A friend's 486 board had a clock generator which allowed switching the settings to 40, 50, and 66Mhz. The IC could actually go up to 80Mhz. The clock was a carry over from the 386 systems. The board layout was set up so the 80 option couldn't be enabled through the DIP settings, however he jumpered it to work. We were able to get a DX2/66 running at 80Mhz. When the Dx4/100 came out, he was overclocking it to 120. His board needed a voltage converter board to allow the use of the DX2 and DX4 chips. The CPU could no longer be passively cooled at that temperature, so he wired up a 120mm fan to force air over it. It helped, but eventually couldn't keep up with the heat.
My grandfather is an IBM fellow. Before going to school for engineering, he messed around with electronics kits and HAM radio leading up to that. One of his earliest jobs at IBM was building RAM modules with vacuum tubes. He also designed and built communications equipment for the military. A lot of those guys he worked with at IBM had similar backgrounds, and that's why the ground planes are so big/thick. He retired for real around the time of the PS/2 launch (they brought him back from retirement for a while, presumably for that). I had hoped to be able to write a book about that era of computing with his help, but he sadly has dementia now, and its progress has been rapid.
@@Epictronics1 As I recall him saying, the large ground planes are to reduce interference and crosstalk, which was required in mil-spec communications equipment. I'm not 100% on that, but he's the expert, not me. :)
Once you fixed the initial issue with RAM errors being detected, I'm pretty sure the further messages and "missing" RAM were the system working as designed. IBM's 386 (and 486) SLC processors are effectively enhanced versions of the 386 SX, so they are limited to a 16-bit data bus and 24-bit address space. The 24-bit address space is 16MiB/16384KiB. In order for peripherals' dedicated memory to be accessible by the system/CPU, it needs to be mapped/reserved within the memory space. When you had less than 16MiB installed, these reservations were made in the upper part of the address space, above the amount of installed RAM. Once you configured the full 16MiB, the system had to disable a portion of system RAM to make room in the address space for the peripherals. Firstly, the 16KiB CPU cache should be subtracted from the 16384KiB, then believe the onboard VGA controller needs a more substantial amount of address space, and I suspect the SCSI controller does as well. I don't know if the 8514/A requires more, since I don't believe its framebuffer memory is made accessible to the CPU.
I was just about to comment with the same! It's similar to how a 32-bit OS (without PAE) will only "see" around 3.5 GiB of RAM if 4GiB is installed. A portion of the top of the 4GiB 32-bit address space is reserved for peripherals like the video card and is otherwise inaccessible to the OS.
It's pretty awesome, (well, or terrible at the time) back in the day they realized they had to build the mainboards and even some of the component boards with multiple options for chips / clocks etc., knowing that manufacturing of certain things sometimes was way behind (especially in the early boom of the 80s) so they absolutely put in any possibility for different setups - through holes for things vs surface mount and whatever other options there were. The thought that those engineers put into these systems.. all written out on paper back then too! Crazy.. things have gotten SO easy on us today, it's great watching you work with this old amazing hardware. I miss the olden days sometimes.. lot of the times.. : )
I always use the regular sockets. Being horizontal, the chance that the crystal will fall out is nihil. It also puts less stress on the legs of the crystal when pushing it in.
Tricked out machine needs a math-co! 487! It exists! And the weird refresh is for 87Hz interlaced, for 43.5Hz direct. This used to give me a real headache in the day. Even so the 8514/a at $5k dollars was a tremendous savings over contemporary mid-80s graphics at 1024x768, which might go for $15k,.
8514/A mode is 1024x768@87i - Interlaced, a good many 15" monitors of the time could do 800x600 non-interlaced and 1024x768@87i - a mode supported by many superVGA cards as well
You probably can't use all 16 MB of RAM for a similar reason you can't use 4 GB in newer 32-bit machines - the CPU has a 24-bit address bus, so it can access 16 MB total, but it needs some address space for ROMs, VRAM and memory-mapped I/O
Possible, however it's also an SLC processor. Those are interesting because they have System Management Mode (SMM) and can partition off some of the memory for BIOS too. Given this is IBM and an MCA PS/2 I would be completely unsurprised if some of those BIOS level drivers actually ran in SMM. This is complete supposition however.
