VOGONS


First post, by cde

User metadata
Rank Member
Rank
Member

The purpose of this computer is to be able to play DOS and Windows games from the late 80s to the early 2000s. It is not period accurate.

Motherboard: Abit KT7A

I chose the Abit KT7A for its one ISA slot. It was updated to the latest BIOS (no longer available on abit.ws but still mirrored in a number of places). The VIA 686B is supposed to have a bug with PCI latency but I haven't felt it so far, presumably because I don't use a PCI sound card. Also I suppose the latest BIOS has a workaround, as well as Windows XP SP3 (on the other hand, Windows 98 SE needs the VIA 4in1 4.43 driver to get the fix).

https://www.theregister.co.uk/2001/04/12/data … s_via_chipsets/

Previously I used a non-RAID version 1.3 KT7A but after flashing a modded BIOS found in this forum for supporting Thunderbred/Barton CPU I started having an insane amount of segfaults under Linux, to the point where sometimes not even init could start. I don't know if it's the modded BIOS or the CPU multiplier experiments I did, but going back to the stock BIOS and normal voltages did not solve the situation, so presumably something got physically damaged making this motherboard unusable 😠 (note that it was flaky at the beginning with one SDRAM slot not working). I will attempt to recap this motherboard using the KT7 caps kit from badcaps.net

EDIT: after recapping, the KT7A-RAID is back to working with up to 512 MB of memory without issues. More than 512 MB memory leads to errors in memtest86. This is a problem specific to this motherboard as another one will work without issues with 1.5 GB under XP or Linux. I also recapped a third, badly damaged KT7A but did not regain full usability - see /viewtopic.php?p=818073#p818073

Memory: 2*256 MB SDRAM PC133

This is the maximum amount with which the system is completely stable regardless of the OS. 1 GB does work with 98 SE (by editing MaxPhysPage and setting it to 1FFFF) however in this situation the NVIDIA driver refuses to offer 32-bit color, only 16-bit at 1024x768 - going back to 512 MB fixes it. My assumption is that the PCI IO space conflicts with memory above 512MB but I haven't been able to prove it conclusively.

CPU: Athlon XP-M "Barton" 2500+ (AXMH2500FQQ4C)

Has a fairly low TDP (45-53W) and is specified for 1867 MHz however the KT7A BIOS only goes to 12.5x multiplier (1666 MHz). This particular CPU can be undervolted to an insanely low 1.15V @ 1666 MHz (1.2V in BIOS, there is a 0.5V offset for some reason) and consumes very little energy, which allows reducing the fan speed without elevating temperatures. Frequency can be brought down to 500 MHz in the BIOS. Together with disabling L1 cache the speed can be brought down to below 486 DX/33 levels which is great for certain speed-sensitive games.

Unfortunately with this desktop motherboard PowerNow! is not being able to accept a live multiplier change, for example in Linux or with the DOS utilities SETMUL/POWERNOW. That's only a small annoyance because the multiplier can still be changed in the BIOS (the list of multipliers the BIOS offers will depend on the CPU, I suppose the XP-M family has the largest range). By the way, thanks to Falcosoft for this link about multiplier changes:
https://web.archive.org/web/20170822100906/ht … html/Tweaks.htm

I'm not sure overclocking this CPU would make a lot of sense, considering the very slow SDRAM is the primary bottleneck.

GPU: GeForce 6800 GT

There is not a lot to say about this GPU. It is quite powerful while working very well with a range of DOS/Win98/WinXP games. The proprietary Linux driver also has great compatibility even with old Linux distributions, and the newer open-source nouveau_vieux driver also works really well in recent distros. However I've noted a regression in performance between Debian 7 and 8 (nouveau_vieux from Debian 7 being much faster). The GPU is actively cooled by a Zalman VF-700, with a fan speed reducer to bring noise down.

Be careful when buying this GPU: some versions have an AGP<->PCIe bridge and do not work with the KT7A. I believe these versions can be identified by their 6-pin power connector (instead of the molex connector).

A very interesting aspect of the NVIDIA drivers, as compared to ATi, is the ability to preserve aspect ratio. In this way, StarCraft at 640x480, and other games, is rendered with perfect 2x scaling to 1280x1024 by adding black bars above and below. This is not necessary when using a CRT (since the native 640x480 resolution is used), but very useful when using a 1280x1024 LCD panel.

