Just one more! 😀 KT133A seems to have been popular with the mobo manufacturers back in the day! This one's by Soltek. It's called SL-KAV75. It's got an AGP Pro port. Currently, I'm not using the "special Pro section" of this AGP port, because it's got a warning sticker covering the "Pro section". (I'm using a Radeon 9800 Pro card ATM.)

It's interesting to note that this board is, I think, the last PCB revision they made. It's F5, and will accept (at least) a t-bred 2000+ rated CPU at 133FSB. It's possible that it will accept higher speed CPUs, but I haven't tested anything faster yet. (Actually, I did try a 2400+ rated t-bred in another identical Soltek board a few days ago, but I got no POST.)

The final BIOS revision is called Q12, and has a December 2003 datestamp. OC'ing is provided inside the BIOS by "Redstorm". To get this to work, you simply have to say "Yes" to OC'ing inside the BIOS set up area, and it tests the board itself, in an attempt to find a safe level of overclocking. I tried this, and it decided to increase the FSB by 8mhz to 141mhz.

I ran 3DMark2001 first edition, with every setting maxed out, including 6x AA but with no FSB OC'ing, and got 6766 which is OK.

The board itself is in average condition. Many caps are domed, and some have a "dirty look" to the tops. One cap is slightly "weeping" goo. Also, the PS/2 mouse port seems to be dead. The PS/2 keyboard port is thankfully OK. I put in a serial mouse in to the one and only COM port, and that fixed the problem nicely. The serial mouse seems very responsive, both at the Windows 98 desktop, and also in DOS games.

I finally found time to play with KT133a again a little bit. I've installed Abit KT7A-Raid V1.3 with a mobile Barton 2200+ (1.35V, 13.5x133Mhz). Interestingly, it doesnt automatically remap the multipliers as it was the case with a former t-bred 2400+ that was running as 20x100Mhz. This one ran only up to 1666Mhz (12.5x133), which was really lousy.

So I found a modified r10 bios by apple_rom that was supposed to allow the higher multipliers and forced-flashed it on the MoBo. It didnt help. So I ended up tickling with registers (the famous 55th register) with wpcredit and cpumsr - successfully; but only up to 13.5x multiplier which is also too modest for my taste.

Currently the system runs at 13.5x149Mhz = ~2Ghz, but as I said this is too modest for this CPU and for the big typhoon cooler, so the next step is opening the L6 bridges to physically unlock the max multiplier. I plan to do this in the next couple of days so keep the fingers crossed - my goal is to reach 2.4Ghz with this setup.

Then it will be really more fun and games with KT133A 😉

You are having an interesting result there. I have never been able to get my KT133 based board to ever adjust the multiplier. So you have updated the BIOS, and then changed the register 55? You should get a dump of your registers and post here, if that isn't too much trouble.

Incedently, I'm using my mobile processor a KT333 based board right now, Where I can adjust the multiplier with software (although not the BIOS, strangely enough) but is won't let me make it higher then 14x, which is to the spec of this chip. Is that what unlocking the L6 bridges on this chip is for?

prophase_j wrote:

You are having an interesting result there. I have never been able to get my KT133 based board to ever adjust the multiplier. So you have updated the BIOS, and then changed the register 55? You should get a dump of your registers and post here, if that isn't too much trouble.

Yes and no. As I said, updating the BIOS didnt do much, as the matter affect - it didnt do anything at all. Tickling with register worked both with the official B4 bios and with the modified r10.

The only thing I needed to do was to load up WPCREDIT and set the second bit of the offset 55 to "1", thus transforming the value of the 55th field from "89" to "8D" (bit counting starts with 0 and not with 1, so its the third bit from the right that needs to be set). With this done, multiplier change was finally working in CPUMSR V0.90, though only up to 13.5x.

prophase_j wrote:

Incedently, I'm using my mobile processor a KT333 based board right now, Where I can adjust the multiplier with software (although not the BIOS, strangely enough) but is won't let me make it higher then 14x, which is to the spec of this chip. Is that what unlocking the L6 bridges on this chip is for?

L6 is controlling exactly that. It defines the maximum multiplier of a true mobile Athlon XP CPU. I am stating "true" because you can also mod a desktop version of athlon XP to a "mobile" version by tickling with the L5 bridges. In case of a normal athlon XP, L6 is unused.

However, even if I do have experience in bridge closing w/conductive pen, I have never cut bridges before and I'm not sure how to do it correctly. It should be easy enough with a sharp knife or a razor - but I didnt try yet. I've heard about some failed attempts, but I hope I wont end up killing the CPU. I've waited like half a year for this particular one (that can reach ~2.4+Ghz).

Tomorrow I'll try to cut only one of the two closed bridges making the max multiplier 17x. That should be enough for my setup... For now... Keep the fingers crossed 😉

Here is the screenshot as I hate making needless dumps...

Thanks for that register info. I was hoping for a dump so I could compare it with mine and possibly load that info aswell... but it's cool I'll figure it out. I'm just excited to see that you got it to work. I must have messed with those registers a hundred times with no success.

A couple of questions, if you don't mind.

