Where did the X800XT-PE go?

R

Robin B

Limp Gawd
Joined
Jul 14, 2004
Messages
209
What happend to the XT ? Have been waiting for one some time now, but here in Norway it seams that i`l have to wait to this fall before i can get my hands on one. Funny thing tho to read all the !!!!!!s claiming that the XT is cloosing in on the 6800U in Doom when the card is nowhere, and by the time the XT is out NV and Ati is about to release theire refresh part. So what happend ??
 
despite all indications to the opposite, ATI apparently hasn't had yeilds sufficient to produce any volume of them. despite hitting the market first, in terms of volume, NV is actually shipping more ultras it seems.
 
Hmm Kombatant over at Rage3d have said that the yeilds are not the problem, but after i read the latest review from bjorn3d.com i think that they are having big problems getting theire XT-PE cards to run at 520Mhz.

Bjorn3d didn`t get more than 10Mhz overclock from the core with the HIS Excalibur X800 XT IceQ II VIVO.

http://www.bjorn3d.com/read.php?cID=668&pageID=881
 
Chips not binning high enough is a yeild problem. Intel has huge yeild problems with the xeon MP chips, that is why there are so many xeon chips w/o the cache, to make a profit on failed top bin chips. every xeon chip intel makes is targetted to be a xeon mp w/ 2mb of cache at the highest speed grade offered (3.2ghz). of them, most fail to reach that mark, so yeilds on the top chips are bad, those rejects are then sold as lower grade chips. the same holds true for r420 chips, the fact that there aren't enough that make the speed and quad grade indicates a problem with yeilds, as ATI clearly would like it if all of them functioned at 520 mhz w/ 16pipes.
 
The way it looks i think the XT was not ment to be out before this fall, sadly. :(
 
I heard elsewhere about not supposed to be out before fall...to my knowledge you can't just flop on 4 extra pipes at the last minute...so I dunno if that makes sense...
 
^eMpTy^ said:
I heard elsewhere about not supposed to be out before fall...to my knowledge you can't just flop on 4 extra pipes at the last minute...so I dunno if that makes sense...
Click to expand...

Hitting XT-PE speed is the big problem..
 
Seems like these two companies just made a really fast card to out do each other and they never really expected to sell them. I am glad because there are a ton of 9800XTs still out there that cost more than the 6800GTs and they are always going to cost $400 even though they are slower and more expensive.
 
Get used to it guys. There will be more yield problems in the future. As the dies have more and more transistors yields are going to head south. Not to mention the oxide is getting so thin that leakage is becoming a serious factor. So no longer are you getting the advantages of reduced power from a die shrink. More heat over smaller area=bad.

I'm not saying there is no way to stop the trend, but it will require some more innovation, but I guess that is what us engineers get paid the big bux for. Of course in the short term, a lot of companies will have problems. Just look at intel, ati, and plenty of others to go around.
 
I guess don't understand how the pipes are setup. Do some just "not work"? Can someone explain this to me? I would think if they designed a chip for 16 pipes then wouldn't all 16 work? Or do some fail when the chips are produced?

How exactly ARE the chips produced, anyways?
 
Well my BBA X800 Pro just died.

I'm very sad.

I'm going to RMA it to ATI.

Given the current shortage of X800s, It may be months before I get another one.

Meanwhile, I just got a cr@ppy PCI VGA card that has no 3D capability at all.

NO GAMES FOR AT LEAST 2 WEEKS AND PERHAPS MONTHS :(
 
Rizen said:
I guess don't understand how the pipes are setup. Do some just "not work"? Can someone explain this to me? I would think if they designed a chip for 16 pipes then wouldn't all 16 work? Or do some fail when the chips are produced?

How exactly ARE the chips produced, anyways?
Click to expand...

Doesn't work that way. When create a wafer of chips you have variation. It comes from many things ranging from the wear of the tools to the quality of the silicon.

Now, up until recently yields were pretty good. However, with the increased complexity of chips and their decreased feature size, every little defect will manifest itself. They are making advances to counter the ever shrinking size of chips, but that is why fabs get more and more expensive. You have to get better clean rooms, better equipment, better tools, etc.

The reason for the low yield on the most recent batches of chips is most likely because of their 200million transistor count. Ideally all transistors would come out identical and you would have near 100% yield. However, because of different factors there are variations in the transistors. Some of this variation has no effect, but lots of the time it can cause speed path issues or requirement of higher voltages to operate, or complete failure of different units of the chip. One of these units could be a 4pipeline quad. So all you do is simply turn that quad off. CPU maker do the same thing when they have cache errors. They just turn off the section of cache that has an error.

These variation in transistors are the reason for the bins. Top bin is usually the highest performing part. It could have the highest frequency, or the highest amount of cache, or both, or some other feature. Then you usually have lower bins where the part didn't test as high on frequency or some cache failed etc.

Now just because something qualified for top bin doesn't mean it will be sold as the highest frequency part. If the yield on the top bin is very good and the demand on lower bins is more than the supply, they will just mark top bin parts as lower grade. That is why some lower speed chips will overclock higher than others, because they were actually a top bin part.

Now through process advancements and different steppings/revisions it is possible to reach higher and higher frequencies by tweaking different things. This is why companies are able to release faster and faster chips on the same basic architecture. They have had the time to test different fixes and different techniques and they can actually get greater yields & speeds out of the same chip with just a few tweaks.

Of course gfx cards are a little different than cpus. With cpus you usually have a life span of a couple years and a development time of several years. This allows you to handpick and layout each transistor. Gfx cards have a life of about 6months and a much shorter development time. They do not have the time to test out as many fixes or do as much tweaking as a cpu. This is one of the big reasons gfx card cores are so much slower than cpus in terms of frequency, but it is also the reason they are able to have many more transistors.

Anyways hopefully that helped you out a little. It is a very small summary of what is actually going on.
 
You must log in or register to reply here.
Back
Top