Point Element

Audio Upgrade - Presonus Eris Studio 5

2026-10-16T10:47:54.247+05:30

For a long time I had Sony SRS-D9 as the desktop speaker. Bought it for 5500 Rs in 2012 and it was used sparingly. Primary reason being I never had a proper setup

for that. And second reason was it was very much bass heavy. Also the bass being boomy did not help a lot.

Finally after 14 years we sold it and I decided to get something much better. Turns out now a days there are many options. The studio monitors are quite affordable now. In fact for about 20-25k there are excellent choices like Kali Audio LP6-v2. It was slightly out of budget and having availability issues. So I looked around and finalized Presonus Eris Studio 5. This one is a proper monitor with a 5” main driver and a separate tweeter. It’s rated at 80W and supports RCA, TRS and XLR balanced inputs.

I was able to find some space on my desk and this cost me about 16000 INR. Came with a nice power cord. Thankfully I had ordered a 2.5mm to RCA cable. And also looked up how to set output to mono on windows as I right now have only one speaker. So I was able to get started quickly.

This is not the ideal solution though. And definitely I can hear the noise seeping in. I had known there would be noise with RCA cable and a desktop. It also means that I need to put more money and get an audio interface which will get digital audio out from PC. And then I can use XLR audio cable for clean audio from interface to the speakers. Yeah more money but the sound is something.

Thankfully I have found couple of Mackie or Behringer interfaces and they are about 6-7k so quite cheap. Or I can spend double and get a very nice high quality audio interface.

So basic expense to kit out the system properly another 23-25k to get the best out. But that’s for another month or another pay cycle. As it is, it’s pretty good. Well, so the gaming system is coming around rather quite well. Cheers!!

Danlite ELM327 Bluetooth OBD-II scanner with TVS 6-pin to 16-pin adapter

2026-09-24T16:37:06.566+05:30

Got this scanner for the car and scooter for 900rs. Good investment. Now I should be able to reset the yellow engine light on the scooter. As well as scan the car for exact fault codes. The stupid Tata service center does not seem to do a great job. Anyway that’s it for now. Cheers.

Aftermath of the Upgrade - Desktop 2.0

2026-08-27T21:41:34.594+05:30

It's actually Desktop 3.0 - 1.0 being a Pentium Dual Core with 2GB RAM on an Intel motherboard - don't even remember the model. But it had an AGP port and I remember adding a NVidia MX4000 GPU to that. And it was a decent machine. After few years it got donated to nearby school. So good EoL. Desktop 2.0 is not sitting in a corner - waiting for memory prices to come down so that I can finally return it's old 2666MHz 16GB sticks and maybe return the old-ass Zebronics SSD too. It would probably go for 12-15K, and then maybe if I manage to get a nice high-end GPU, then I would combine this junk with the RTX 3050 and sell it for 40k. Would work out. Or just sell the parts one by one. Whatever.

Anyway back to the Desktop 3.0, the desktop is quite good at 2-tasking as long as those tasks are consuming different storage media. The CPU runs well, memory is slow-ass being 2666MHz. And yes I did pay 7k for a 8GB 3000MHz module but it does not matter since dual-channel is still better than 333MHz ram speed. Just for gags, I asked Gemini AI about which RAM I should pair with AMD Ryzen 5 5600 and it said to get 3600 MHz RAM, and yes that is as bad as DDR5. So might as well spend an arm and a leg and get AM5 and everything else. No!

Anyway, coming back to the performance, it's good. The CPU is running well. Cabinet is airy and things are cool, and even the RTX3050 GPU is running well. Wiring the damn thing was a pain in the ass, and it took half-a-day to figure out optimal cable routing etc and then got to disassemble it twice because the mainboard took 6 screws, and another time because the RGB lighting won't work.

Well the RGB - or rather poverty spec RGB lighting, just blinks on various settings. Maybe I'll spend some amount and just get a bunch of true RGB fans. Anyway right now it seems okay.

The storage is limiting but then all these are used low perf disks and they don't really matter. Another couple of years and we can get a 2TB Samsung SSD for an arm. The OS installation could have been challenging but Windows did not even bother with license related prompts - asked me to login to Microsoft account and then it installed all the drivers. Had to grab some from MSI wesbsite and then it was all good. Getting Ubuntu did not work well, because turns out secure boot related issues meant no more MBR. But then I figured out F11 gets us multi-boot options.

So all good on that front.

Alright so what next? A nice 27Inch OLED and the Eris Pro 5 is en queue. Cheers!!

Upgrading the desktop

2026-08-19T09:09:28.889+05:30

Finally I have decided to upgrade the desktop. I have booked a Ryzen 5 5600 with a B550M motherboard. I would have liked to put in another 16GB RAM to make total 32GB but since I'm not changing the graphics card, it makes sense to stick with 16GB. I'm also going to get a better cabinet and that will make this another PC. Then I can use whatever parts from the old system and when next year I upgrade the new system further with a new graphics card if things become cheaper, I'll sell the old system as a basic gaming CPU with a RTX 3050 which should fetch decent price for it.

So overall I would have spent 45000 + 25000 = 70000 on this system. And I think the performance is worth it.

