Posts

Aftermath of the Upgrade - Desktop 2.0

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 st...

Upgrading the desktop

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

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 r...

Periods when to make money

Image

The state of Alcatel Pop 4 Tablet

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 ye...

How to delete duplicate file on linux

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

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.