GPU: Radeon 9600 Pro

Although somewhat less powerful than the 6800 GT, I have been able to patch the BIOS ROM to force 720x400@70 Hz on the DVI output, so that games that rely on the VGA refresh synchronization signal for timing run at the correct speed (such games are Lotus III and Jazz Jackrabbit). While this change is certainly not needed when using a CRT, it is nice when using a LCD with DVI to avoid pixel "doubling" artifacts that will appears on VGA (see /viewtopic.php?p=775155#p775155) but retaining the 70 Hz refresh rate.

More details on the BIOS ROM patch: /viewtopic.php?f=46&t=68506 and/viewtopic.php?p=777414#p777414

Power Supply Unit: Seasonic S12II-620 82+ 80PLUS Bronze

This PSU seems extremely overkill considering the whole system uses at most 85W at the wall (the CRT consumes almost more power). Nonetheless it is the only one with a reasonable price and beefy 5V output (here 24A, so 120W in total), the issue being that motherboards of this era do not use the 12V rail and take all their power from the 5V rail. Fortunately since the CPU is extremely undervolted the requirements for 5V power are lessened.

Case: Cooler Master Force 500

Although I used previously a Cooler Master N400, I switched to a CM Force 500 as the PSU is at the top instead of the bottom. In fact it's one of the last cases that you can find with such an arrangement. Unfortunately the KT7A is slightly too short, I suppose the ATX form factor was expanded a bit at the beginning of the 2000's. Because of this there is no mechanical support at the right edge so care is necessary when plugging eg. IDE/floppy cables.

There are four fans (Corsair AF120 and AF140): two (at the back and on the side) suck air in, and two at the front blow air out in addition to the PSU fan. The northbridge has no fan and so the heatsink is cooled by the airflow from the back fan. This reduced noise considerably, and even under a hugh memory load (like memtest86) the northbridge does not get hot.

The case front panel headphone and microphones jacks were connected to the audio cards line out and microphone. On the Edison Gold 16 Plus, this required soldering on the back of the ISA card 2.54mm jumper cables that can be plugged in the HD audio header, in a manner similar to:
/viewtopic.php?f=62&t=67254&p=760442#p760375
(note that I no longer use the Audigy 2 ZS, so the microphone input is indeed connected to the ES1868F). On the TerraTec 16/96 Gold, headers are already provided on the PCB (which is excellent) and so no soldering is required, only female-male cables are used to add the connection.

The motherboard has a USB 1.1 connector that is pin incompatible with USB 2.0 headers. Still, I was able to use this connector. The cleanest solution would have been to use 2.54mm male-female cables to match Vcc, GND, D+, D- from the motherboard to the header. However it is also possible to rotate the header 180 degrees, slide it by one and now Vcc is connected to Vcc, and GND to GND. D+ and D- are now inverted, but since this is a differential pair this works fine and I was able to connect USB pen drives.

An alternative is to connect the front panel USB 2.0 header to the type A USB port on the NEC PCI card using an adapter such as https://www.ebay.com/itm/332407752005 but this does not provide good results as the D+ and D- of two USB ports are connected.

Display: iiyama 17" Vision Master 1403

It was rather cheap to buy at 20E, has a shadow mask and gives a very authentic look and feel. The only drawback I have is that at 640x480 60Hz (eg. in Windows 3.11) there is visible flickering of the screen which is visually tiring. Also I noted Windows XP in 800x600 will default to 60 Hz despite the monitor supporting much higher refresh rates. The refresh rate can be set up to 100 Hz (800x600) or 85 Hz (1024x768) fortunately. Certain games require the use of the "RefreshLock" (attached) to force the highest possible refresh rate instead of the default 60 Hz (in addition, in Half-Life fps_max must be increased). Overall this CRT was an excellent and cheap purchase. I strongly recommend getting a CRT screen, as it provides a much more faithful experience than any LCD.