Cheers!!

Screwing with AI on a core ultra 7 155U

2026-08-11T18:11:52.902+05:30

I was getting bored with work so decided to fiddle with wsl Ubuntu and AI. I decided to go big and downloaded a 14B model in ollama. The office laptop has 32GB of ram. It’s waste because coding takes hardly 10-12GB. Anyway there was a proxy issue with wsl which I solved by importing certificates from windows. Then got ollama installed. Then got a 1.3B model installed for autocomplete. And then got the 9GB 13B parameters model.

Running the 13B model on CPU was a pain. But the integration with continue worked decently. I ran into some code applying issues but nothing crazy. Only issue was slowness.

Next I asked copilot how it can be better. It asked me a bunch of questions to figure out if the GPU is accessible in Linux. Turns out not. After a bunch of tries I reached a point where intels openVINO runtime was able to detect the Gpu. Then I asked the actual question that how much this gpu will help and the answer was 50% which is not really much. So I stopped. But it was rollercoaster journey and fun time. Cheers!!

Periods when to make money

2026-06-11T09:40:55.608+05:30

The state of Alcatel Pop 4 Tablet

2026-03-03T17:09:00.000+05:30

This tablet had kind of died few months back. I tried to revive it using the Raspberry Pi 5V 2.5A power adapter but it did not work. Today after few more months I decided to use AI to see if it can be resuscitated. I used the power station this time and saw it's taking 2W charge for some time and later it went to 6W. And then voilla!! It powered on and showed battery image on the screen with battery in red. I was completely floored.

Anyway I queried chatgpt and it said the tablet is recovering. And it asked me to let it charge for about 45 minutes and then start it. I did that and after 45 minutest the tablet did start up. But then I ran into next roadblock. The first problem is old google account. So added a new account and removed the old - probably dead - account. Next hurdle - no updates for apps. Did even more research and found out it's version 6 of android is outdated and apps need at least version 8 or 10. By this logic my realme 3 will also go dodo in next couple of years.

Anyway right now there's no other fix. No way to get a custom rom that has all the hardware working and won't break. I saw the issues with custom rom with my earlier phones and if you don't have a popular phone - custom rom's don't make any sense. Anyway right now I'm going to let it charge fully and use it for a while. Also going to turn off WiFi and see how long it stays charged. Found out my Realme 3 stays alive way longer when I turn off the wifi. SO great from storage perspective. Maybe same thing here works. Lets see.

I think I'll keep charging this tab and keep it alive. And maybe we find some use case for it. Or it goes into junk electronics bin. That's what happens to outdated hardware. Cheers!!

How to delete duplicate file on linux

2026-02-11T19:18:00.001+05:30

fdupes is a program that can help.

fdupes -r . -d -N

Above command will recursively scan current folder & subfolders, and keep first instance and remove following duplicate instances.

Playing with audio Pt.2

2026-02-08T20:06:00.001+05:30

Audacity could not handle it properly so chatGPT produced below ffmpeg command to split it.

ffmpeg -i input.wav -map 0:a -c copy -f segment -segment_time 3600 out_%02d.wav

We can still use audacity to see the waveforms and targeted playback.

Playing with audio

2026-02-08T19:48:00.001+05:30

I recorded an audio of 34 hours as a test. Wanted to go through that hour by hour and figure out a good way to do that. ChatGPT suggested Audacity. I ended up with version 3.7.7. The installer worked fine on windows.

But the instructions chatGPT provided, did not work exactly but I was able to find the menus and split the file into hour chunks by using labels. It although detected only 14 hours of audio. The split chunks played fine, so seemed like some kind of issue with either data or the audio-file. In the recording software I checked the size and playback and it was fine and the file size reported of ~10GB was also fine. Then further R&D on chatGPT told me it could be erroneous header. It provided a command using ffmpeg to fix that so that Audocity sees full 34 hours of file. Below is the command

ffmpeg -err_detect ignore_err -i input.wav -c copy fixed.wav

But first I had to install ffmped. That involved downloading a latest build, then creating c:/ffmpeg. And then edit path so that we can run the command. Thankfully it worked fine.

After this I have to again split the file if Audacity sees it properly. Splitting has a good side - the 10GB original file loads very slow. But the split files are easy to handle and load quite fast. The labels have numbering and provide enough context to figure out timestamps from start.

Lets see how it does.

How is my preowned desktop doing

2026-01-29T21:16:00.001+05:30

Well its about a month and half since I got the desktop. Since I have put a bunch of AI models on the desktop in both windows and Linux. On windows I am mostly gaming. On Linux I'm doing mostly college-projects and I'm also doing AI things. But one thing I have found out is local AI models are just not up to the mark. And with 4-5 free providers we have enough free AI that we are not limited to local AI. Another problem is I could not get the VS code Continue extension to really do great. It was kind of average and it was not really smart.

So we are not really doing vibe coding. Looks like vibe coding is an expensive hobby. Anyway I do plan to put in some efforts to run some visual models and learn more in coming months. Right now though I'm in survival mode. So I have put everything on hold for next six months.