Not really a correct analogy. 32 bit CPUs from the Pentium Pro onward can access up to 64 GB of RAM in 4 GB chunks using PAE. The 4 GB limitation is entirely artificial in Microsoft Windows, because they instituted memory licensing. Only server versions of Windows 32 bit allowed more than 4 GB of RAM. Linux had no such limitation with PAE/PSE enabled.
The 8514/A and XGA running at 1024x768 is actually an interlaced mode. 43.5 frames per second but 87Hz fields per second. You'll likely have some trouble finding monitors that are happy wih that. Of course the matching 8514 monitor will do it. The 640x480 mode does at least offer 256 colours unlike the vga
I kept troubleshooting after I was done with the video. The 8514/A issue is still not solved, but the IBM CRT display I use in the video turned out to be able to handle the weird 43.5Hz :) It's on the bench right now, running glorious 1024x768@43.5 :)
The silkscreen suggested that can use an square can oscillator. I usually bend slightly the pins of the component when I use a machined socket, it happened to me once for a DIP-8 IC to fall off from that type of socket. By upgrading the video RAM for the onboard VGA would not achieve a higher resolution? If I remember correctly, the video memory directly afected the display resolution and color depth on old VGA cards (at least on those that supported up to 2M of VRAM). I guess that the onboard video RAM is 512k? I never had an MCA computer, altough I many times could get one. I just did not liked the proprietary stuff, the scarcity of adapters and the high prices so thanks, watching your videos I can appreciate how an MCA IBM is.
The crystal oscillator hack instead of soldering to leg of crystal solder to leg of the pin on socket so you can swap crystals without needing to desolder the crystal leg every time.
Thanks, yeah, that's how I ended up doing in the Commodore 486 Overclock video. If I turned the board upside down, the crystal would fall out of the socket lol
@@Epictronics1 The 386SX only has a 24 bit address space, thus is only able to directly address 16 MB. Some of that space is reserved for system ROMs. There is also the space from 640K to just above 1M that is reserved for memory mapped devices, like video ROM and high memory area.
The Windows 95 device manager might be able to tell you which of the onboard components is reserving your missing 1MB of address space. Click on Computer and select Properties; a new dialog should apppear, click on Memory there.
May I ask an on-it-s-face stupid question? Could you hypothetically get a 386, add a Cyrix '486 clip on upgrade, and put a Pentium Overdrive in that socket?
I'm wondering why you're using round hole sockets instead of the wiper style for crystals. Those round hole ones don't seem to work as well from my experience as well as what I've heard. Maybe somebody can explain why some people prefer one over the other, though I've had better luck with the square hole single wipe type.
Dual wipe sockets are much better than machined sockets for regular ICs in my experience. These crystals, however, have round pins and should fit better in machined sockets. That being said, crystals seem to fit well in dual wipe sockets too.
The display drivers claim to be version 3.1 which suggests they are actually drivers for Windows 3.1 and just happen to work with Windows 95. 16 bit Windows 3.x drivers will usually work with Windows 95 (even severely limited very old drivers like ega can be made to work) but they will be slow and be missing the more advanced features introduced in Windows 95 like refresh rates, the drivers do not tell windows what refresh rates are supported. The XGA drivers will might have worked too and had a better result. Did you run the add new hardware wizard after installing to card to let Windows attempt to detect the 8514? The strange 43.5 HZ refresh rate is because it is running 1024x768 in 87 Hz interlaced mode. Modern monitors do not support this mode but 90s CRTs capable of 1024x768 will. Modern Monitors might support 30/60i (for 1080i) but not 43.5/87i (for 768i). 8514 does not ordinarily let you change the refresh rate so you're stuck with it like this unless you use a scan converter.
I did some more tests after the video was done. The card works in a 50Z with Win 3.1 with the included drivers on an IBM 8515 and P97 display. But no progress with the 57 yet, unfortunately.