Display: AOC G2590PX (1920x1080) @ 144 Hz

An excellent LCD panel that can work with different refresh rates on its VGA and HDMI inputs without dropping frames. Also features 1:1, 4:3 (17" or 19") and 5:4 modes which preserve aspect ratio. For more details, see: /viewtopic.php?p=790180#p790180

Display: Acer V176L bmd (1280x1024)

Very cheap and yet decent screen that is lightweight and works great. It features VGA and DVI and is compatible with the 9600 Pro BIOS ROM patch described above. For more details, see: /viewtopic.php?p=781249#p781249

Sound card: ESS AudioDrive ES1868F

I'm extremely satisfied with the two cards I have: Edison Gold 16 Plus and TerraTec 16/96 Gold. They both have an extremely low noise floor, do not pick up RF noise from the rest of the system, and have (as far as I can tell) very good SoundBlaster Pro capability. Installation is extremely easy. I have added a DreamBlaster X2 + Buran 0.9 which sounds really awesome and Serge is a super nice guy. Compatibility with DOS/98/XP/Linux is excellent. The only caveat this I discovered is that (on the Edison) by default the output is set to speakers which was way too loud for headphones. I switched the jumpers to Line out without 3D and the result is excellent. I wish in fact I had bought this card first instead of buying the ALS100/ALS100+ (horrible, crappy sound) and YMF719/ESS Solo-1/Audigy 2 ZS (very decent but somewhat difficult to set up and got compatibility issues). Note that not all ES1868F have a low noise floor; see this post for a picture of the card: /viewtopic.php?p=770322#p770322

Sound card: OPL3LPT

The ES1868F is a good OPL3 clone, but the OPL3LPT is an excellent choice to get authentic OPL3 sound. Just be aware that the power requires clean 5V, it is best IMO to reuse the 5V from the PSU and avoid phone USB chargers which can sometimes not provide a clean voltage. Also be careful of ESD when manipulation the OPL3LPT as the back is not protected.

Sound card: BLASTERBOARD 2.0

Expensive but excellent sound blaster clone. I still use the EF1868F on a daily basis due to its wavetable header, stereo ability and excellent Windows/Linux compatibility, but the BLASTERBOARD remains an very good alternative.

Storage: Seagate Barracuda 7200.7 (ST360014A)

Very silent and fast, a great alternative to the SSDs presented below.

Storage: Samsung 860 EVO (250 GB) mSATA SSD

My current storage is the Samsung mSATA 860 EVO (250 GB) together with a mSATA to IDE 2.5 board and passive 2.5 to 3.5 IDE adapter: /viewtopic.php?p=808477#p808477 . It has the advantage of being fully compatible with DOS/Windows 9x/XP without additional drivers. The only issue I have with it is that I have to force it to PIO in the BIOS: if UltraDMA is enabled, Windows 98 refuses to use UDMA and is extremely sluggish, and XP outright bluescreens. OTOH PIO mode 4 works fine with up to 33MB/s which is largely enough.

Storage: SI3114 with Samsung 850 EVO (250 GB) SATA SSD

Another option I've tried is the SI3114 which was reflashed with the 5.5.00 BIOS and works fine with a SATA SSD. It has the disadvantage of consuming 10KB of conventional memory for its buffers, leaving 630KB for DOS, in addition to having to share the PCI bandwith with eg. network and USB 2 cards. XP installation did require slipstreaming of the SI3114 drivers but that was no issue. I'm using driver version 1.3.10.0 which is compatible both with 98 and XP (3114-vista-x86-logo-1.3.10.0.zip).

Storage: NEC IDE DVD ND-4570A / ND-4571A

I bought a good number of IDE drives that arrived either completely dead or failed to read different kinds of media, or were very noisy. I finally found the ND-4570A and ND-4571A which are very silent and have no problems with different kind of media (CD-R(W), DVD+/-R(W)). It has RGGL audio output but unfortunately lacks a volume knob.

Storage: Gotek with HxC2001 and FlashFloppy

After having a floppy drive and a couple disks fail on me - very retro but also annoying 😉 - I switched to a Gotek "SFR1 M44-U100 K" with HxC2001 firmware and I'm very satisfied with it. It's extremely easy to use: just put the indexed mode HXCSDFE.CFG file at the root and name the disks DSKA****.IMG. The array of options this firmware offers is quite huge and very impressive. In fact Jeff added my suggestion of customizing the idle 7 segments display splash screen (https://hxc2001.com/custom_fw/) so that instead of displaying HxC all the time the screen can remain blank. I have a second Gotek on which I installed FlashFloppy, and it works great too. Be aware that newer, non-SFR Gotek drives might not be usable due to a different pinout of the serial port and BOOT0! see https://github.com/keirf/FlashFloppy/wiki/Got … k-Compatibility

Miscellaneous PCI card: 3Com 3C509B / NEC USB 2.0

The 3C509B has the advantage over the Intel Pro/100S that its lack of a ROM BIOS makes booting slightly faster. It also has great compatibility. The only drawback I found so far is that the DOS packet driver (3c90xpd.com) requires it to have a interrupt by itself, if the interrupt is shared eg. with the SI3114 then disk access fails. So I have placed this card above the ES1868F and the BIOS will assign it IRQ 9 which does not conflict with other PCI cards.