So was it a good value? Very much so. Blowing 50-60k on PS5 would not have given me this much value. Also I could spend another 50K and upgrade this to a serious system with much less efforts. And that works for me. Cheers!

Canon 200D image coloring tips

2026-01-09T17:46:00.001+05:30

Color grading JPG photos from a Canon 200D requires a slightly more delicate touch than RAW files because JPGs have less "dynamic range" (the amount of detail stored in highlights and shadows). In GIMP 3, you have access to powerful non-destructive-like workflows and high-bit depth processing that make this easier.

Here is a professional workflow tailored for your Canon 200D JPGs.

1. Preparation: The "Do No Harm" Setup

Before you touch the colors, you need to protect your original image data.

Duplicate Layer: Press Ctrl + Shift + D. Always work on a duplicate so you can compare with the original.
Check Precision: Go to Image > Precision. Even though your JPG is 8-bit, GIMP 3 works best if you set this to 32-bit float and Perceptual gamma (sRGB). This prevents "banding" (ugly stripes in gradients like skies) while you edit.

2. Color Correction (The Foundation)

You must "fix" the photo before you "grade" it. Canon JPGs tend to have a slightly warm, reddish "skin tone" bias.

White Balance: Go to Colors > Auto > White Balance. If it looks too robotic, undo it (Ctrl + Z) and use Colors > Levels. Use the middle (gray) eyedropper and click something in your photo that should be neutral gray (like a concrete road or a grey shirt).
Shadows-Highlights: Canon 200D JPGs often crush blacks. Go to Colors > Shadows-Highlights. Raise the "Shadows" slider slightly (5–10%) to recover detail in dark areas without making the photo look "flat."

3. The Color Grading Process

This is where you add the "vibe" (cinematic, vintage, or punchy).

A. The S-Curve (For Contrast & Depth)

Go to Colors > Curves.

Value Channel: Create a subtle S-curve by clicking the line to add a point in the top third (drag up slightly) and the bottom third (drag down slightly). This makes highlights pop and shadows deep.
Color Channels: Switch the dropdown from "Value" to Blue.
- Warm/Golden Look: Pull the bottom of the blue curve down (adds yellow to shadows).
- Cinematic/Teal Look: Pull the top of the blue curve up (adds blue to highlights) and the Red curve slightly down (adds cyan).

B. Color Balance (The "Film" Secret)

Go to Colors > Color Balance. This is the most intuitive tool for grading.

Shadows: Add a tiny bit of Cyan and Blue for a modern, "cool" shadow look.
Highlights: Add a tiny bit of Yellow and Red to make sunlight feel natural.

C. The "Orange and Teal" Pop (Optional)

Go to Colors > Hue-Saturation.

Select the Cyan or Blue radio button and shift the "Hue" toward Teal.
Select Yellow or Red and boost the "Saturation" slightly to make skin and wood tones "glow."

4. Final Polish

Vignette: Go to Filters > Light and Shadow > Vignette. A subtle dark edge focuses the eye on the center of the frame.
Sharpening: Since JPGs are already sharpened by the camera, use Filters > Enhance > High Pass on a new layer (set layer mode to "Overlay") for a more natural look than "Unsharp Mask."

Pro-Tip for Canon 200D Users

The 200D has a "Picture Style" setting in-camera (Standard, Portrait, Landscape). If you find your JPGs are always too saturated or sharp, try setting your camera to the "Neutral" or "Faithful" Picture Style. This gives you a "flatter" JPG that is much easier to color grade in GIMP later.

Would you like me to walk you through how to create a "Cinematic Teal & Orange" look specifically using the Color Balance tool?

Canon 200D color enhancing guide by Gemini AI - It works!

2026-01-09T17:44:00.001+05:30

📸 Mastering the Canon 200D in GIMP 3: The Color Grading Guide

If you are shooting with a Canon 200D (SL2), you know that the "Canon Colors" are beautiful but the JPGs can sometimes feel a bit "heavy" or overly warm. This guide walks you through a professional-grade workflow to transform your photos using the power of GIMP 3.

Phase 1: The "Digital Negative" Setup

Before you touch the colors, you must maximize the data available in your 8-bit JPG to prevent the image from "breaking."

Create a Safety Layer: Press Ctrl + Shift + D to duplicate your image.
Enable High Precision: Go to Image > Precision and select 32-bit float and Perceptual gamma (sRGB). This prevents "banding" (ugly stripes) in your gradients.
Recover the Shadows: Go to Colors > Shadows-Highlights. Set Shadows to 12 to lift the dark areas without making the photo look flat.

Phase 2: Technical Correction

Fix the "foundation" of the photo before adding the creative vibe.

Levels Tool: Go to Colors > Levels.
Neutralize: Pick the Middle (Gray) Eyedropper and click on something in your photo that should be neutral (like a grey stone or a white wall).
Black & White Points: Move the outer sliders until they just touch the edges of the histogram peaks.

Phase 3: The Creative Grade

This is where you give your photo its unique "look."

The "Matte" Curve (Cinematic Feel)

Open Colors > Curves. Grab the point at the very bottom-left (the blacks) and pull it slightly upward. This turns deep, "crushed" blacks into a soft, cinematic charcoal gray.

