Safe to mount a G400MAX on a i845D chipset?

[IceStorm]

Heh. Haig posted his last $0.02 just under the wire in the P4B thread. He doesn't seem to get that the notching on the card must correspond with the appropriate TYPEDET# signal - it's part of the AGP spec. I really don't care who screwed up, Matrox or the PCB fabbers, the fault, in my opinion, is with the company that sold the card to the end-user. It looks like they're going to have an official statement on the issue next week. I hope they don't cop-out and put the onus of this problem on the buyers and Asus.

Anywho, the P4B266 will not power up if the card inserted has TYPEDET# left open. Since the early G400 MAXes left TYPEDET# open, the board shouldn't power up. If you want to use an early G400 MAX (or any other early G400 that wasn't wired up properly), don't use a 1.5v board that supports TYPEDET# detection.

[Haig]

He doesn't seem to get that the notching on the card must correspond with the appropriate TYPEDET# signal - it's part of the AGP spec.

Follow me here The 4x indication notch will allow the card to fit in a 4x or a universal agp slotted motherboard. Typedet has nothing to do with this.

The poster wants us to remove the notch for the 4x indication. I am telling you that removing the 4x notch would not allow the 2x version of our card to work in a 4x only motherboard.

Haig

[Kruzin]

And how do you remove the notch from a card that's no longer in production? Drive to every early G400 owner's house, and epoxy the slot closed?

[Greebe]

Naw... just use some scotch tape

[Kruzin]

Or perhaps bubble gum?

[IceStorm]

The poster wants us to remove the notch for the 4x indication. I am telling you that removing the 4x notch would not allow the 2x version of our card to work in a 4x only motherboard.

He did not say he wants Matrox to remove the notch. What he said was:

If Matrox didn't want the G400 cards to go into 1.5volt AGP slots they should have written AGP2x on the box and keyed the bottom of the card to prohibit it.

He's not refering to the present, he's saying that before Matrox made the cards they should have properly keyed the G400 MAX cards to match TYPEDET# (in this case 3.3v). That would have meant a single notch, like the G200 has.

Follow me here The 4x indication notch will allow the card to fit in a 4x or a universal agp slotted motherboard. Typedet has nothing to do with this.

First, the notches have absolutely nothing to do with the way Universal slots determine AGP signaling level. Universal slots have no dividers. They rely exclusively on TYPEDET# to determine signaling level.

Second, TYPEDET# has everything to do with this, both with what JohnK is talking about and the larger issue, that the older G400 MAX doesn't work with the P4B266 line.

The notches and TYPEDET# are supposed to be one in the same. The notch on the card has to match the TYPEDET# level from the board - either ground or open. I've already made explanation of the only "violation" of that rule which works with the spec. The only slot that doesn't have explicit TYPEDET# checking is the 3.3v one. Universal and 1.5v slots support checking of the TYPEDET# signal as per AGP spec. And yes, I realize that the slot isnt't doing the checking - the AGP Master or an external circuit wired to those slots is doing the checking. I'm assuming the Master and circuitry are wired to the appropriate slot.

There's a notch on my two G400 MAX cards indicating 1.5v support. There's a TYPEDET# signal on both indicating 3.3v support. The two are NOT mutually exclusive - they're supposed to match. JohnK was stating that since the first-run G400 MAX cards are wired up with TYPEDET# open, they shouldn't have a 1.5v notch. He's right. They shouldn't, as per spec. I've already stated that there aren't any card vendors that I know of who actually followed the spec when they moved to 1.5v AGP signaling support (dual mode cards aren't part of the spec). But the majority seem to have wired their cards with TYPEDET# grounded. My G550 cards do, and they violate spec. My GF2U does, it violates spec. My GF2MX-400 does, it violates spec. They all have TYPEDET# wired to ground. My G400 MAXes left it open, they violate spec. The difference is the G400 MAX violates spec in the "wrong" way.

