[X] My Assistant

Discovered a few things about the supermicro power supply when poking about inside.
The 3x 12V rails are actually a single rail, but split via 3 current shunts. It has no way to independently regulate or switch them. It had provisions for a 4th 12V rail, but the output pad isn't used and shunt isn't populated.
There's also provisions for a -5V rail. If I add an output wire, populate positions for 2 capacitors, an LM7905, and two high speed diodes, and add a winding onto a filter choke, I'll have that rail. Oh, and I need to drill out the tapped hole in the heatsink from the back, so I can use a screw and nut to clamp the vreg, it's impossible to get at where the screw should be. Or I could desolder the LM7912, pull it and the heatsink, then solder them both back in. The transformer even has the pin connected, so dropping the rail was just to shave a few cents off the design. The -12V is made via a very similar method and all the parts are right beside it.
I think I'm going to do that just because.

Alpine edit: my parts bin was actually lacking in every single part (a linear regulator, useful small capacitors, magnet wire, white thermal paste) so I've placed a couple of assortments, a 4oz spool of 26awg (existing choke is probably 25awg, as the wire measures 17 thou overall; 26awg is good enough and I don't have to buy 1lb of it), and a 20g tube of white thermal grease. I'll likely just rewind the choke to make sure the turn count is even.

This post has been edited by Wayward_Vagabond: Feb 28 2021, 05:31

Discovered a few things about the supermicro power supply when poking about inside.
The 3x 12V rails are actually a single rail, but split via 3 current shunts. It has no way to independently regulate or switch them. It had provisions for a 4th 12V rail, but the output pad isn't used and shunt isn't populated.
There's also provisions for a -5V rail. If I add an output wire, populate positions for 2 capacitors, an LM7905, and two high speed diodes, and add a winding onto a filter choke, I'll have that rail. Oh, and I need to drill out the tapped hole in the heatsink from the back, so I can use a screw and nut to clamp the vreg, it's impossible to get at where the screw should be. Or I could desolder the LM7912, pull it and the heatsink, then solder them both back in. The transformer even has the pin connected, so dropping the rail was just to shave a few cents off the design. The -12V is made via a very similar method and all the parts are right beside it.
I think I'm going to do that just because.

Alpine edit: my parts bin was actually lacking in every single part (a linear regulator, useful small capacitors, magnet wire, white thermal paste) so I've placed a couple of assortments, a 4oz spool of 26awg (existing choke is probably 25awg, as the wire measures 17 thou overall; 26awg is good enough and I don't have to buy 1lb of it), and a 20g tube of white thermal grease. I'll likely just rewind the choke to make sure the turn count is even.

Neat.
I really need to get used to/learn about coils; that's like the one thing I never really touch. Why is it important for the turn count to be even, out of curiosity?

Interesting that modern PSU's still use linear regulators like the 7912. I wasn't sure if they would or not. I'm guessing they don't for the positive rails that get more use, though (switched, I bet).

This post has been edited by dragontamer8740: Feb 28 2021, 07:48

It's a circa 2006 power supply, and server insted of consumer. That being said, the dell one had an LM7912 in it too- I think that's just the cheapest way to get such a small legacy rail.
I've not traced out the whole area, but I suspect the choke of being common mode- if the DC currents don't cancel out across it, there will be a net DC flux, which could saturate a core not designed for it, and reduce it's effectiveness. In this case, that means the regulator having to deal with a lot more high frequency noise.
Also nearby is what looks to be a very low power rail based on it's choke and capacitor that powers something on the controls card. Board is double sided and lots of parts have that silicone cement on them.

This post has been edited by Wayward_Vagabond: Feb 28 2021, 11:25

Neural network image generation can create human faces based upon rules or samples of faces used to train it. If you want anime, then of course all your training examples would be anime images from Haruhi or Idolm@ster.....

[thispersondoesnotexist.com] https://thispersondoesnotexist.com/

If you need a realistic human face when someone demands a picture of your face, try the above URL....

Neural network image generation can create human faces based upon rules or samples of faces used to train it. If you want anime, then of course all your training examples would be anime images from Haruhi or Idolm@ster.....

[thispersondoesnotexist.com] https://thispersondoesnotexist.com/

If you need a realistic human face when someone demands a picture of your face, try the above URL....

See also:
[thiscatdoesnotexist.com] https://thiscatdoesnotexist.com/
[www.thiswaifudoesnotexist.net] https://www.thiswaifudoesnotexist.net/
[www.thisfuckeduphomerdoesnotexist.com] https://www.thisfuckeduphomerdoesnotexist.com/

(IMG:[files.catbox.moe] https://files.catbox.moe/m42ks0.jpg)
simp

I wish there was an option to temporarily set graphic cards fans to full power.
I'm not sure, but I suspect my current temperatures being above the usual (though not especially high) is because fans have cooled a bit the cpu and gpu, but not properly evacuated some residual heat, that now keep thing hotter than they could be.

I wish there was an option to temporarily set graphic cards fans to full power.

