My PhD Laptop Setup (2025): Why I Chose Portability Over Power for AI Research

---

The most common piece of hardware advice for AI researchers is wrong. Here’s why you don’t need a powerful GPU laptop, what you actually need, and the exact setup I’ve refined over 6+ years of research across IIT Kharagpur, Germany, and the US.

The question I get asked most on LinkedIn: “What laptop should I buy for ML research?”

Most answers on the internet say the same thing: buy the most powerful NVIDIA GPU you can afford. I disagree, and this post explains why — with the specific setup I’ve refined across 6+ years of research in different countries and contexts.

---

Table of Contents

• The GPU Laptop Myth
• What Your Laptop Actually Does in ML Research
• My Primary Machine: The Portability Philosophy
• My Current Laptop Recommendation for Indian Students (2025)
• The Remote Server Workflow
• Peripheral Setup: What Actually Matters
• Software and Tools
• Budget Breakdown
• Frequently Asked Questions

---

The GPU Laptop Myth

The internet is full of “best laptop for machine learning” lists that lead with GPU specs. These recommendations systematically ignore the realities of how research actually works:

Reality 1: All serious training happens on remote servers. Any research lab, university computing cluster, or company provides GPU servers. AWS, Google Cloud, and vast.ai provide on-demand GPU access for under $1/hour. Your personal laptop is not where models train.

Reality 2: GPU laptops are poor work companions. The best gaming/ML laptops weigh 2.5–3.5 kg, run hot, throttle under sustained load, and last 2–3 hours on battery. You’ll carry this to every conference, every library session, every café working session.

Reality 3: The battery life tradeoff is severe. A laptop with an RTX 4080 and 16 hours of battery life does not exist. You’re choosing between GPU power and daily usability.

The right mental model: Your laptop is a client that connects to powerful remote servers. Its job is to be the best client possible — fast startup, long battery, great display, comfortable keyboard, and enough local CPU power for data preprocessing and running small experiments.

---

What Your Laptop Actually Does in ML Research

Here’s what I actually do on my laptop day to day:

• Write and edit code in VS Code (SSH’d into a remote server)
• Read and annotate papers (PDF reader)
• Write papers and documentation (LaTeX on Overleaf, Google Docs)
• Preprocess and explore data (pandas, runs fine on CPU)
• Run small experiments locally to check code before deploying to server
• Attend video calls (Zoom, Google Meet)
• Manage experiments remotely (SSH, tmux, wandb dashboard)
• Browse, email, Slack

None of these tasks require a dedicated GPU. All of them benefit from long battery life and light weight.

---

My Primary Machine: ASUS VivoBook S Series

I’ve been using an ASUS VivoBook S series (Intel Core i5 8th Gen, 8GB RAM, 256GB SSD + 1TB HDD) since 2019. After 6 years across multiple countries, it’s still my primary research machine.

Why it’s held up:

• Weight: 1.5 kg — I carry this to conferences, on flights, and between offices without thinking about it
• Battery life: 8–10 hours of real usage with WiFi and VS Code open
• Build quality: The chassis has survived airports, rainy commutes, and being packed in a laptop bag under train luggage
• Display: 15.6″ FHD IPS — comfortable for long reading and coding sessions

What it can’t do: Train a ResNet-50 in reasonable time locally. But as I explained above — that’s not what I use my laptop for.

Current equivalent recommendation (2025):
The original model I bought is discontinued, but here are the closest current equivalents:

• ASUS VivoBook 15 (AMD Ryzen 5, 16GB RAM, 512GB SSD) — my current top recommendation for Indian students (~₹55,000–65,000)
• ASUS ZenBook 14 OLED — step up if you want OLED display and premium build (~₹80,000–90,000)
• HP Pavilion 15 (Intel Core i5) — budget option that handles the daily research workflow (~₹45,000–55,000)

What to look for in any laptop for research (priority order):

1. Battery life (8+ hours real-world)
2. Weight (under 1.8 kg)
3. RAM (16GB minimum; 8GB is a real constraint)
4. SSD (256GB minimum for OS and active projects; store datasets externally)
5. Display quality (IPS panel minimum; anti-glare coating matters for long sessions)
6. CPU (Intel Core i5/i7 12th gen or AMD Ryzen 5/7 — either is fine)
7. GPU (not a priority — integrated graphics is acceptable)

---

The Remote Server Workflow

This is the setup that makes a lightweight laptop viable for serious ML research.