Split Toning (Color Balance)

Open Colors > Color Balance and apply these targeted shifts:

Shadows: Add +4 Blue and -2 Red (Cyan). This creates moody, deep shadows.
Highlights: Add +3 Yellow and +2 Red. This mimics the warmth of natural sunlight.

Phase 4: Pro Sharpening

Instead of using standard sharpening, try the High Pass method:

Create a "New from Visible" layer (Shift + Ctrl + Alt + E).
Go to Filters > Enhance > High Pass.
Set the Layer Mode (in the layers dock) to Overlay. Adjust the opacity until it looks crisp but natural.

Pro-Tip for Canon Users: For the best results, set your 200D's in-camera "Picture Style" to Neutral or Faithful. This gives you a "flatter" image that responds much better to color grading in GIMP later!

CMake useful shortcuts

2025-12-11T20:40:00.000+05:30

Add below code block to .bashrc and then you have these shortcuts
available to make life easy.

# === CMake Shortcuts ===
alias cmkconf='cmake -S . -B build' # Configure
alias cmkbuild='cmake --build build --parallel 8' # Build with 8 threads
alias cmkinstall='sudo cmake --install build' # Install to system
alias cmktest='ctest --test-dir build -V' # Run tests

Got a desktop PC

2025-12-05T21:39:00.001+05:30

I found that my laptop was lacking when it came to AI related stuff.
So now the thing is I could buy a new desktop but the prices were
quite crazy and it did not seem like a good value. So I did a bunch of
R&D and finalized that I should be buying a refurbished desktop and
then upgrade it in limited capacity.
So I looked around and bought an Acer corporate refurbished desktop
for 19500. Then I added a beefier PSU and a graphics card to that by
spending 24000 and some change for some fans and sata cables etc. So
now finally I have below config:
Intel core i5 8500T 2.1GHz
Acer H310ch5-M19 mainboard
Zebronics 256GB SSD
500GB HDD
The desktop came with a 200W generic PSU. I replaced that with a
CoolerMaster 550W 80 Bronze PSU.
Added a Zotac RTX 3050 8GB graphics card.
I had an extra WD Green 240GB SSD here with me, so ordered couple of
SATA cables and plugged it in. That took total storage to almost 1TB
and it also gave me a separate drive to install Linux on.
So have created a separate user account on windows. Have to install
and setup Linux. And then setup all the other tools and stuff on that.
But it's going to take time that I do not have right now.
Lets see if this things earns me something or I end up using it for
playing games!!

A FULL PRICE-TO-TOPS VALUE TABLE for all major AI hardware

2025-11-22T11:08:00.001+05:30

Recently I was doing some research on various hardware available in
all segments. I did not expect below results.
The surprising thing was the amount of options available and the
various price points for that. Anybody looking into AI will definitely
find this helpful.

Global MSRP (approx.)

Converted to Indian Rupees (₹) using a neutral rate:
$1 = ₹85

Computing ₹ per TOPS so you can instantly see value-for-money.

⚠️ Prices vary, but this gives clear comparative value.

________________________________

🔥 PRICE PER TOPS (₹ per TOPS)

Sorted from BEST value → WORST value

________________________________

🟩 1. Desktop GPUs (Best Value in the Consumer Market)

Hardware TOPS Approx Price ₹ Price ₹ per TOPS
RTX 4090 1500 $1599 ₹1,36,000 ₹90 per TOPS
RTX 4080 Super 780 $999 ₹85,000 ₹109 per TOPS
RTX 4070 Ti Super 450 $799 ₹68,000 ₹151 per TOPS
RTX 4060 Ti 353 $399 ₹34,000 ₹96 per TOPS
AMD RX 7900 XTX 300 $999 ₹85,000 ₹283 per TOPS

➡ Winner: RTX 4090 & RTX 4060 Ti (best value-per-TOPS)

________________________________

🟦 2. Embedded AI Accelerators (Industrial / Edge)

Hardware TOPS Price ₹ ₹ per TOPS
Hailo-8 (AI Hat / M.2) 26 $199 17,000 ₹653 per TOPS
Hailo-10H 40 $299 25,000 ₹625 per TOPS
Coral Edge TPU USB/PCIe 4 $80 6,800 ₹1700 per TOPS
Sipeed Maix4-HAT (INT4) 72 $170 14,500 ₹200 per TOPS
Sipeed Maix4-HAT (INT8) 18 $120 10,000 ₹555 per TOPS

➡ Winner: Sipeed Maix4-HAT (INT4) = ₹200/TOPS

________________________________

🟥 3. Jetson Modules

Hardware TOPS Price ₹ ₹ per TOPS
Jetson Orin Nano (8GB) 40 $499 42,000 ₹1050 per TOPS
Jetson Orin NX (16GB) 100 $699 59,000 ₹590 per TOPS
Jetson AGX Orin (64GB) 275 $1599 1,36,000 ₹495 per TOPS
Jetson Xavier NX 21 $399 34,000 ₹1619 per TOPS