Connecting to a recent Linux machine through ssh requires the -g option, and on the server end this configuration options:

HostKey /etc/ssh/ssh_host_dsa_key
HostKeyAlgorithms ssh-dss, ...
KexAlgorithms diffie-hellman-group1-sha1, ...
Ciphers aes128-cbc, ...
MACs hmac-sha1, ...

Run ssh-keygen -A to generate the DSA key. Otherwise ssh2dos will fail to connect with a handshake failure error message.

The NEC USB 2.0 card works without issues, although unfortunately like the USB 1.1 found on the motherboard it can only provide a small amount of current. Therefore it is not capable of powering eg. the OPL3LPT. Presumably when USB was still new it was only designed to be used with low power devices such as keyboard and mices, and only later was the 500mA current delivery option added. I'd be wary of plugging a device that draws a lot of current since I have no idea if the card or the motherboard has overcurrent protection. Nonetheless it's fine for connecting USB pen drives.

Operating systems: MS-DOS 6.22 / Windows 98 SE / Windows XP SP3 / Debian 7.11

The Samsung SSD has four primary partitions.

The first one is for MS-DOS 6.22. After installation, the partition type has been changed from 06 to 60 and IO.SYS has been altered to recognized this partition type instead of 06 with the following patch:

at offset 4E610 from the beginning of the partition
74 06 26 80 3F *06* 75 06 0A F6 74 3A FE CE 83 C3 becomes
74 06 26 80 3F *60* 75 06 0A F6 74 3A FE CE 83 C3

This change may seem strange but it will prevent Windows 98 / XP from creating files on the partition with Long File Names, which I dislike and may not be compatible with certain disk tools (the partition can still be mounted in Linux with msdos filesystem type).
Also you must edit the disk directly to make this modification: editing IO.SYS at the filesystem level may relocate the file rendering the system unbootable.

By the way, EMM386.EXE from DOS 6.0 has a bug that causes the system to hang. The version shipped with 6.22 is OK which is why I switched to this version. The second and third partition host Windows 98 SE and XP SP3. Finally the last partition hosts Debian. Initially I used 7.11 which is the most recent version of Debian that has an Xorg version compatible with the NVIDIA driver for legacy cards (NVIDIA-Linux-x86-96.43.23-pkg1.run), here the GeForce4 Ti 4200. This is also the last Debian version to use a real SysV init by default which has a nice retro feel to it. After changing the video card for the 6800 GT, I installed Debian 8 which has faster booting times and includes the NVIDIA driver in its non-free repository.

At first I thought browsing the web with this machine would be impossible, but in fact it is possible with the trick of using an old version of Firefox that does not require SSE2 (48 or below I believe) plus the noscript and ublock origin add-ons. addons.mozilla.org does not keep versions of addons that are more that a couple years since they made changes to the addon API. Fortunately old add-ons versions can be found on legacycollector.org (I presently use noscript-2.9.0.14.xpi and ublock-origin-1.9.16.xpi). As a matter of fact, I'm posting this on the very PC I'm describing. I find Firefox 38.8.0esr to be veyr usable with the aforementioned plugins, and faster than the 4x series.

In conclusion, this was a triumph. I'm making a note here, huge success! It's hard to overstate my satisfaction 😀

Attachments

Last edited by cde on 2020-02-09, 17:16. Edited 32 times in total.

Reply 1 of 24, by melbar

User metadata
Rank Member
Rank
Member

Interesting approach to see the KT7A with a mobile CPU, an Athlon XP-M.