@@Epictronics1 ok so we have dx...now we only need dx2 and that magic bus doubling sorry 2 must have not registered when i first typed i ment something that run core ad faster speed than the memory bus, floating point math was not neseserly needed (mistical 486 sx2 from intel comes to mind)
![[#2024MAMA] ROSÉ (로제), Bruno Mars - APT. | Mnet 241122 방송](http://i.ytimg.com/vi/dgGqD28J6aQ/mqdefault.jpg)
Support me on patreon.com/Epictronics
Buy me a coffee: ko-fi.com/epictronics
Join me on Twitter: twitter.com/epictronics1
A friend's 486 board had a clock generator which allowed switching the settings to 40, 50, and 66Mhz. The IC could actually go up to 80Mhz. The clock was a carry over from the 386 systems. The board layout was set up so the 80 option couldn't be enabled through the DIP settings, however he jumpered it to work. We were able to get a DX2/66 running at 80Mhz. When the Dx4/100 came out, he was overclocking it to 120. His board needed a voltage converter board to allow the use of the DX2 and DX4 chips. The CPU could no longer be passively cooled at that temperature, so he wired up a 120mm fan to force air over it. It helped, but eventually couldn't keep up with the heat.
Sounds like fun :) I managed to run my DX4 (ODPR100) at 120 too in another video two weeks ago
My grandfather is an IBM fellow. Before going to school for engineering, he messed around with electronics kits and HAM radio leading up to that. One of his earliest jobs at IBM was building RAM modules with vacuum tubes. He also designed and built communications equipment for the military. A lot of those guys he worked with at IBM had similar backgrounds, and that's why the ground planes are so big/thick. He retired for real around the time of the PS/2 launch (they brought him back from retirement for a while, presumably for that). I had hoped to be able to write a book about that era of computing with his help, but he sadly has dementia now, and its progress has been rapid.
What an awesome career. I'm sorry to hear about his health. Do you mean they needed thick ground planes for HAM stuff?
@@Epictronics1 As I recall him saying, the large ground planes are to reduce interference and crosstalk, which was required in mil-spec communications equipment. I'm not 100% on that, but he's the expert, not me. :)
Once you fixed the initial issue with RAM errors being detected, I'm pretty sure the further messages and "missing" RAM were the system working as designed. IBM's 386 (and 486) SLC processors are effectively enhanced versions of the 386 SX, so they are limited to a 16-bit data bus and 24-bit address space. The 24-bit address space is 16MiB/16384KiB. In order for peripherals' dedicated memory to be accessible by the system/CPU, it needs to be mapped/reserved within the memory space. When you had less than 16MiB installed, these reservations were made in the upper part of the address space, above the amount of installed RAM. Once you configured the full 16MiB, the system had to disable a portion of system RAM to make room in the address space for the peripherals. Firstly, the 16KiB CPU cache should be subtracted from the 16384KiB, then believe the onboard VGA controller needs a more substantial amount of address space, and I suspect the SCSI controller does as well. I don't know if the 8514/A requires more, since I don't believe its framebuffer memory is made accessible to the CPU.
I was just about to comment with the same!
It's similar to how a 32-bit OS (without PAE) will only "see" around 3.5 GiB of RAM if 4GiB is installed. A portion of the top of the 4GiB 32-bit address space is reserved for peripherals like the video card and is otherwise inaccessible to the OS.
It's pretty awesome, (well, or terrible at the time) back in the day they realized they had to build the mainboards and even some of the component boards with multiple options for chips / clocks etc., knowing that manufacturing of certain things sometimes was way behind (especially in the early boom of the 80s) so they absolutely put in any possibility for different setups - through holes for things vs surface mount and whatever other options there were. The thought that those engineers put into these systems.. all written out on paper back then too! Crazy.. things have gotten SO easy on us today, it's great watching you work with this old amazing hardware. I miss the olden days sometimes.. lot of the times.. : )
I'm glad they did, otherwise, we would have been stuck at 46MHz :) Yeah, the transition from through-hole to SMD must have been a mess!
You're really pushing the envelope trying to upgrade this thing to run Windows 95! Great work
Thanks. It actually runs W95 ok. It's not perfect, but it's usable :)
It seems like crystal sockets have always been sketchy. A lot of old boards that used them had zip ties to improve contact. Nice video card! 🤤
Thanks. I think regular dual-wipe sockets are probably the best solution so far.