1. What kind of power supply are you using? Specifically I want to know what your 5v rail is rated for.

2. What kind and how are you running that ram with the 149FSB, PC150 or did you tell the BIOS PC100, so as to use the higher divider ratio meaning your really at PC112.

prophase_j wrote:

Thanks for that register info. I was hoping for a dump so I could compare it with mine and possibly load that info aswell... but […]
Show full quote

Thanks for that register info. I was hoping for a dump so I could compare it with mine and possibly load that info aswell... but it's cool I'll figure it out. I'm just excited to see that you got it to work. I must have messed with those registers a hundred times with no success.

A couple of questions, if you don't mind.

1. What kind of power supply are you using? Specifically I want to know what your 5v rail is rated for.

2. What kind and how are you running that ram with the 149FSB, PC150 or did you tell the BIOS PC100, so as to use the higher divider ratio meaning your really at PC112.

1. Levicom silent 450W , -5V @ 0.3A, +5V @ 30A, +5 VSB @ 2A
2. Infineon 3x512MB PC133 7ns (CL2) RAM, but ran with very loose timings (3-3-3) and 4-way interleave @ 149Mhz. For some reason, RAM completely fails to overclock even the slightest bit if I use 2-2-2 or 3-2-2. I really hate this RAM... I still remember my good old Chinese NCP SDRAM that worked at 145Mhz CL2 (!), but I sold it almost ten years ago 🙁

BTW either its a blatant lie that 38MHz PCI (or higher ISA for that matter) will harm components, or I'm damn lucky. I've got a promise RAID controller (that profits from higher PCI freq btw), LAN, Voodoo 5 5500 AGP (which also profits from higher AGP believe it or not), SB AWE32 and Voodoo TV/FM. Full rack of components and so far, everything works pretty stable.

Concerning the PC-133 SDRAM, stability at higher speeds not only depends on the chips, but also on who assembled the modules.
I'm the lucky owner of some Infineon 256MB sticks which easily run at 150+ MHz CL2 without hiccups.

High PCI clocks do not harm components, but may cause them to behave erratically. For example, the onboard IDE controller on KT133A boards does not like PCI speeds above 40 MHz (160+ MHz FSB).
The "38 MHz" value comes from the time when people ran Cyrix 6x86 CPUs on Intel chipset boards. Some models needed 75 MHz FSB, resulting in 37.5 MHz PCI clock, which usually worked fine.

I did it!!! I've cut only the third L6 bridge and made the max multiplier 18x. It works and now Im in a process of fine tuning... Let u know the results a bit later 😉

Results:
System passes generic tests in Sandra at 2.35Ghz, but locks up in 3Dmark 01. At 2.25Ghz it passes correctly all the tests, however I lowered it a little bit more for the sake of additional stability so I ended up with 2195Mhz (15x146Mhz). Here is the screen-shot:

Wow super impressive man. I'm kinda suprised that your power supply is only rated 30a for the -5v, in my experiences with motherboard class I would have at least 38a for what your trying. Is your motherboard new, or recapped?

Its kinda brand new, or its supposed to be brand new. I've gotten it on ebay, boxed, unopened, hell even that "brand new"-kind of smell was there 😉 I'm also a bit surprised that everything works, since the power supply is powering 5 HDDs (4 of them in RAID 0) and a CDROM drive beside the overclocked athlon 😉

I have read that when the capacitors are new they won't require as much current. Still, it wouldn't be a bad move to get power supply a little more geared for this application. I guess it depends on whether you have long term plan for this build or not.

gravitone wrote:

ok lets see: […]
Show full quote

ok lets see:

Writeback buffer corruption when using UDMA on IDE channel 1.
SLow memory controller, VIA disabled support for memory banks 7&8 due to stability issues.
PCI bus mastering not properly implemented,(try a sblive and you'll see)
IRQ allocation issues (bios?)
Soft reboot bug which causes the screen to remain black (can be fixed in some cases by upping the Vcore voltage).

It may also be a slight personal bias, but I have yet to encounter a decently stable kt133 system. Having owned one, and having to troubleshoot a whole load of them throughout the years has given me a sour taste for VIA chipsets that will never disappear again.

The km133 series is a nightmare to work with with its horribly slow integrated savage 4 graphics core draining the already weak memory bandwidth to the point of starvation for any decent CPU.

Ok, lets see again 😀 :

Writeback buffer corruption when using UDMA on IDE channel 1:

Not true 😀 (tell us on which mainboard) - I'm using Abit VP6 board with HDD + DVD on IDE1 - no problems. And guess what sound card I use - the very SB Live! 😀...There have been lots of discussions regarding the 686B southbridge (with kt133a boards) and its faulty management with storage system + worsening this issue by corrupting large files & causing hang-ups when SB live is present in the system...I'm a living proof that this is a myth, no such thing occurs. But I have to say you are right, the 686B bug really exists, but in a specific combination of socket / NB / SB. KT133A with 686B DO have unstable times ONLY when used with socket 462 (as are my experiences), not sure if other sockets too. The S370 based mobos with these chipsets (and SB live used) do have absolutely no file damaging problems at all...there is only one tine itsi bitsy thingie, that can be easily set up in the BIOS menu where the PCI bus is to be configured thoroughly, such as options for PCI latency timer, bus mastering etc. The thing is when I'm playing Rainbow Six Las Vegas 2, in the beginning, I get horrible hiccups and lags - poor bus mastering on the PCI side. A little meddling with the PCI subsystem and I diminish the lags by approx. 70%. By removing the SBLive, no problems at all. Using an external USB soundcard is the best choice, as I free one PCI slot and it seems that they quite like my PCI USB/FireWire card - no hiccups. All to be assumed I'm testing this on my VP6 system, not others. To conclude all of this, the problems with data corruption affect specific combination of socket / chipset combo, not the chipset itself.