Have got also an Abit socket 462 board, but it's only the KT7 with out the "A", that means without additional onboard raid IDE ports.
Because it's KT133 only 100Mhz FSB, it does not make sense to combine it with a Barton. Personally i would go for periond correct thunderbird cpu.
Up to know, i have to replace some CAP's. The CAP's are of quite bad quality, we all know that from ABIT 😵 .
I assume you have the same issue right 😊

What is with your northbridge fan? The original one. It is noisy or have you replaced it?

Reply 2 of 24, by cde

User metadata
Rank Member
Rank
Member

So I got the idea of using an Athlon XP mobile from another thread on vogons, this allows lower power consumption and a good range of multipliers.

I haven't replaced the caps yet, as the board seems to be working fine. I might do it eventually altough it is a bit of work. Did you replace the caps on your KT7?

About the fan, it was the original noisy one which I removed entirely. Both CPU and NB get their cooling from the 120mm case fan sitting next to them. It works surprisingly well and is very silent.

EDIT: I reinstalled the CPU cooling fan (albeith with a speed limiter) as it allows further undervolting and resolves segfaults under Linux.

Last edited by cde on 2020-02-09, 10:59. Edited 1 time in total.

Reply 3 of 24, by melbar

User metadata
Rank Member
Rank
Member

I have already removed some of the critical caps on my board.

Because i've removed also caps of two other boards, this KT7 is still not finished.
Soldering is something, i do not for fun 😐 -> may i have less patience than others...

Reply 5 of 24, by FFXIhealer

User metadata
Rank Oldbie
Rank
Oldbie

I have an Abit KX7 motherboard paired with an AMD Athlon XP 3000+ Barton core. It originally had a 1800+ Palomino core when I originally built the computer.

It currently has 2x 1GB DDR-400 memory sticks in there running Windows XP SP3 perfectly. It originally was built with 256MB DDR-333, then doubled that to 512MB, but after SP2, this simply wasn't enough. But when I first tried to put 2GB of RAM in there, MemTestx86 threw lots of errors. Also, the hard disk drive would constantly power down and stop working mid-task in Windows XP... sometimes even preventing the OS from booting.

I sent the motherboard off to a guy in Minnesota or something like that and he completely recapped the board for me. Yeah, it was expensive, but it worked. Both sticks of 1GB RAM, absolutely solid - no errors at all. That same old hard drive that kept "failing"? Yeah, that stopped acting up too. It was clearly a fault of power delivery on the motherboard due to bad caps. Now I have GOOD caps on there and probably don't have to worry about it ever again....or at least it will work well for another 20 years.

292dps.png
3smzsb.png
0fvil8.png
lhbar1.png

Reply 6 of 24, by bofh.fromhell

User metadata
Rank Member
Rank
Member
FFXIhealer wrote:

I have an Abit KX7 motherboard paired with an AMD Athlon XP 3000+ Barton core. It originally had a 1800+ Palomino core when I originally built the computer.

It currently has 2x 1GB DDR-400 memory sticks in there running Windows XP SP3 perfectly. It originally was built with 256MB DDR-333, then doubled that to 512MB, but after SP2, this simply wasn't enough. But when I first tried to put 2GB of RAM in there, MemTestx86 threw lots of errors. Also, the hard disk drive would constantly power down and stop working mid-task in Windows XP... sometimes even preventing the OS from booting.

I sent the motherboard off to a guy in Minnesota or something like that and he completely recapped the board for me. Yeah, it was expensive, but it worked. Both sticks of 1GB RAM, absolutely solid - no errors at all. That same old hard drive that kept "failing"? Yeah, that stopped acting up too. It was clearly a fault of power delivery on the motherboard due to bad caps. Now I have GOOD caps on there and probably don't have to worry about it ever again....or at least it will work well for another 20 years.

My restoration projects shows similar results.
Hardware that was dodgy even as new suddenly works perfectly when introduced to really good caps.
AMD CPU with VIA chipset and an Matrox graphics card was generally consirered a bad combo.
But now ? not a single issue whatever 4-in-1 or graphics driver I throw at them.

Reply 7 of 24, by cde

User metadata
Rank Member
Rank
Member
bofh.fromhell wrote:
FFXIhealer wrote:

Hardware that was dodgy even as new suddenly works perfectly when introduced to really good caps.
AMD CPU with VIA chipset and an Matrox graphics card was generally consirered a bad combo.
But now ? not a single issue whatever 4-in-1 or graphics driver I throw at them.