You should be able to set that in afterburner or any other program that lets you set the fan curve

This post has been edited by Scumbini: Feb 28 2021, 20:28

You should be able to set that in afterburner or any other program that lets you set the fan curve

…Unless you have a NoVidia card newer than Kepler/Maxwell 1 and are using the Nouveau drivers with reverse engineered firmware (since their power management FW requires an NoVidia signature).


Got my GBA Pokémon games' real-time clocks synchronized with reality (They have no provision for setting an actual date; just a time of day which starts in 2000). This way I can copy saves back and forth between my cartridges and Wii without losing all of the berries that I plant. Did it by sending a specially crafted GBA ROM to the GBA via multiboot (they can be sent small ROM's over a link cable and boot from them). The ROM pokes a new value into the game's RTC. I have another multiboot ROM for sending save data over the cable, and a third multiboot rom so my GBA's can imitate Gamecube controllers (mapping 'select' to the Z trigger). Two of those were readily available (in source code form) online, but I had to heavily modify the GBA-as-controller sources to make it not try to imitate joystick movements on the d-pad, analogue triggers, and so on. So now I can play GBA Pokémon on larger screens and then send the saves back to the cartridges to play on the go without losing data (berries) each time.

Why yes, I would like a Gamecube and a Game Boy Player and a component (or RGB) cable. But those cables are super expensive and making my own still requires purchasing an FPGA. And then to get non-interlaced video with the GB Player you need to run homebrew stuff, and homebrew is a little more annoying on a Gamecube than on the Wii. (The picture is a little sharper, though; it appears the Wii does some chroma subsampling, even with component video. It's on my to-do list to try a mod on the video encoder chip that I saw online to enable 480p RGB output and see if that also suffers from the same degradation).

As a side effect of synchronizing the clocks, of course, my cartridges think that 20 years have passed since I last played them.
Click for full size; pic is as described in the pic
[i.imgur.com] (IMG:[i.imgur.com] https://i.imgur.com/n98Qd9Tg.jpg)

This post has been edited by dragontamer8740: Mar 1 2021, 00:21

Traced out the area fully. It wasn't a common mode choke, just two windings in series. Board is setup to use LM79XX or LM337 regulators. I had to desolder the choke, pull the heatsink and regulator, and capacitors to trace everything out though.

After quite some time, I finally identified the existing diodes- Vishay ES1D in a DO-214AC/SMA package. It's a 200V 1A 15nS ultrafast rectifier diode.
Other diodes on the same transformer included Sirect MBR30100CT 100V 30A and Fuji Semi ESAD83-004R 48V 30A common cathode dual Schottky barrier diodes.
Best thing I had were Motorola 1N4934 100V 1A 150nS fast rectifier diodes, bog standard DO-41 axial package. I think they'll be good enough. I saw UF4007 diodes in the same SMA package, but can't justify ordering a big bag or an SMD diode kit.

…Unless you have a NoVidia card newer than Kepler/Maxwell 1 and are using the Nouveau drivers with reverse engineered firmware (since their power management FW requires an NoVidia signature).

I guess you forget just how good Nvidia on linux is after a few short years of swearing them off. I'm almost impressed they managed to make the experience worse.

I guess you forget just how good Nvidia on linux is after a few short years of swearing them off. I'm almost impressed they managed to make the experience worse.

I've sworn them off for five years as well, however I bought my last GPU six years ago (and at that point it was still Stallman's recommended supplier, just because thanks to Nouveau there was totally 100% free and open firmware for their GPU's. ATI/AMD has some binary blobs to this day that no one's bothered to reverse). He said it with the caveat that NVidia was no friend of ours, and that it was just due to the Nouveau project that he unfortunately felt he had to recommend them over AMD/ATI.

Nowadays you can't control the fans, and on account of that even though re-clocking the GPU's is possible, they don't enable that because the fans will continue running at the speed they run at in the BIOS with the card's initial clock rate and that probably will cause hardware damage.

Mine's a Maxwell 1 (750 Ti), so I still don't need signed firmware, but my card's the end of the line.
Nvidia® GeForce® Experience™

Anyway the real solution is to connect the fan directly to the 12VDC rail

This post has been edited by dragontamer8740: Mar 1 2021, 00:26

Any "bad" software players still out there that can't re-sync audio-video? One of my converted movies has an audio delay of 1s 1ms. Can't be bothered to convert the audio portion to Opus because bizarrely the source only contains a single lossy Dolby Digital (2-channels) audio stream. It appears The only video types which have no audio delay (based on metadata read by Mediainfo) are Bluray remux.

Any "bad" software players still out there that can't re-sync audio-video? One of my converted movies has an audio delay of 1s 1ms. Can't be bothered to convert the audio portion to Opus because bizarrely the source only contains a single lossy Dolby Digital (2-channels) audio stream. It appears The only video types which have no audio delay (based on metadata read by Mediainfo) are Bluray remux.

Not that I know of, but if you're doing this for other people then you should fix it properly.

