- Memory on the GPU. Large VRAM — for example 96GB — lets you hold big datasets or larger local models entirely on the card, which matters for GPU dataframes and local sentiment models.
- ECC RAM
- Error-correcting memory that detects and fixes bit errors. It is important for long backtests and servers where a single silently flipped bit could quietly corrupt a result you trust.
- Kernel bypass
- Moving network data straight from the NIC into the application, skipping the OS network stack to cut latency — for example with Solarflare/OpenOnload or DPDK. Real for active algo traders, irrelevant for most retail. The low-latency PC page maps who needs it.
- NIC (network interface card)
- The network adapter. Low-latency or FPGA NICs can shave microseconds off tick-to-trade time — meaningful at the algo end of the ladder, invisible to a swing trader.
- Tick-to-trade latency
- Total time from receiving a market event to your order hitting the exchange; the core HFT metric. For genuine sub-microsecond figures you are usually in colocation territory, not on a tower under a desk.
- Colocation
- Renting rack space inside or near an exchange data center to minimize network latency. It is the domain of genuine HFT and usually beyond an on-prem desk — we will say plainly when you do not need it.
Risk and operations
The unglamorous layer that protects capital — the controls and rules context every automated system should understand. Built right, it is what a live dashboard watches over; see trading risk systems and dashboards.
- Kill switch
- A control that immediately halts an automated trading system when losses, errors or anomalies cross a threshold. One runaway algo can erase a year, so a tested kill switch matters more than the strategy it guards.
- Circuit breaker (system)
- An automated limit that pauses trading activity when predefined risk thresholds are breached. It is the softer cousin of the kill switch — a pause rather than a full stop.
- Pattern Day Trader (PDT) rule
- A FINRA rule requiring $25,000 minimum equity for four or more day trades in five business days in a margin account. This is informational context, not financial or compliance advice — verify the current rule for your situation.
Where to go next
Now that the terms make sense, these guides put them to work:
- Start with custom trading computers — the multi-monitor desk rig built for all-day uptime.
- Capturing and storing your own feed? See the market-data feed server guide.
- Slow pandas backtests? Read GPU-accelerated backtesting on which workloads a GPU actually speeds up.
- Chasing lower execution time? The low-latency trading PC guide draws the honest line between tuning and HFT.
- Protecting capital? See trading risk systems, kill switches and live dashboards.
- Doing ML on your own data? Pair a research box with a backtesting server, and read the honest take on machine learning for the stock market.
We translate the jargon into a rig you own
You don't need to master every term on this page — that's our job. Tell us how you trade and we'll spec the CPU, the displays, the VRAM and the storage, then hand-build and burn-in test the machine here in Texas and set it up in person across Katy, Fulshear and the Fort Bend area. See our Texas service areas.
Still decoding the spec sheet?
Skip the acronyms — tell us how you trade and we'll spec a Texas-built rig you own outright, in plain English.
TIS builds the hardware and software you own — not financial advice, signals, or guaranteed performance. Trading involves substantial risk of loss.