➡ Winner: Jetson AGX Orin = ₹495/TOPS

________________________________

🟨 4. Qualcomm / Mobile Dev Kits

Hardware TOPS Price ₹ ₹ per TOPS
Snapdragon 8 Gen 3 Dev Kit 75 $799 68,000 ₹907 per TOPS
Snapdragon 8 Gen 2 Dev Kit 49 $699 59,000 ₹1204 per TOPS
Qualcomm RB5 (865) 15 $379 32,000 ₹2133 per TOPS
Qualcomm RB3 (845) 3 $199 17,000 ₹5666 per TOPS

➡ Winner: SD8 Gen 3 dev kit = ₹907/TOPS

________________________________

🟧 5. MediaTek Genio + Dimensity Dev Boards

Hardware TOPS Price ₹ ₹ per TOPS
Dimensity 9300 Dev Kit 46 $899 76,000 ₹1652 per TOPS
Genio 1200 4.8 $199 17,000 ₹3541 per TOPS

➡ Mobile chip dev kits are not value hardware, they are R&D tools.

________________________________

🟪 6. Rockchip AI Boards

Hardware TOPS Price ₹ ₹ per TOPS
RK3588 AI Boards (Orange Pi 5+, etc.) 6 $109 9,500 ₹1583 per TOPS
RK3399Pro 3 $149 12,500 ₹4160 per TOPS

➡ RK3588 boards = best Rockchip value

________________________________

🟫 7. SOPHGO Industrial Accelerators

Hardware TOPS Price ₹ ₹ per TOPS
BM1684X 32 $599 51,000 ₹1593 per TOPS
BM1688 70 $699 59,000 ₹842 per TOPS
SG2042 RISC-V 8 $199 17,000 ₹2125 per TOPS

➡ BM1688 = ₹842/TOPS

________________________________

🟫 8. Apple iPhone Neural Engine (MSRP-based)

(prices relative to launch MSRP)

iPhone TOPS Launch Price (₹) ₹ per TOPS
iPhone X (A11) 0.6 89,000 ₹148,333
iPhone 12 (A14) 11 79,000 ₹7,181
iPhone 13 (A15) 15.8 79,900 ₹5050
iPhone 15 (A17 Pro) 35 1,34,900 ₹3854
iPhone 16 (A18) 40 1,39,900 ₹3497

➡ Newer iPhones increased TOPS faster than price → good AI value but
still very expensive per TOPS compared to GPUs.

________________________________

🟦 9. Raspberry Pi Ecosystem

Hardware TOPS Price ₹ ₹ per TOPS
Pi 4/CM4 (0.4 TOPS est.) 0.4 ₹6,000 ₹15,000 per TOPS
Pi 5/CM5 (0.4 est.) 0.4 ₹7,500 ₹18,750 per TOPS
Pi 5 + AI Kit (Hailo-8) 26 ₹16,000 ₹615 per TOPS

➡ Pi 5 with AI kit = actually VERY GOOD value.

________________________________

⭐ SUMMARY OF VALUE (₹ per TOPS)

Best Value Hardware (₹/TOPS low → high)

NVIDIA RTX 4090 → ₹90/TOPS

RTX 4060 Ti → ₹96/TOPS

RTX 4080 Super → ₹109/TOPS

Sipeed Maix4-HAT (INT4) → ₹200/TOPS

Jetson AGX Orin → ₹495/TOPS

Hailo-10H → ₹625/TOPS

Hailo-8 → ₹653/TOPS

BM1688 → ₹842/TOPS

Snapdragon 8 Gen 3 dev kit → ₹907/TOPS

Pi 5 + Hailo-8 → ₹615/TOPS (very good!)

➡ Absolute best performance-per-rupee: NVIDIA consumer GPUs
➡ Best embedded value: Hailo, BM1688, Jetson AGX Orin
➡ Best budget option: Pi 5 + Hailo AI Kit

Using AI for work

2025-11-20T22:33:00.001+05:30

I have started using copilot AI more and more for work. It seems that
if you have a general idea of the work, it can generate the code
pretty much at about 70%. It's not perfect - for example it almost
always messes up lambda functions but that 30% of problematic work one
can fix.
Meanwhile using other AIs also provide different different outcomes.
So turns out the training data and usage matters a lot. Anyway I think
human intuition and other hard won skills still matter and when AI
screws anything then knowing the nitty-gritty of things really comes
handy. But yeah it's going to mess with people's learning habits and
probably going to make more scatter-brained people. Learning is a hard
process and the inception of knowledge is a difficult thing. Looking
around we already find so many idiots with dirt for brains, I think
for disciplines like engineering where really complex knowledge needs
to be ingested, AI is going to probably tip over people into
un-learnables. But that's problem for the future.
Anyway another topic was if I should invest in a PC that could run
local AI models. But it' s costing more than 1Lac for decent usable
setup. And that's a bit too much money right now. I guess if I had
15-20L then spending couple of lacs on such a PC would not have been
all that difficult. But anyway right now my work is getting done via
the free models. So there's that. Regarding hardware I found out that
NVidia GPUs have the best AI support right now and they do provide
230+ TOPS performance right now. Comparing that to my inbuilt Vega 7
graphics's piddly 1.6 TOPS is an astronomical increase in performance.
But I have another arena to explore and that is small models. I see in
AI generated results that they could also perform certain tasks well
enough. So I'm going to explore that. Lets see if we find something
meaningful there. Cheers!!