That makes complete sense. The faulty board I have is probably not capable of supplying enough power, and as a result the CPU segfaults a lot. In any case I didn't throw the board away and may considering trying recapping it.

Reply 8 of 24, by FFXIhealer

User metadata
Rank Oldbie
Rank
Oldbie

Just a point of correction so that you don't make any incorrect assumptions in the future.

It's not that it's not supplying enough power.... it's that it's supplying INCONSISTENT power. What that means is, when a cap dies, electricity flows through it, but doesn't get stored.... so it doesn't get filtered. That means every single dip in power..... also passes through. Every single spike in power - also passes through.

When a capacitor is working properly, it holds electrical charge and and releases it at a specific rate, so capacitors are usually used for one of two reasons - to filter out power fluctuations (like in electronic components such as your computer and cell phone)... or to provide burst power (like starting a fan or the big motors running your air conditioner outside. Those caps are the size of old 6-volt batteries).

Inconsistent power can result in your RAM losing bits and getting thrown off, causing data errors.
Inconsistent power can result in circuit faults in the IDE controller, causing the HDD to shut down due to lack of command and control signals from the controller chip (the HDD thinks it's unplugged).

If it were me, choosing to trash a board or spend money to fix it depends on the perceived value of the product to you. Since this is entirely subjective, no one here can give you a recommendation. But I can tell you that I was ok with paying to recap my motherboard because it was the 2nd computer I ever bought and I gamed pretty hard on it back in the day, so there are memories there.

292dps.png
3smzsb.png
0fvil8.png
lhbar1.png

Reply 9 of 24, by KT7AGuy

User metadata
Rank Oldbie
Rank
Oldbie

I don't have a v1.1 revision KT7A, but I do have v1.0, v1.2, and v1.3 models. Some things I've noticed over the years:

My v1.0 board doesn't like to run with more than 512MB RAM. Trying to run with more than that causes all sorts of problems and nothing I've tried has helped. It's perfectly fine with 512MB RAM though.

My v1.2 and v1.3 boards work fine with 768MB RAM (3 x 256), but I noticed that they get slower results in 3DMark2001SE than when tests are run with only 512MB (2 x 256). I'm not sure why that is, but I know that Win98SE is generally happier with 512MB (or less) anyway, so I just use two sticks and everything works great.

Just like in the link that cde- posted, I've avoided the 686B bug by using only ISA sound cards and a single hard drive with the board's onboard controller. I'm not quite sure why anybody would run a PCI sound card in a KT7A. The ISA slot is what makes the KT7A so cool.

Another fun little quirk with KT7A boards: Voodoo 2 v3.02.02 drivers will cause lockups and other problems. Don't use them. Use either the older v3.03.00 drivers (yes, the version numbers are weird) or an unofficial package like FastVoodoo2 that has the fix built in already. I recommend the FastVoodoo2 drivers for a trouble-free experience with the KT7A.

I've had all of my KT7A boards recapped by Chris from badcaps.net. I know I would ruin the boards if I tried to do it myself, so paying Chris was worth it to me. He does really good work. If you need a board recapped and are unsure of your own skills, I can personally recommend his service.

Reply 10 of 24, by cde

User metadata
Rank Member
Rank
Member

Thanks for the precisions, KT7AGuy. I have come to the same conclusion that 512 MB is the only amount where no corruption is guaranteed to happen. On the 1.3 board I also had issues with 768 MB or more, possibly because of bad caps.

By the way, I have tried the USB type A to 9-pin USB header cable bought from ebay, so as to connect the front panel USB to the NEC PCI card (I previously used a hack to connect the front panel header to the motherboard, by rotating it).

It works fine but the NEC card will be limited by the power it can draw from the motherboard. So I tried disconnecting the 5V from the type A and instead getting 5V from the PSU directly. Unfortunately any connected device will fail high-speed negotiation and falls back to full-speed. I assume this is because part of the negotiation involves the power supply itself.