--------------------------
SLow memory controller, VIA disabled support for memory banks 7&8 due to stability issues:

A new info to me I must admit (about the banks), but you're right, KT133 isn't a quite a good racer in memory bandwidths, but when a good bios is written and properly set, you'll gaze on the results. With the VP6 board and 2 PIIIs I'm beating the slower Xeons, athlonsXPs and even P4s with SDRAM (Tonicom) at 150mhz CL2 all turbo settings. Will post some pictures later...

--------------------------
PCI bus mastering not properly implemented,(try a sblive and you'll see):

Discussed in the first article.

--------------------------
IRQ allocation issues (bios?):

With good bios, easily to be manually defined (thus issue solved).

--------------------------
Soft reboot bug which causes the screen to remain black (can be fixed in some cases by upping the Vcore voltage):

Where did you get this? 😕 I owned a few QDI manufactured KT133 based boards and no experience like this.

All in all, KT133/As are a great and very cheap chips, which are ONLY competitive to intel counterparts when BIOS is rich and allowing you to tweak very deep things as AGP Driving Control and its values, classic RAM settings and OC possibilities. Then you can make wonders with these mobos.

Lots of the problems with VIA boards can be attributed to a poor BIOS. That's why George Breese's PCI latency patch works. It just reconfigures the chipset after the BIOS screws things up.

With regards to PCI IRQ issues, believe it or not 440BX has serious problems there too, again possibly down to board and BIOS quality. Often it also seems that some expansion cards don't interact well together on the same INT, so the cards themselves can be the problem.

And when considering PCI issues like its latency problems and bandwidth, remember that you can still run into that problem with much newer chipsets/boards. I had trouble with nForce4 in that way. Also, you really only notice these issues with RAID and hardware-based sound cards. Other cards usually don't need enough bandwidth or bus time to be affected, or the issues are just not apparent.

swayee's right, the older sister BP6 has those IRQ issues, I'm just lucky to have the setup right...But I cannot concur with the poor bios design - jut need to play with it. PCI latency timer can be basically set up on every board, as far as I have had the opportunity to do so. Nforce 4 - it's a beutiful one, but has some serious bugs, OC is not comfortable and so...good design, bad implementation I guess.

The problem with the PCI Latency Timer setting in the BIOS is that drivers override it very frequently. Video cards in particular will crank it way up so they can hog the system bus and this does cause problems with sound cards. The move to PCIe video cards has stopped this however.

I once had a notebook with a crap cardbus controller and wanted to use a PCMCIA Audigy with it. The video chip was a Mobility Radeon 9600 AGP. I had to find a PCI Latency tweaker for Windows and tweak the cardbus bridge's latency higher and video card lower in order to minimize the extreme audio problems I was having.

http://downloads.guru3d.com/download.php?det=951

About IRQ allocation issues - what are you guys talking about? I've used KT133A continuously since it was developed and I have never had any issues with IRQs. I've had Asus A7V133-C, MSI KT7a, K7T-Turbo (the RED one 😉 ), K7T-turbo2 and of course Abit KT7a, v1.0, v.1.1 and v1.3. Neither of these mainboards had any issues with IRQ handling. Just go to bios and select "set IRQ resources manually" and choose an IRQ for legacy ISA that you want to reserve.

If I may say so, I find VIA KT133A one of the best chip sets VIA ever produced. However, I cannot say much about using Live! with them, as I've used CMI8738 at that time (which worked perfectly) and now SBAWE32 and Terratec DMX Fire! (which also works flawlessly).

Seriously guys, I think most of you are still remembering the crappy MVP3 and avoiding VIA because of it 😜

Well in my experience it's more an issue with some cards not getting along on the same PCI INT. For example, I've seen problems with sound cards sharing IRQ/INT with other cards, to the point of the board not even POSTing.

swaaye wrote:

Well in my experience it's more an issue with some cards not getting along on the same PCI INT. For example, I've seen problems with sound cards sharing IRQ/INT with other cards, to the point of the board not even POSTing.

Oh that. Well... I do remember having issues with Miro PCTV back at the time. That card always wanted to steal the IRQ from video card... until I disabled "assign IRQ to VGA" and placed Miro in another PCI slot.

It seems that the first and the last PCI slots are sharing IRQs with AGP and ISA slot respectively. To avoid the fuss, avoid using both of these at the same time. Then everything should be fine.