I always use the regular sockets. Being horizontal, the chance that the crystal will fall out is nihil. It also puts less stress on the legs of the crystal when pushing it in.
Tricked out machine needs a math-co! 487! It exists! And the weird refresh is for 87Hz interlaced, for 43.5Hz direct. This used to give me a real headache in the day. Even so the 8514/a at $5k dollars was a tremendous savings over contemporary mid-80s graphics at 1024x768, which might go for $15k,.
I have math co for it but I'm not aware of any game I like that takes any advantage of it :)
8514/A mode is 1024x768@87i - Interlaced, a good many 15" monitors of the time could do 800x600 non-interlaced and 1024x768@87i - a mode supported by many superVGA cards as well
It turned out to be a compatibility issue. The card and monitor are now running perfectly in a Model 50
This was awesome! I'm impressed at how much you were able to push this machine. :D
Thanks, yeah, that's one crazy fast 386!
A more elegant (and universal) solution for the crystal botch would be to solder the botch wire on the underside of the socket.
I vote for AWE32 as the sound card!
MCA slots in this IBM. SB 1 clone will have to do :)
That system is a looker for sure
I loved the hack with the crystal ocilators. It's always fun to see homebrew hacks to motherboards to push these systems.
You probably can't use all 16 MB of RAM for a similar reason you can't use 4 GB in newer 32-bit machines - the CPU has a 24-bit address bus, so it can access 16 MB total, but it needs some address space for ROMs, VRAM and memory-mapped I/O
Possible, however it's also an SLC processor. Those are interesting because they have System Management Mode (SMM) and can partition off some of the memory for BIOS too. Given this is IBM and an MCA PS/2 I would be completely unsurprised if some of those BIOS level drivers actually ran in SMM. This is complete supposition however.
Not really a correct analogy. 32 bit CPUs from the Pentium Pro onward can access up to 64 GB of RAM in 4 GB chunks using PAE. The 4 GB limitation is entirely artificial in Microsoft Windows, because they instituted memory licensing. Only server versions of Windows 32 bit allowed more than 4 GB of RAM. Linux had no such limitation with PAE/PSE enabled.
So. Many. Errors! This was a sea of errors, speed and bloopers!
Still the usual quality content one can expect from Epictronics Studios.
Thanks :)
You could always try OS/2 to see if it had support for that graphics card. OS/2 2 would be era appropriate.
It's almost Chrisтmas! Are you going to get another smoking special this year?! 😂
I probably will :D
Hi Epictronics, Nice overclock what you did. The system is faster then normal. Greetings from Steven from the Netherlands
Thanks :)
The 8514/A and XGA running at 1024x768 is actually an interlaced mode. 43.5 frames per second but 87Hz fields per second. You'll likely have some trouble finding monitors that are happy wih that. Of course the matching 8514 monitor will do it. The 640x480 mode does at least offer 256 colours unlike the vga
I kept troubleshooting after I was done with the video. The 8514/A issue is still not solved, but the IBM CRT display I use in the video turned out to be able to handle the weird 43.5Hz :) It's on the bench right now, running glorious 1024x768@43.5 :)
The silkscreen suggested that can use an square can oscillator. I usually bend slightly the pins of the component when I use a machined socket, it happened to me once for a DIP-8 IC to fall off from that type of socket.
By upgrading the video RAM for the onboard VGA would not achieve a higher resolution? If I remember correctly, the video memory directly afected the display resolution and color depth on old VGA cards (at least on those that supported up to 2M of VRAM). I guess that the onboard video RAM is 512k?
I never had an MCA computer, altough I many times could get one. I just did not liked the proprietary stuff, the scarcity of adapters and the high prices so thanks, watching your videos I can appreciate how an MCA IBM is.
Thanks, I need to check and upgrade the VRAM on this IBM. I have not looked into it yet, but it has two chips and two empty sockets
as always... please run "fdisk /mbr" to rewrite the master boot record. You don't know where that drive has been.
That's probably good advice. I've had my share of boot-sector viruses
The crystal oscillator hack instead of soldering to leg of crystal solder to leg of the pin on socket so you can swap crystals without needing to desolder the crystal leg every time.