-vcodec copy -acodec (whatever) will do the job faster. Or just extract the audio stream (-vn -acodec copy, of 'no video, copy audio codec'), add a delay, and then combine the two (-an -i video_source -i audio_source -vcodec copy -acodec copy outfile). Arguments should be in that specific order to tell it to not source the audio from the video source file.

I would like to turn off my computer for a while to see if letting it cool off would make it better, but I have H@H to consider, and also if I don't wait long enough before restarting, it would be all for nothing anyway, and I'm always doing stuff on my computer, not the kind of thing I can just randomly/arbitrarily rush (IMG:[invalid] style_emoticons/default/mellow.gif)

I would like to turn off my computer for a while to see if letting it cool off would make it better, but I have H@H to consider, and also if I don't wait long enough before restarting, it would be all for nothing anyway, and I'm always doing stuff on my computer, not the kind of thing I can just randomly/arbitrarily rush (IMG:[invalid] style_emoticons/default/mellow.gif)

Just replace a side of the computer with a box fan, lol
(temporarily; it'll get dusty)

Decided to permanently make my docking station a part of my laptop; popped a secondary 1TB HDD in there so I can hopefully distribute my disk I/O a little more.

Edit: It seems to be working pretty well. SW that was previously making my computer grind to a halt is feeling snappy (including Audacity, as well as Photoshop in Wine; I am directing the "scratch disk" location to my secondary drive). I missed having two HDD's in this thing. The 1TB drive is the one that I migrated my laptop from when I upgraded to a 2TB disk around a year and a half ago (probably). So I just erased it and as an upshot I also have a full terabyte more of space in my laptop (it's MBR-based, so I can't really get much bigger individual HDD's than the 2TB one that's already inside it).

I also moved my normal "swap" space to the other drive (added 'nofail' to fstab so it shouldn't make my laptop unusable if I boot it undocked).

This post has been edited by dragontamer8740: Mar 2 2021, 11:58

Rewound the choke for the power supply last night. Ended up doing that so I could make a tifllar winding, and see how many turns it had. My component tester claimed both windings were 680uH and I counted 144 turns of wire I assume was 25awg. Rewound it with 114 turns of 26awg loosely twisted into a bundle, and they all came out to 750uH somehow. Got that soldered back, and the heatsink with both regulators, and the capacitors, so the new rail should be working. Globbed that silicone all over everything, or at least stuck the inductor down with it, and filled in under the caps etc like it was before I messed with it.
Capacitor kit didn't include 100uF 50V, so I stuck two 47uF in parallel. Regulator inputs got 0.01uF 50V ceramic discs and outputs got 0.1uF disks added on as well.
Gotta add the power switch today, and terminate the leads in some usable manner.

This post has been edited by Wayward_Vagabond: Mar 2 2021, 13:51

Still not sure if it's really tech chat, but doing more pixel art today.

Lighting on skin is hard. This is a WIP.
(IMG:[i.imgur.com] https://i.imgur.com/UUcLVPG.png)
Update:
(IMG:[i.imgur.com] https://i.imgur.com/ZVPrHiL.png)
(IMG:[i.imgur.com] https://i.imgur.com/rtty9vS.png)
Update 2 (fixed hair, tried to fix eyes):
(IMG:[i.imgur.com] https://i.imgur.com/Qlx6hx6.png)
(IMG:[i.imgur.com] https://i.imgur.com/8jBaw7F.png)

This post has been edited by dragontamer8740: Mar 4 2021, 07:16

Power switch.


Ugly choke, and bodged double capacitor. Choke used to sit upright.


1K resistor to make the rail go down when the supply is off. The -12V rail had one. Picked a random ground pad, and jammed the other end in beside the -5V out wire.


Ceramic discs I added pre and post regulator to both negative rails, and my kludged on diodes.

The silicone is Chipquik neutral cure electronics silicone, btw. It's a slightly more posch[sic] version of the "704" white silicone OEMs use, because I couldn't readily source it domestically.

Thanks; I needed to know what that silicone stuff was called because I need some for my CRT monitor repair (I had to remove and re-seat its yoke).

My Galaxy J3 Luna Pro's screen cracked yesterday. Upgrading to a NOS unlocked Galaxy S7 32GB. Ordered an SD card, a glass screen protector, and an Otterbox Symmetry case for it. Trying to find just a glass, or a screen/digitizer/glass assembly for my current phone, but not having a great deal of luck.

I dropped it about 6' onto concrete, and a crack that looked like a gravel hit a windsheild appeared in the top corner. Was kind of poking at it, and a deep scratch that interested that divot cracked, and went all the way across the phone. I have electrical tape over the divot cause loose glass and it doesn't cover the screen proper.

Picked the S7 cause it's a flagship and still has some community support, has a home button, an sd card slot, and comes in a verizon CDMA version. Home button, official android 10, and CDMA in the same phone seems to be a pipe dream... (somebody please tell me who to give money to if I'm wrong)