Reply 11 of 24, by KT7AGuy

User metadata
Rank Oldbie
Rank
Oldbie

If I remember correctly, I had to use the schematics in the manual to wire up the USB connections for the front panels correctly. It's not hard to get the wires out of the USB connector blocks so you can rearrange them. You just use a needle or something pointy to lift up the little plastic tabs, then the wires come right out and you can rearrange them so they'll work with the KT7A correctly. You can see an example of this operation here:

https://youtu.be/W_4VCm-0CwM?t=177

It sucks that the KT7A only has USB 1.1 onboard. In situations where I've wanted front USB 2.0 connections with older boards like this, I use a USB 2.0 PCI card and then connect it to one of these:

https://www.newegg.com/syba-sd-u2hub-4-usb/p/N82E16817801001

There are other versions of 3.5" USB panels like that on sites like NewEgg, Amazon, eBay, etc. I've just always used that particular model because it has a power connection and a female USB-B connection on the back. This makes it easy to use a cable with male USB-A and USB-B ends to connect it to a PCI USB 2.0 card.

If you can't find a USB 2.0 front panel solution anymore, I know I've seen USB 3.0 to USB 2.0 header adapters. I've just never tried them so I don't know how well that would work.

Reply 12 of 24, by cde

User metadata
Rank Member
Rank
Member

Thanks. For the record, here's the adapter I bought: https://www.ebay.com/itm/-/332407752005

If at some point this item becomes unavailable, search for something similar to "1Pcs Single USB Male To USB 9pin Motherboard Adapter Cable Cord Wire 20CM".

In the future I'd like to to a modification of the NEC card to pull 5V for USB from the PSU directly instead of the PCI bus, this would allow more power for USB than what the PCI bus maximum specs allow (25W per card). Could there be any adverse effects of such a modification?

pvkF9i0.jpg

Reply 13 of 24, by KT7AGuy

User metadata
Rank Oldbie
Rank
Oldbie

I like the idea of using a cable like that, but I can see right away that it would cause exactly the sort of problems that you're describing. That cable is essentially taking one USB port and splitting it in two without a hub. I'm not an expert on such things, but I'm pretty sure it can't work this way.

The internal female USB-A connector on a PCI USB 2.0 card is a single port, so you can only connect a single device or a hub to it. That Syba front panel I mentioned earlier is actually a hub, so this type of situation works with it. Attempting to connect two front panel USB ports to a single internal port won't work. This is more than a problem of supplying enough power.

Reply 14 of 24, by cde

User metadata
Rank Member
Rank
Member

Yes, it does work as long as you only plug one device into it. I didn't try plugging two devices but I have no doubt the result would be bad.

Now I'm thinking of a new solution, which is to take this as well as the header on the MB and have one USB 1.1 port and one USB 2.0 port on the front panel.

Reply 15 of 24, by looking4awayout

User metadata
Rank Member
Rank
Member
cde- wrote:

At first I thought browsing the web with this machine would be impossible, but in fact it is possible with the trick of using an old version of Firefox that does not require SSE2 (48 or below I believe) plus the noscript and ublock origin add-ons. addons.mozilla.org does not keep versions of addons that are more that a couple years since they made changes to the addon API. Fortunately old add-ons versions can be found on legacycollector.org (I presently use noscript-2.9.0.14.xpi and ublock-origin-1.9.16.xpi). As a matter of fact, I'm posting this on the very PC I'm describing.

In conclusion, this was a triumph. I'm making a note here, huge success! It's hard to overstate my satisfaction 😀

Why using an outdated version of Firefox (48) when you can use Roytam1's Firefox 45 ESR SSE which has better support, it has TLS 1.3 and webp implementations and has several more updates backported from newer versions of Firefox? I use an overclocked P3 Tualatin as my main machine and I can tell you that 45 ESR SSE coupled with uBlock Origin 1.10.0, NoScript 2.9.0.1 RC1, Bluhell Firewall, Classic Theme Restorer 1.2.9 and the UOC Patch, make it very fast and usable. With the YT2Player plugin, you can even watch Youtube videos in VLC.

If you haven't heard about it, the UOC Patch is a patch I made myself to optimize Firefox and other Mozilla based browsers in order to run as good as possible on old machines. You can find more info about the patch here:

https://msfn.org/board/topic/178306-experimen … r-old-machines/

While you can find more info about Firefox 45 ESR SSE here:

https://msfn.org/board/topic/177125-my-build- … -targetting-xp/

If I can use a Tualatin as a daily driver, you can too... Although if you want your system to perform as fast as possible with the UOC Patch, you'll need a DirectX 9 capable graphics card, such as the Geforce 6800GT.