Matrox is getting bit in the ass for promoting their G400 parts as "AGP 4x compatible" at lauch over two years ago - You remember, when we found out later that all it meant was that the card would work at 1.5v AGP signaling instead of actually supporting the bulk of AGP 4x features. The chipset might support 1.5v signaling, but a chipset is not the only thing which determines if the card meets AGP spec. Certain revisions don't meet spec, even in the "wrong" way. I look forward to seeing how Matrox is going to deal with this situation.

[Wombat]

The real problem here is one of timing. Intel was short-sighted when they didn't spec out a "universal add-on component." It looks like they just did a copy-pasted-edit with the 5v/3.3v part of the PCI spec (see page 119-121).

But it looks like intel didn't have enough coffee the morning they wrote this, trying to leverage the PCI methodologies. The PCI spec says that 5V motherboards supply 5V, and +/-12V, with 3.3V optional; 3.3V motherboards supply <B>all four</B> rails (page 142). With PCI, it is the <I>card</I>, not the motherboard, that has the "universal" capabilities. It seems that intel took a working protocol, changed a few subtle but critical aspects, and didn't realize they had opened a new hole with this new way of doing things. Short (G-rated) answer: "Whoops!"

So, Matrox and everybody else found some way of showing the card as the motherboard needed it. Matrox's method would seem to emulate the "universal" style PCI boards. Not everyone took that approach, but they had to do something. So, any video card that supports both voltage levels does not meet the spec.

I just read the AGP spec cover-to-cover (again). There's not a whole lot about the initialization sequence. Matrox determines TYPEDET one way, and this MB is trying to detect it in a different way. There's no spec for this part or the process. It would be nice if it were in section 5.4.5 or 6.1.2.4, but it's not.

As far as I'm concerned, both Matrox and the P4B are working as well as the AGP spec allows them. Unfortunately, they handle the gray area differently.

I am with Ice that Haig has part of things wrong. There are no "2x/4x" keyings. What there are are "1.5V/3.3V" keyings. 4x transfers can only be run at 1.5V, but I saw nothing in the specs that says that a 1.5V motherboard <I>must</I>support 4x, even if it is about the only reason to go through the effort of supporting that signalling level.

[Greebe]

I've read the AGP 2.0 spec over several times now and have a few chronological clairifications to what both IceStorm and Wombat have said... there is a "2x and 4x" versions of the AGP slot. Follow me on this (just like Haig said
Back prior to the AGP 2.0 spec there was only a 2x slot MB's as it was known to the average Joe. These operated at 3.3v. When the AGP 2.0 spec was released it was all about AGP 4x operation and the average Joe saw that the AGP connector had been reversed. Why, because they didn't reverse the slot until AGP 4x capable MB's were produced (ie 1.5v signaling which only 4x boards needed to support the differential inputs that were neccessary to insure low noise operation/signal integrity that operation demanded). When the G400 was designed there wasn't a finalized AGP 2.0 spec. It came shortly afterward (May of '98) and back then we were still using the AGP 1.2 spec as a guideline and a unfinished spec for what nobody honestly knew was truely going to be fully unvailed until these Intel 'AGP4x' MB's actually became reality for them to test on... and that was still many months away. Besides you could only build in what you could foresee and Miss Cleo wasn't around back then.

So now while the G400 in it's various forms has had a wonderful success in the marketplace it is only now with the latest incarnation of Intels interpretation of the spec that's causing problems for the early revision of the card. Too much grey area in that section of the spec and not enough grey matter... or was it?! Why was this done possibly... because Intel sees Matrox (and all other Gcard manufactures) as competition. Especially when you write the spec, anyone you can snub makes perfect business sense.

Naw

Now wether that is true or not isn't the issue. What has happened is just too loose of terms in the spec. Besides... who wanted to wait even longer for G400 back then?! It would have upped the developement cost even more... and delayed the card that much longer... for an early version of the G400 with nothing to test it on that would challenge it's design for nearly ~4 years (<3.5 for most of us)

I don't think so!

