Update: I made some high resolution photos.
Version 2.3:

Version 2:

The most obvious difference is additional capacitor on the back in Ver 2.3. I read somewhere that it should be 47p. I only had 18p of that size and soldered it, but didn't help. Other difference is value of RNs. They are 750 in Ver 2.3 and 560 in Ver 2. And different revision of TVC chip. Should I try to replace resistor arrays on Ver 2 to match these on Ver 2.3?

No idea what that tvc15222 does. If it's done kind of transceiver, the "A" suffix on the v2.3 could easily mean a higher speed grade. If that is the case, the resistor network is probably another value in order to impedance match at higher frequencies.

Maybe it's tvc16222 as this chip from Texas Instruments seems to be GTL+ bus compatible https://www.ti.com/product/SN74TVC16222A

Its the voltage clamp. The most important is to have one as to not get the wrong voltage.

Yes, it is. But i found an image of 2.3 revision with non-A clamp chip: https://apress24.pl/media/catalog/product/cac … _patrycja_1.jpg
it still has R750 resistor networks. So maybe i should try replacing them with fine nozzle hot air station and some kapton tape. At least if no-one can report a successfully working ver 2 slotket with 133MHz FSB... I can't spot any differences in PCB layout between these two revisions.

Let's do some google magic, and check the datasheet directly.
https://www.ti.com/lit/ds/symlink/sn74tvc16222a.pdf

Older TVCs are early revisions of this one, but the newest one has facilities for Tualatin's
https://forums.overclockers.com.au/posts/726372/

Google is your friend.

hyoenmadan wrote on 2020-11-24, 21:57:

Let's do some google magic, and check the datasheet directly. https://www.ti.com/lit/ds/symlink/sn74tvc16222a.pdf […]
Show full quote

Let's do some google magic, and check the datasheet directly.
https://www.ti.com/lit/ds/symlink/sn74tvc16222a.pdf

Older TVCs are early revisions of this one, but the newest one has facilities for Tualatin's
https://forums.overclockers.com.au/posts/726372/

Google is your friend.

Thanks, but I already have seen that datasheet and there is no mention of older TVC16222, so I can't compare the specs. They may or may not perform the same. I checked that overclockers forum, but I don't think that it directly proves or disproves that -A version is needed for newer CPUs. They are comparing totally different slotkets. Anyway I think i could just try replacing all components to match 2.3 revision, it shouldn't make anything worse I guess. Probably it will be easier to start with RNs (better availability)...

adalbert wrote on 2020-11-24, 22:28:

hyoenmadan wrote on 2020-11-24, 21:57:

Let's do some google magic, and check the datasheet directly. https://www.ti.com/lit/ds/symlink/sn74tvc16222a.pdf […]
Show full quote

Let's do some google magic, and check the datasheet directly.
https://www.ti.com/lit/ds/symlink/sn74tvc16222a.pdf

Older TVCs are early revisions of this one, but the newest one has facilities for Tualatin's
https://forums.overclockers.com.au/posts/726372/

Google is your friend.

Thanks, but I already have seen that datasheet and there is no mention of older TVC16222, so I can't compare the specs. They may or may not perform the same. I checked that overclockers forum, but I don't think that it directly proves or disproves that -A version is needed for newer CPUs. They are comparing totally different slotkets. Anyway I think i could just try replacing all components to match 2.3 revision, it shouldn't make anything worse I guess. Probably it will be easier to start with RNs (better availability)...

It's not a v2 / 133MHz thing as I've just tested a SL52R coppermine in a MS6905 v2 on MS6163 Pro (Ver 2) BX board and it happily boots at 1GHz (7.5 x 133 OC)

Hmm, so I wonder what may that be. I swapped the CPUs between slotkets and same SL5XL didn't work in V2. Do you have R750 or R560 resistor networks?
I found out that under ver 2.3 sticker is ver 2 silkscreen. So ver 2 may be ver 2.3 without a sticker.

Maybe my motherboard is more sensitive to signal level it gets from the cpu than others.

