<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Sergey Rychagov on Medium]]></title>
        <description><![CDATA[Stories by Sergey Rychagov on Medium]]></description>
        <link>https://medium.com/@s.rychagov?source=rss-40d1f7b82411------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*mhtbE4f4WlW5CGa5CcJaKA.jpeg</url>
            <title>Stories by Sergey Rychagov on Medium</title>
            <link>https://medium.com/@s.rychagov?source=rss-40d1f7b82411------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 00:19:20 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@s.rychagov/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[My RTX 5090 lives in Russia. I live in Asia. Here’s how I use it every day.]]></title>
            <link>https://medium.com/@s.rychagov/my-rtx-5090-lives-in-russia-i-live-in-asia-heres-how-i-use-it-every-day-31f0801e4dd7?source=rss-40d1f7b82411------2</link>
            <guid isPermaLink="false">https://medium.com/p/31f0801e4dd7</guid>
            <category><![CDATA[home]]></category>
            <category><![CDATA[mqtt]]></category>
            <category><![CDATA[arduino]]></category>
            <category><![CDATA[esp32]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Sergey Rychagov]]></dc:creator>
            <pubDate>Sat, 19 Sep 2026 11:14:12 GMT</pubDate>
            <atom:updated>2026-09-19T11:14:12.973Z</atom:updated>
            <content:encoded><![CDATA[<p>A few months ago I moved to Asia for an extended stay. My laptop came with me; my main computer stayed behind in Russia. That computer is built around an RTX 5090, and it’s the machine where the actual work happens: local language models through Ollama, image and video generation through ComfyUI. A laptop runs neither well.</p><p>The obvious options all failed. Renting a comparable GPU in the cloud runs about a dollar an hour, and I use the machine two to three hours a day, so the months add up. Keeping the desktop powered on around the clock in an empty apartment felt wrong on every level. And asking family to press the power button is a gamble when the fix for “it froze” is “just reboot it”.</p><p>The option that worked: let the computer sleep, and wake it myself. From a phone. From another continent.</p><h3>The one insight everything hangs on</h3><p>A shut-down PC is nearly unreachable. It runs no OS, no agent, nothing that could accept a command. The single thing its motherboard still listens for is a Wake-on-LAN packet: a tiny broadcast frame that the network card hears even in a powered-off state.</p><p>But here’s the trap. That packet only helps while the machine is off. The moment Windows boots, you’re dealing with a different problem entirely: running a machine you can’t see, through a network you don’t fully trust, with no keyboard attached. Waking a computer and commanding a computer are two different tasks with two different lifetimes. So I built them as two separate channels instead of forcing one tool to do both.</p><ul><li><strong>The wake channel.</strong> A small password-protected web panel runs on a VPS I already pay for. Pressing “Power on” leaves a short command there. At home, an ESP32 board, roughly the size of a flash drive and powered by a phone charger, checks for that command, and when it finds one it screams Wake-on-LAN at the desktop until it boots.</li><li><strong>The command channel.</strong> Windows, once it’s up, reaches out on its own to a message broker on the same VPS and waits for instructions. Whatever I type in the panel gets delivered to the machine, and the output comes back to my Telegram.</li></ul><p>Both sides dial out to the server; nothing ever dials into my home network. That single property does more for my peace of mind than any firewall rule.</p><h3>What the little board actually does all day</h3><p>Almost nothing, which is the point. The ESP32 sleeps nearly all the time, drawing microamps. Every five minutes it wakes, joins the Wi-Fi, checks for a pending command, does its one job, and goes back to sleep. The first version I built skipped the sleep and just kept the radio alive in a loop, and the board was warm to the touch around the clock. I chased the heat through radio power settings before accepting the boring answer: a device that works once every five minutes shouldn’t be awake the other four minutes and fifty-five seconds. Now it barely registers above room temperature, and a phone charger powers it for months.</p><p>The subtle part is what happens when “wake the PC” and “go back to sleep” collide. The board sends its packet, the PC takes a minute to boot, and the board, being asleep, has already forgotten what it was doing. My fix was to store the “I’m waking it” flag in the small pocket of RAM that survives deep sleep. On the next wake the board checks whether the machine answers; if yes, it reports success to Telegram, if no, it sends the packet again. The goal was never “transmit a packet”. The goal is “PC is on”, full stop.</p><h3>What I can actually do from abroad</h3><p>The wake-up is the boring half. Once Windows is up, three PowerShell scripts make it usable:</p><ul><li>A listener connects to the broker and executes what arrives: keywords like <em>shut down</em>, <em>restart</em>, <em>lock</em>, <em>sleep</em>, and arbitrary PowerShell commands whose output gets sent back to me. There’s a 5-second debounce per command, because brokers replay messages and humans double-click buttons.</li><li>One command raises a reverse SSH tunnel, and suddenly Ollama and ComfyUI on the home GPU are reachable from my laptop as if they were a cloud endpoint. The 5090 works exactly as long as the task takes, then the tunnel closes.</li><li>A monitoring script reports CPU, RAM, disk space, and GPU temperature and VRAM load straight into the chat. From ten time zones away, “the 5090 is cold and idle” is a surprisingly comforting sentence.</li></ul><p>And a guard rail against myself: if I wake the PC and forget it, a tray app notices an hour without mouse or keyboard activity, warns me, and shuts the machine down. There’s a 30-second window to cancel. So far I’ve needed it twice.</p><h3>The honest limits</h3><p>No setup survives contact with reality unharmed, so the fine print. Wake-on-LAN is a one-time setting buried in the motherboard’s BIOS, and forgetting to enable it before leaving is the classic way this project fails; test from outside the house, not from your couch. The whole scheme also assumes the home internet and power stay up; a dead router means no wake-ups, though at least the board reports its Wi-Fi status on request. A Windows that’s hung hard can’t be managed remotely either: the scripts need a live OS, and Wake-on-LAN is useless on a machine that’s already powered on. The only real cure for that is a smart plug cutting the mains, which I now keep as a backup layer. And the wake channel carries up to five minutes of latency by design, fine for “have it running by evening”, not fine for “right now”.</p><h3>What it added up to</h3><p>The board cost about five dollars. The charger and the VPS were already in my life. Three scripts took an evening each to write and debug with the help of an AI assistant, which is great at turning a precise description into code and useless at the two decisions that actually matter here: that wake-up and control must be separate channels, and that one flag has to survive a reboot.</p><p>Two or three hours of daily use would cost me a cloud GPU rental that dwarfs the whole setup within a couple of months, or ten to twenty dollars a month of electricity for a machine that spends the other twenty-two hours heating an empty room. Instead, the only thing running at home is a board the size of a matchbox, and the big machine wakes up when I ask it to, does its job, and puts itself to bed.</p><p>If you want the same thing, the product page with the details is here: <a href="https://rsa-labs.dev/robotics-lab/esp32-power/?utm_source=medium&amp;utm_medium=article&amp;utm_campaign=turn-on-home-pc-from-vacation">ESP32 Power Switch</a>.</p><p>The one scenario I still haven’t solved cleanly is a Windows that’s frozen solid. <strong>If you’ve solved that remotely: smart plug, hardware reset IC, something else?</strong> I’d love to hear it in the responses.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=31f0801e4dd7" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>