Office laptop fan failed

2025-11-17T17:21:00.001+05:30

So apparently this is a thing with Lenovo T14 laptops. The fan starts
making high pitched whiny noise and becomes useless.

https://www.reddit.com/r/Lenovo/comments/1lnbmir/my_lenovo_thinkpad_t14_gen_1_fan_started_making/

Going to have to go to office and get it fixed.

Non-tech: Regarding petrol NA vs Turbo engines

2025-10-23T10:56:00.001+05:30

I owned a Maruti Celerio VXI AGS MY2020 for almost 5 years. And recently switched to a Tata Nexon Creative S+AMT. This post will talk about the difference in engines on both these cars and the effect it has on driving dynamics and mileage.

First of all Maruti has 1L NA engine which is tuned frugally and returns very good mileage. I have seen it return 15kmpl in city with AC ON. And on highways it can hit 18-19kmpl fully loaded and with AC ON.

Although this car is an economy hatchback and is not really tuned to cruising. That's why at 80-90kmph, engine is around 2800-3000 RPM and good mileage is observed around 80kmpl, and after that it drops. Plus the 1L 3-Cyl engine means not enough torque and so jerky driving in low speeds and ghats. Still for the low price even with AMT gearbox, very good value for money.

Moving on to Nexon, the engine is 1.2L turbo-petrol. The turbo really spins at around 2000RPM and provides max boost. Although the engine maps provided are such that there's enough power at low end also. There are 3 modes provided eco, city and sport. They primarily change AMT gear shifting points along with engine power characteristics. So in economy mode power is less and AMT shifts to higher gears quickly. So this mode is very good for sedate driving. City mode increases power and makes gear shifts longer, and is good for usual city driving with enough power. Sport mode improves this further and usually tanks the economy except for a band of speeds where turbo provides maximum boost and runs the engine leaner and thus increasing mileage.

Looking at the turbo-effect closer, what happens is below 1500-1800 RPM the engine works similar to a NA engine with turbo not adding a lot of boost. Around 1800 RPM turbo spools high enough to improve boost and providing more power and mileage. So if we are able to cruise between 2000-2500 RPM we get optimal conditions with maximum mileage. Cruising this way on Delhi-Mumbai Expressway, I was able to get mileage of 18-19kmpl at 90kmph. This would not be possible on a NA engine.

Thankfully my use case is of primary highway usage and so this helps a lot. If the highway is busy, we can shift to sport mode and so this ensures engine RPM is high enough even at 70kmph to get max boost and better economy. That's why on many forums they mention that with sport mode, Nexon gets better highway mileage.

Let's talk about other fuel types and EVs and let me tell why I did not get a different vehicle. So first of all my current use case for next 3 years at least is primarily highway usage. CNG was rejected due to less number and overcrowding at the fuel stations. Diesel was rejected since my vehicle does not see daily usage and is stationary for the most part. Plus with BS -VI part 2, diesels have become much more complex and difficult to maintain with DPF related issues and what not. Also the arbitrary govt bans don't help. Last left are EVs. Looking around EVs have neither become cheaper in last 5 years nor have they become defect free. In Indian market as of 2025, there's no cheap and reliable EV. I really wanted to switch to an SV for low running costs but the headaches are just too much as of now. Unfortunately the Japanese auto makers seem to be dragging their feet so no high quality EVs in the market as of now. The homegrown car OEMs have EVs but the quality control leaves a lot to be desired. For example if the turbo blows up on my Nexon, it could cost about a lac or so, but if the battery blows up, and the OEM is not supporting, that would leave me in a hole of 5-6Lac. And that means scrapping the car and buying a 2L economy hatch again. So no, EVs need 5-10 more years to mature.

Well, long post but it does cover all I wanted to talk about. So far the Nexon seems to be decently reliable. Hoping it remains the same. Cheers!!

Replacing the western digital 1TB hdd

2025-10-17T15:54:00.001+05:30

Bought a 1TB Samsung T7 SSD to replace the western digital one.
Cost was 9000 INR. Quite costly but I'm hoping it pays off.
It's specs as below:

Samsung T7 1TB Up to 1,050MB/s USB 3.2 Gen 2 (10Gbps, Type-C) External
Solid State Drive (Portable SSD) Grey(MU-PC1T0T)

Performance of up to 1,050 MB/s, up to 9.5x faster than external HDD
Interface: USB 3.2 Gen 2 (Up to 10Gbps). Backwards compatible. The
host device and connection cable must support USB 3.2 Gen 2 to reach
the maximum speed.
Durable: Shock resistant internal frame lets the T7 withstand falls
from up to 2 meters
Security: Optional password protection and AES 256 bit hardware encryption
Windows 7 and higher, Mac OS X 10.10 and higher, or Android 5.1
(Lollipop) and higher.
Includes USB Type C to C and USB Type C to A cables

Going to sync all the data to this one. I'm not moving the data right
away. But I'm going to slowly move all of it to here while keeping
other copies. If there's no issue, I'll get rid of unwanted data.