I have a number of these MS6905 slokets, but only one v2.3 (sticker rather than silkscreen) which like yours has R750 resistors but the non-A version of the TVC16222. The rest, including the one I tested here (new out the box) are v2 (silkscreen) with R560 resistors and again, the non-A version of the TVC16222. The only jumpers I changed out the box were setting J1 & J4 to 'open' for 133MHz FSB and left the voltage on 'auto'.

The version numbering of these is always a bit of a lottery and in fact there are various boards labelled as v2.3 - with and without the later version of the voltage clamp IC and the 'true' v2.3 which has the 4-pin J3 jumper for dual coppermine support without the need for the N33 wire mod.

adalbert wrote on 2020-11-25, 01:01:

Hmm, so I wonder what may that be. I swapped the CPUs between slotkets and same SL5XL didn't work in V2. Do you have R750 or R560 resistor networks?
I found out that under ver 2.3 sticker is ver 2 silkscreen. So ver 2 may be ver 2.3 without a sticker.

Maybe my motherboard is more sensitive to signal level it gets from the cpu than others.

Well, if ver 2.3 has ver 2 in silkscreen, I would suspect that you're right about PCB being identical.

If you're ok with the risk of maybe bricking your ver 2 slotket, it would probably be worth the while changing the resistor networks. Especially since it sounds like the TVC16222 may not need to be an "A" ..

adalbert wrote on 2020-11-24, 18:49:

I successfully modded MS-6905 VER 2.3 (i made 3 holes larger and insulated them), it works with wire-modded Pentium III-S SL5XL 1.4GHz. I wanted to add a second CPU, so I performed the same mod with VER: 2 slotket.

Is the mod done right on the ver 2? I can't see the insulation, however, I do see the pins in the enlarged holes on your photo?

I don't have any Intel Socket 370 CPUs, but my MS-6905 V2 works perfectly with VIA C3 CPUs at 133 MHz FSB.

H3nrik V! wrote on 2020-11-25, 05:18:

Is the mod done right on the ver 2? I can't see the insulation, however, I do see the pins in the enlarged holes on your photo?

It was insulated but i removed the insulation before taking the photo as I was trying to check if it works with regular 133MHz Coppermines. And it doesn't work with any 133MHz CPUs, even unmodded Coppermines. And I tested regular Coppermines even enlarging the holes, didn't work at 133MHz FSB.

I'm trying to find MS6905 v2.3 locally, I already contacted someone so maybe I won't need to do the soldering.

//EDIT: i just found this info about MS6905 recall https://www.overclockers.com/msi-replacement-offer/

Therefore, some early shipments of the MS-6905 Ver 1.1B were labeled “Coppermine Ready” on the color box. Due to a sudden change notice from Intel recently, design specifications were altered to properly support Coppermine FC-PGA processors. In order to support Coppermine FC-PGA processors, motherboards and CPU converter boards need to provide a different pull-up resistor on legacy CMOS signals. These changes require hardware modifications.

they are mentioning wrong pull-up resistors. Ver 2 is supposed to be "fixed" version and I can't find any image of version 1.1B. But maybe these resistors still were unsuitable for some motherboard/CPU combinations and it was only ver 2.3 that was the most compatible.

Now my Rev 2 slotket stopped working completely. I measured 7 ohms on some ceramic and electrolytic capacitors, so I hooked up a battery pack and sprayed the board with alcohol to find out where it evaporates quckly. It was that ceramic cap close to the edge. Who would've expected such failure. Works again after being replaced. But it still doesn't work with 133MHz CPUs in my dual CPU board, while Rev 2.3 works.