Step 1: Get server access
Your institute almost certainly has a GPU cluster. Talk to your supervisor or department IT. Free options also exist:

• Google Colab (free tier) and Colab Pro (~₹1,000/month)
• Kaggle Notebooks (free GPU time)
• Lambda Labs, vast.ai, or RunPod (on-demand, ~₹60–150/hour for an A100)

Step 2: Set up SSH with VS Code
VS Code’s Remote-SSH extension lets you edit code on a remote server as if it were local. This is the most important tool in my workflow.

# On your laptop: configure SSH config
# ~/.ssh/config
Host my-research-server
    HostName 192.168.1.100
    User yourname
    IdentityFile ~/.ssh/id_rsa

# Then in VS Code: Ctrl+Shift+P → Remote-SSH: Connect to Host

Step 3: Use tmux for persistent sessions
When you disconnect, your training run should keep going. tmux keeps your terminal sessions alive:

# Start a new session
tmux new -s training

# Detach (training keeps running)
Ctrl+B, then D

# Re-attach from anywhere
tmux attach -t training

Step 4: Track experiments with Weights & Biases

import wandb
wandb.init(project="my-experiment")
wandb.log({"loss": 0.42, "accuracy": 0.87})

The wandb dashboard is accessible from your browser on any device. You can monitor a training run from your phone if you want.

Step 5: Sync data with rsync or rclone

# Sync local data to server
rsync -avz data/ yourname@server:/home/yourname/data/

# Download results back
rsync -avz yourname@server:/home/yourname/results/ results/

---

Peripheral Setup: What Actually Matters

I’ve learned that peripherals matter more than the laptop itself for daily productivity.

External monitor (highest ROI investment): An external display makes everything better — paper reading, coding, writing. I use a 24″ 1080p IPS monitor at my desk. In India, BenQ and LG offer good options in the ₹10,000–15,000 range.

Ergonomic keyboard and mouse: After developing wrist pain in my second year, I switched to an external keyboard and vertical mouse at my desk. The laptop keyboard is fine for travel; for 6+ hours of daily work, external peripherals are a health investment, not a luxury.

Webcam (for remote collaboration): If you do frequent video calls with collaborators in different time zones (which you will, in international research), a decent external webcam matters. The built-in cameras on most budget laptops are poor.

---

Software and Tools

Development environment:

• VS Code with Remote-SSH extension
• Conda or venv for environment management
• Git + GitHub for version control (every project, from day 1)

Research workflow:

• Zotero for reference management (free, syncs across devices)
• Overleaf for LaTeX paper writing
• Notion for research notes and experiment logs
• Weights & Biases for ML experiment tracking

Communication:

• Slack (lab communication)
• Zoom / Google Meet
• WhatsApp (unfortunately, this is standard in Indian research contexts)

---

Budget Breakdown (2025 India Prices)

For a student starting a research program:

Item	Budget option	Mid-range	Notes
Laptop	₹45,000	₹65,000	Focus on battery + weight
External monitor	₹8,000	₹15,000	Any 24″ IPS
External keyboard	₹1,500	₹4,000	Compact mechanical optional
External mouse	₹600	₹2,500	Vertical mouse if prone to wrist issues
USB-C hub	₹800	₹2,000	If laptop has limited ports
Total	~₹56,000	~₹88,000

A ₹56,000 setup will serve a researcher well for 4–5 years. The ₹3 lakh GPU laptop will be heavier, shorter-lived, and solve the wrong problem.

---

Frequently Asked Questions

Q: What if I want to do computer vision research that requires large models?
Use cloud compute. Even 10 hours/week on an A100 via vast.ai costs less than ₹5,000/month — far cheaper than buying a laptop with an A100.

Q: What about MacBook for research?
MacBooks are excellent research machines — great build, long battery, excellent trackpad. The M3/M4 chips can run smaller models locally. If budget allows (~₹1.2–1.8 lakh), a MacBook Air or Pro is a strong choice. For students on a tighter budget, the VivoBook S series provides 80% of the value at 40% of the cost.

Q: My institute provides a desktop with a GPU. Should I still get a lightweight laptop?
Yes. The desktop handles training; the laptop handles everything else, everywhere else.

---