My Retro Daily Driver: Pentium !!!-S 1.7GHz | 2GB PC164 ECC SDRAM | Gecube Radeon X1950 Pro 256MB OC | 128GB Lite-On SSD + 500GB WD Blue SSD | Creative Sound Blaster Live! CT4620 | Windows XP Professional SP3

Reply 16 of 24, by cde

User metadata
Rank Member
Rank
Member

That is an extremely informative post looking4awayout, and I commend you for writing it!

To answer your question, I did not use Roytam1's Firefox 45 ESR SSE because I did not know it existed.

I will use it on Windows XP. However my main OS to browse Internet is Linux 7.11. I might recompile Roytam1's patched source for Linux if I have the time.

By the way the type A to USB 2.0 header is not working very well. presumably duplicating D+/D- to two ports is bad for the signal integrity and I cannot connect a hispeed device. I'm back to plugging the header on the motherboard with orientation inversed.

Reply 17 of 24, by looking4awayout

User metadata
Rank Member
Rank
Member
cde- wrote:

That is an extremely informative post looking4awayout, and I commend you for writing it!

To answer your question, I did not use Roytam1's Firefox 45 ESR SSE because I did not know it existed.

I will use it on Windows XP. However my main OS to browse Internet is Linux 7.11. I might recompile Roytam1's patched source for Linux if I have the time.

Would be interesting if the UOC Patch works on Linux too. If you will need assistance I'll be here to help you at finding the most lightweight versions of the plugins you need to get a better browsing experience alongside the UOC Patch.

My Retro Daily Driver: Pentium !!!-S 1.7GHz | 2GB PC164 ECC SDRAM | Gecube Radeon X1950 Pro 256MB OC | 128GB Lite-On SSD + 500GB WD Blue SSD | Creative Sound Blaster Live! CT4620 | Windows XP Professional SP3

Reply 18 of 24, by cyclone3d

User metadata
Rank l33t
Rank
l33t

As for the multiplier only going to 12.5 on the KT7A, you can do a simple wire mod to the socket to get higher multipliers.

No soldering required, you just use a single strand of wire in either 1 or two places and you can get up to about 2.3Ghz with the KT7A.

I have a Barton 2800+ Mobile running in mine.

I had a KT7A back in the day when they were on the market and then in one of my overclocking / homemade waterblock experiments I tightened the cooler bracket down too much and killed the board.

The one I have now I bought off of eBay. It was NOS. First time I powered it up, 3 of the capacitors exploded. Had the insides of those capacitors blown all over the place and it was quite loud.

See here for the multiplier mod info:
Interactive guide: https://www.ocinside.de/workshop_en/amd_pinmod/

Enable high multiplier bit (what you need for the KT7A) - you will have to convert this to the socket side if you want to do the wire strand method: http://www.emboss.co.nz/amdmults/

Yamaha YMF modified setupds and drivers
Yamaha XG resource repository - updated November 27, 2018
Yamaha YMF7x4 Guide
AW744L II - YMF744 - AOpen Cobra Sound Card - Install SB-Link Header
Epstein didn't kill himself

Reply 19 of 24, by cde

User metadata
Rank Member
Rank
Member

That is some awesome advice cyclone3d, and I thank you very much for it.

For the time being I will stick to 1666 MHz, the reason being that I can undervolt this mobile CPU to 1.3V, it allows me to cool it using only the rear case fan and removes the need for a CPU fan. This way both CPU and NB are cooled by the rear fan, this makes the whole PC a lot more silent. The CPU and NB get a bit hot, but not that much - CPU goes to about 46C under a sustained 100% load and NB I estimate is roughly the same (at 25C room temperature). This particular case, the N400, has ventilation holes on the top which is good; however an older style case with the PSU at the top instead of the bottom would allow even better hot air extraction. Right now only the two 120mm fans at the front extract the hot air.

EDIT: after readding the CPU fan, it can be further undervolted (down to 1.15V) and has excellent stability.

Switching to a Zalman VF-700 on the GPU (GeForce 4 Ti) also allows it to be cooled passively by the 140mm side case fan that blows air inside. I have also put a VF-700 on my 6800 GT but here is it needs active cooling, albeit with a speed reducer. I was lucky to find a brand new VF-700, so the fan is in pristine condition.

Last edited by cde on 2020-02-09, 11:02. Edited 1 time in total.