I just found in Intel's datasheet that 56 ohms (not 560; SMD R560 code = 56 ohms) is valid for "Coppermine-T UP" plaftorms, while "Coppermine-T DP" requires 68 ohm. But maximum resistance is up to 115 ohms.

https://www.dexsilicium.com/Intel_Pentium3.pdf

Tualatin datasheet has similar specs. "Uni-processor platforms require a 56Ω resistor and dual-processor platforms require a 68Ω resistor. Tolerance for on-die Rtt is +/-10% (56, 68Ω resistors). Rtt is +/-15% (100Ω resistors)." http://www.frsn.utn.edu.ar/tecnicas3/manuales … ata%20Sheet.pdf
I tried runing 1.4GHz Tualatin on Rev 2 slotket with 100MHz FSB, but it didn't work at all. Only 100MHz coppermines work, or 133 ones downclocked to 100MHz.

I installed two matching 933@700MHz Coppermine P3's on both slotkets and they worked one at time, but the board didn't manage to go to the POST screen in dual setup. But I was getting some POST codes on diagnostic card. I did N33 mod and it didn't help.

So I will try replacing these with 82 ohms ones (75 ohm ones are out of stock everywhere). These are 10 pin bussed type resistor networks. I might replace that TVC chip to newer revision as well. Both can be bought at Mouser. Probably will do the TVC chip first and check results.

Please use this guide for performing mods (at least for getting to work in single mode): http://www.rom.by/articles/S370/_english_part3.htm
And you will not get it to boot or stable running in most cased without bridging AK4 to VTT (preferrably through resistor, but works mostly without too).

Paadam wrote on 2020-12-22, 07:59:

Please use this guide for performing mods (at least for getting to work in single mode): http://www.rom.by/articles/S370/_english_part3.htm
And you will not get it to boot or stable running in most cased without bridging AK4 to VTT (preferrably through resistor, but works mostly without too).

Thanks, this is a very detailed documentation. Anyway I already had bridged AK4 to VTT on cpu (forgot to mention it). But still the main problem is that MS-6905 ver 2.3 works with every CPU i install there, and MS-6905 ver 2 doesn't (even in single CPU mode), even though it has identical mods and jumper configuration. Only some 100MHz FSB cpu's works. The problem has to be with the slotket itself (i can't imagine any other reason of this issue). They have different resistor values and different revision of TVC chip. So I will wait until parts arrive and resolder everything (TVC chip, resistors) with the same models as in 2.3 slotket.

Important factor in this failure may be that Tyan mainboard, which is probably one of the earliest Via 694x dual CPU mainboards, was pulled out market very quickly and may have compatibility problems. But this is the only dual slot mobo with AGP (and ISA, but that's not so important) and >1GB memory support I was able to get. It is then possible, that these slotkets would work perfectly fine in other mainboards, but I don't have any way to check it, and even if they would, i need them to work with this particular hardware 😁

//edit: I also found two small resistors with different values and I'm trying to investigate if it can relate to the failure
//edit2: whatever, they seem to be the same, just different tolerance (150ohm / 151 vs 18A marking)

//edit3: upgradeware-T native tualatin slotkets seem to be using 120 ohm resistor ladders, so it seems that higher resistance really may be needed. With good understanding of documentation suitable values could be probably calculated. I guess that wrong pull-up resistor values may cause CPUs signals to be unreadable.

Update: Now I have two MS6905 2.3 slotkets. They both have 75 Ohm resistor ladders, but one has TVC16222A chip and other has TVC16222 chip. One with TVC16222 chip also doesn't have additional small SMD capacitor on the back.

The result: Each one of them works with 1400MHz 133FSB Pentium 3 Tualatins (SL5XL) in single CPU mode. So, it looks like resistor ladder values were more important than the TVC chip itself. However...

For dual Tualatin mode i had to connect X2 pin (not N33) to B75 pin on both slotkets with a wire. That's the only mod required so far. BUT it works in dual mode only with 100MHz FSB set...

With 133MHz it gets stuck on C0, C1, or CF code.

Maybe I will try soldering TVC16222A instead of TVC16222, as I bought some spares and see if it works, I can also try soldering missing capacitor. But I have proper dual S370 rig now, so that's not high priority. Anyway, maybe someone will find this info useful: resistor values on slotkets can be important.