All the earphones/headphones/megaphones yada yada

2025-10-14T20:44:00.001+05:30

Current ones - Sennheiser HD439, Marshall Monitor II, Sony WI-C100

Past Ones from latest to oldest: Realme buds, Philips Wired Earbuds,
Klipsch Image S3, Panasonic Headset, Generic Blues headset, Sony Wired
earbuds, Generic earbuds.

In last 20 years I have been through a bunch of earphones and
headphones/headsets. I started with FM Radios and generic earbuds in
my college days. This is around 2007-08, seems like an eternity away.
We did not have mobiles back then. And the FM radios would take pencil
cells. And would need new cells every fortnight. Anyway the music was
good to ears. The first major upgrade was when I got my first Sony
Mobile and it had inbuilt FM. I bought a decent pair of Sony earbuds
at 600 bucks back then. That was $10 in 2008 I think. Those sound
astronomically better than my earlier generic earphones. They did not
last all that long I remember. Then thanks too recommendation from one
of my gamer friends, I bought a Blues headset which was locally made
but sounded very good at 1/3rd cost of Sony. It lasted 3-4 years too.
Then as usual when one speaker failed, I did not know what to replace
it with. Then one day while roaming around in a mall, a Panasonic
Headset with HD sound label caught my eye. It cost same as the Sony
and this is probably 8-10 years later. Bought it and used it a lot. It
did not have a lot of bass but very good sound-stage and overall clear
sound.
It lasted long and I bought my next pair of earbuds only when I
switched to a new job and it involved a lot of commute. So I looked
around in audiophile circles and bought a Klipsch Image S3. It was a
decent upgrade. But did not last very long. About a year and half
later one side died. Then not wanting to spend a lot, I bought some
cheap Philips earphones. Their construction was just cheap but the
sound was very decent. Only problem - about a year and half later,
they literally fell apart. Finally I decided to spend big bucks and
bought Sennheiser HD 439's. This was a great decision and they have
lasted long even with my rough use. So much so that when I wanted to
buy a pair of Bluetooth Noise Cancelling headphones, I spent big and
bought Marshall Monitor II's last year. Cost ~12000 INR, that's about
~$140. The noise cancelling is quite good but they don't seem to like
a lot of interference.
Somewhere in between I bought a pair of Realme Buds - but they fell
apart in about a ear with their flimsy construction. Later I bought
Sony WI-C200's - wireless neckband type earphones. And my primary use
case for these is Teams calls. I don't use them for music since they
are just too bass heavy. Later I played around a bit with Sony Music
Connect app and tuned them a bit to sound better for music. But still
they are best for calls only. And I have all the others for music and
movies.
P.S. I bought a 2.1 Sony System for desktop-audio and it's going
strong for last 13 years. It's just a 2.1 but is so much loud that the
windows shake if we turn up the volume to max. So Cheers for Sony!!

Data Managment Ep 2

2025-10-14T20:07:00.000+05:30

So I'm in the middle of another HDD death. This time it's my WD 1TB.
And by the name of this post I'm referring to my original post which
talked about my earlier WD 320GB external HDD which died. And somehow
I never made a post about it. And in fact I have never made a post
about the tens of headphones which I tried and they died. Maybe I
should make a catalogue of them. I don't really know what value such a
thing to be for anyone except for a story. Anyway I don't listen to
much music anymore. I just watch soaps most of the time like an old
man. Anyway the thing is I observed that this drive was acting up for
some photos when I copied them to the drive. It would hang up. So I'm
thinking maybe it's about to die. Ideally I should buy another drive
but as usual money is tight and so I just moved data here and there
and then left non-critical data on the drive.
Got some time today to dig into and found a very good program called
Victoria that analyses the drive and tells you how good is it doing. I
checked CrystlDiskInfo first and it flagged a warning about pending
sector count. Anyway now I'm scanning it and will try to see if it can
be fixed and if I could use it to store some junk.
So long story short, I went on Amazon and bought a Samsung 1TB SSD.
That should last a while. Plus I have another WD 240GB SSD in a USB
enclosure. That is my second copy for important data. I would have a
third copy on laptop but I'll probably give my laptop to wife and
start using her awful Asus Vivobook. But that's alright, because for
project work I am thinking to switch to the Old Blue Lenovo. It's slow
as hell but does not matter. Works fine for that use case.
What else? I wanted to compile my code for Raspberry Pi and run it.
Thought to setup a cross-toolchain so did that. The Beaglebone Black
has too many issues with performance and wifi and other things thanks
to TI chip inside it. So Raspberry Pi is anyway better candidate. The
Nucleo 144 will not see any sun for a while now since I don't have any
embedded work.
Alright, what else? Ah I'm not using the marshalls bluetooth
headphones a lot. But I'll have to ensure the battery does not drain
completely. Have to keep them charged a bit every few weeks. Meanwhile
it was good that I spent a bit of money and got Old Blue Lenovo
working well enough again. It's good to have a decent backup laptop.
And exchanging it also does not make sense as I may get at max 800-900
rupees. The work I can do on it for a day is worth 1000 rupees. Ok
long rant about life and electronics. But yeah, overall things are not
too shabby. Cheers!!