[IceStorm]

Follow me on this (just like Haig said

Except Haig got his tenses wrong and started this whole thing. You and Kruzin just followed along without actually reading what was posted.... "Do as I say, not as I do," as it were.

When the G400 was designed there wasn't a finalized AGP 2.0 spec. It came shortly afterward (May of '98)

You're suggesting that Matrox designed the G400, then sat on the chip for a year? The G400 card was released around May/June 1999, not 1998. The G400 chips were done and sitting around doing nothing in April of 1998? Besides, this problem certaintly doesn't look like a chip-level issue. It looks like a PCB-level issue.

Besides you could only build in what you could foresee and Miss Cleo wasn't around back then.

It's taken me about an hour to wrap my head around the mechanical section of the spec well enough to understand that the only loophole is to set up any dual mode cards with TYPEDET# tied to ground. Matrox couldn't do the same in 12 months time?

The real problem here is one of timing. Intel was short-sighted when they didn't spec out a "universal add-on component."

You don't always get what you want, although I can see why Intel would prefer to spec a universal slot over a universal card.

Now wether that is true or not isn't the issue. What has happened is just too loose of terms in the spec. Besides... who wanted to wait even longer for G400 back then?!

It wasn't the most straightforward spec ever written, but it's not entirely ambigious.

I don't have a problem with Matrox screwing up, provided they do the right thing in the end. What I have a problem with is when people tell me I'm wrong for saying so without providing any proof to back up their assertions. Saying that the card notching and TYPEDET# aren't related is false. If you want me to believe otherwise, you're going to have to prove it.

[KevinST]

Simple question...... did Matrox intentionaly not connect the TYPEDET to ground on the very first revision of the G400, or was it a small design / manufacturing mistake???

All subsequent G400's have had TYPEDET tied down to ground, and even Haig has said the TYPEDET should be grounded, so I assume there's no question as to "if" the TYPEDET pin should be grounded (like there was back at Christmas when I raised this issue, or even at the beginning of this thread).

[Greebe]

Kruzin nor I misunderstood what Haig said... and I don't believe he did get his tenses wrong either... it's just the difference between talking to Joe Schmoe off the street 24/7 and Engineers 24/7... if you have to address one predominately more often than another (especially in forums like this) then you typically do the same always... unless it's neccessary to clairify in further detail.

Besides Ken and I were only havin fun... can you not see that? (it was late on a Friday night... geesh)

OK I was off by a year... big flyin monkey poop! I knew that didn't sound right when I posted it but it was late and said what the heck and posted anyway, then crashed.

I don't have a problem with Matrox screwing up, provided they do the right thing in the end.

Did they not make the neccessary change once they could test it on the very first AGP 2.0 slot mb's?

YES... to TYPEDET#

What I have a problem with is when people tell me I'm wrong for saying so without providing any proof to back up their assertions. Saying that the card notching and TYPEDET# aren't related is false. If you want me to believe otherwise, you're going to have to prove it.

The keying has nothing todo with it in the real world. Do the keyings actually signal anything, provide some form of feedback or the like? No not at all. Back then there weren't universal slots, they were all keyed one way or another. Sure the spec says somewhat differently in what todo, but then Matrox's solution was completely different that what others had done and that's called innovation. Make it FIT in whatever AGP slot which will ensure it'll work with all MB's available so maximum compatability (and sales) is insured. This is business ie the REAL world... which is something that the spec does not take into concideration.


When you have a company like Intel which writes a spec with as many loopholes and grey area as this spec has init and has a history of making agonizing changes to how they implement a spec... well you should see how that's a major problem, NO?!