Do you have any other channels? I kinda recognize the voice and stubborn intonation
No, but I probably should have one
Are you sure that ancient video card isn't just stuck at 1024x768 43Hz interlaced? That was super common back in the bad old days.
It turned out to be a compatibility issue. The card and display works fine in a Model 50 now
Try a small zip tie to secure the crystal to the socket.
Thanks, yeah, that's how I ended up doing in the Commodore 486 Overclock video. If I turned the board upside down, the crystal would fall out of the socket lol
Good Video a lot of fun 😊
Thanks :)
If the graphics card is integrated, it might use 2MB out of the onboard RAM
There is dedicated VRAM on the mobo, so it must something else
@@Epictronics1 The 386SX only has a 24 bit address space, thus is only able to directly address 16 MB. Some of that space is reserved for system ROMs. There is also the space from 640K to just above 1M that is reserved for memory mapped devices, like video ROM and high memory area.
I think I have one of those graphics cards laying around, or atleast a similar one.
The Windows 95 device manager might be able to tell you which of the onboard components is reserving your missing 1MB of address space. Click on Computer and select Properties; a new dialog should apppear, click on Memory there.
Unfortunately, I kept troubleshooting and replaced W95 with W3.1. It would have been interesting to see
You should be using double-wipe sockets
I only use double-wipe sockets
22:45 Sniff Sniff
I was like # 486! I feel lucky ^-^
May I ask an on-it-s-face stupid question? Could you hypothetically get a 386, add a Cyrix '486 clip on upgrade, and put a Pentium Overdrive in that socket?
Not a stupid question at all. Creating a 386 interposer for a POD would probably work on some 386 systems
XGA/2 is what you wanted.
Unfortunately, no 32-bit slots in this IBM
The missing ram might be for the onboard video
The graphics has it's own VRAM on this mobo
I'm wondering why you're using round hole sockets instead of the wiper style for crystals. Those round hole ones don't seem to work as well from my experience as well as what I've heard. Maybe somebody can explain why some people prefer one over the other, though I've had better luck with the square hole single wipe type.
Dual wipe sockets are much better than machined sockets for regular ICs in my experience. These crystals, however, have round pins and should fit better in machined sockets. That being said, crystals seem to fit well in dual wipe sockets too.
The display drivers claim to be version 3.1 which suggests they are actually drivers for Windows 3.1 and just happen to work with Windows 95. 16 bit Windows 3.x drivers will usually work with Windows 95 (even severely limited very old drivers like ega can be made to work) but they will be slow and be missing the more advanced features introduced in Windows 95 like refresh rates, the drivers do not tell windows what refresh rates are supported. The XGA drivers will might have worked too and had a better result. Did you run the add new hardware wizard after installing to card to let Windows attempt to detect the 8514?
The strange 43.5 HZ refresh rate is because it is running 1024x768 in 87 Hz interlaced mode. Modern monitors do not support this mode but 90s CRTs capable of 1024x768 will. Modern Monitors might support 30/60i (for 1080i) but not 43.5/87i (for 768i). 8514 does not ordinarily let you change the refresh rate so you're stuck with it like this unless you use a scan converter.
I did some more tests after the video was done. The card works in a 50Z with Win 3.1 with the included drivers on an IBM 8515 and P97 display. But no progress with the 57 yet, unfortunately.
You should get a signal generator.
I'd love to have one but fast enough signal generators for this type of OC are unfortunately very expensive
@@Epictronics1 that's fair enough. I'd like to get an oscilloscope, but they are just too expensive and I probably wouldn't use it very much.
did any dx cpu ever came out for that wierd socket in the middle of the adapter or is that impossible since it uses 386 memory bus?
Yes, I have one but I'm not aware of any game that makes any use of it, so I left it out
@@Epictronics1 ok so we have dx...now we only need dx2 and that magic bus doubling
sorry 2 must have not registered when i first typed i ment something that run core ad faster speed than the memory bus, floating point math was not neseserly needed (mistical 486 sx2 from intel comes to mind)
@@kokodin5895 We already have clock doubling. The next step is clock tripling :)
16gb of memory loses 1024 so hence it will be at 14.9gb. 16000 - 1024 = 14.9gb.