Playing with AI - Pt. 2 and some Beaglebone Black fun

2025-09-30T22:30:00.001+05:30

So I did continue playing with AI and got local GPT4All installation
on my new lenovo ideapad 3. Before that I did one thing I had been
dreading. I got rid of partition 2. Emptied it and deleted it, and
then expanded Partition 1 into that area. This was needed because my
WSL installation has been swelling up and up and I really needed more
space in C: drive. And managing junk in 100GB had become a chore.
Anyway after doing this, now I'm seeing regular hangups and issues
with volume-mgr. So there's some negative effect, but screw it. I
played around with GTP4All and found it to be a good enough solution
for local offline AI. Unfortunately I don't have any use case for such
a thing. I tried to configure it as an agent in Visual Studio Code and
things just would not work. There were errors, documentation said one
thing and another happened at runtime, and there were issues. So I
gave up after some time. So yeah tools like Cursor which give you an
amazingly well-working agent are of some value.
What next? I have been using ChatGT, Claude etc to create some
software code and they work decently. Good for a lazy person. Coming
back to Beaglebone black, I don't remember if I have mentioned it here
on this blog at any time. Let me check. Yes I did. I played around
with it and actually did not use it at all. I had bought it as a
placeholder for an office project and I had to get another dev-board
eventually on which I did all the development. So coming back to the
beaglebone black, I thougt I could check out some of these libraries
I'm making with the help from AI on this board and maybe get some more
learnings out of it. Boy oh boy, was I in for a shock!! So much
learnings.
Let me start from what the board had - a 32GB card with debian 11 on
it and working to a console GUI. I tried to get wifi to work and it
did not work. Then I used my spare Asus router that I use as kind of
wifi-extender and basically wired ethernet hub. Connected both via
ethernet cable and beagle had internet. Tried to get weston working
but no luck. Most all suggested to use X11 with OpenBox. But those
days are quite in past so I looked around and found out there is
debian 12 image. Downloaded it to SD Card and got it booted. Getting
weston to work was a pain. But a lot of R&D finally pointed to
settings of use-pixman=true and backend of drm and color of rgb565.
Very specific settings and with these only it works. Anyway got weston
working and copied my code, built demo app and had it launched. Then
ran into performance issues. Turns out debian 12 does not have any
drivers for powervr sgx gpu. And all the rendering is happening in
software. No problem for a 4.3/7Inch screen with max 800x480
resolution. I was in 1280x720 resolution. Anyway spend 8 hours trying
to go back to kernel 5.10 (debian 12 comes with 6.13) and failed to
get weston working. So lots of trial and error and lots of
frustration. Anyway at last I have given up on hardware assisted
rendering and HMI performance. I can see the apps launch and work
decently and that's good enough. Maybe I'll try to figure out GPIO
channels and try to use them. That would be awesome. Cheers!!

Playing with AI

2025-08-26T12:12:00.001+05:30

So we have gone on an AI bandwagon. Turns out AI in agent mode can be
tremendous help in generating code and documentation. Prompt AI is
good only for handling small problems but agent can integrate with
IDEs and increase productivity by a few times.
The next frontier is locally running AI and so this way one is free
from AI provider plans and the only cost is for electricity and
hardware. Now a good AI system needing a bunch of AI chips is still
costly. But the control could be worth it.
For an experiment I started on my now 10 year old Blue Lenovo laptop
with Pentium B950. This processor is so old that it does not support
AVX2 and so can not really do anything. I can still use this for
coding. Actually the bootup was way slow at about 2 mins. But with
generous help from chatgpt I was able to bring it down to a minute
which is quite good. I also removed a bunch of unneeded things thus
making it quite faster to work too.
So my goal is to use my new laptop with windows to host the AI. I see
that it can run a decent good model like codellama-7b-instruct. Then I
can figure out on how to let my coding machine talk to this hosted AI.
Hopefully I can run this in complete offline mode and see how good it
is. Maybe it needs online help for search, and I should be able to
configure something.
But that's the goal and a way to use AI at cheap. Otherwise cursor,
trae etc AI agents work great but the cost is something I don't want
to pay. Well, that's life. Something interesting going on. Cheers!!

Realme 3 firmware back to Android 10

2025-07-29T14:18:00.000+05:30

It was a pain to find latest Android 10 firmware for Realme 3. The
official website did not have anything for these old devices.
One tool I found was a paid service. And unlocking bootloader also
stopped working officially. The unofficial process was too complex.
And what baffled me was I was just unable to find stock update files
anywhere.
After a lot of searching, finally found stock Android 10 update at
below website.
It contains lot more updates besides Realme 3.
https://realmefirmware.com/realme-update/

So now I'm back to Android 10. The phone camera is probably as useless
as earlier and I think maybe it's damaged by lasers or something else.
Anyway I have a DSLR or iPhone to take photos anyway. So quest failed,
but I think this phone should run for couple more years at current 66%
SOH. It's not perfect, like the GPS conks off once in a while which I
think has to do with loose connections due to cracked frame. Anyway
let it run for as much as possible, and then I'll probably completely
switch to iPhone 15. No issues. Just want to extract maximum usage out
of this thing. Cheers!!