Besides the bottem line is this... there still wasn't a single mb on the market which had an AGP 1.5v slot til well after the G400 was introduced and it's only now there's a problem with how Intel decided to implement the spec differently yet again. It's much easier to further clairify a spec which has been around for better than 3.7 years on paper than to address what would have been done in the REAL WORLD at the very beginning of it's conception in hardware. I know I've been there when designing electronics... it's a real PITA... especially when your waiting for multiple entities to actually cement the standard which affects what you're all trying to achieve.

[IceStorm]

Kruzin nor I misunderstood what Haig said... and I don't believe he did get his tenses wrong either...

So you three joking that JohnK wants Matrox to go around removing the notch on old Matrox cards was something that came out of JohnK stating Matrox shouldn't have put the notch there in the first place?

No, you two and Haig misread what JohnK said in the first place. JohnK pointed out that the notch shouldn't be there, not that it should be removed after the fact. You all got the tense of his argument wrong. I can see you and Kruzin not bothering to read the Matrox forum thread I linked to, but ignorance is a poor excuse.

Besides Ken and I were only havin fun... can you not see that? (it was late on a Friday night... geesh)

Sure, except that you're doing it at the expense of an end-user, and that end-user didn't say what you three think he said.

Did they not make the neccessary change once they could test it on the very first AGP 2.0 slot mb's?

YES... to TYPEDET#

But this isn't relevant. The specification didn't change between the time the card was designed and the first AGP 4x cards hit the market (unless we're going with the PCB-built-in April, 1998 theory). The cards should have been built properly in the first place.

The keying has nothing todo with it in the real world.

I'd call keeping the card OUT of slots it doesn't belong in a "real-world" use for keying.

Do the keyings actually signal anything, provide some form of feedback or the like?

No, but the TYPEDET# signal which is supposed to match the keying is used in two out of the three slot types in the spec. Keying should tell you what TYPEDET# is set to.

Make it FIT in whatever AGP slot which will ensure it'll work with all MB's available so maximum compatability (and sales) is insured.

Making it fit while violating spec in the "wrong" way is not ethical.

Besides the bottem line is this... there still wasn't a single mb on the market which had an AGP 1.5v slot til well after the G400 was introduced and it's only now there's a problem with how Intel decided to implement the spec differently yet again.

This is not the first time that I've heard comments about how Intel is implementing their own spec "differently". The spec hasn't changed since May, 1998. Exactly how is Intel implementing it differently? Tell me when/where the spec was appended between now and May, 1998, to cause this situation. I get tired of people saying Intel changed something, yet never pointing out what it is.

The "bottom line" is that Matrox read the spec, didn't fully understand the ramifications of leaving TYPEDET# open, and built PCBs that not only violate spec, but violate it in the "wrong" way. Now there are people left with cards that won't work in the P4B266 because Asus added a feature to prevent non-1.5v compliant cards from destroying the motherboard.

[Greebe]

drop back two steps and PUNT!

Ice do you have a clue as to how any electronics design and specs apply to reality... I honestly don't think so.

[xortam]

Much to do about nothing. How many people are affected by this? Matrox might even exchange their cards for the few customers that run into this.

How about all the HW that is now useless because the manufacturers don't see a business case for delivering new drivers for new OSes? That's the biz.

[Wombat]

But this isn't relevant. The specification didn't change between the time the card was designed and the first AGP 4x cards hit the market (unless we're going with the PCB-built-in April, 1998 theory). The cards should have been built properly in the first place.

The spec formed while the G400 was in late stages of development. It's not a "theory we're going with." If Intel had really cared, they would have taken outside input or something. And then every video card maker out there would have had something to say to fix the gap in the specs.

Making it fit while violating spec in the "wrong" way is not ethical.

This was not a choice about ethics. It's a quality card, with an admitted kludge to overcome a hurtle all video card mfrs encountered. I also don't know how you deemed Matrox's implementation "violating the wrong way." I checked, there's no appendix to the spec on "How to Correctly Violate the Spec."

but ignorance is a poor excuse.

And being smart enough to read a spec doesn't excuse a combatant, shitty, arrogant, personality, but you keep going anyway.