R600 Delay - The Reason

[the Chase]

Originally posted by: Wreckage

Originally posted by: terentenet

Originally posted by: the Chase
ATI has been respinning this chip trying to best the G80 performance.

Another month or so delay would not really allow for any major hardware changes. They may have just had similar problems with their Vista drivers and are gun shy after all the crap NVIDIA has been getting lately.

Well yes but mid Q1 to mid-late Q2 would give them up to 4+ months for a respin. If it was just a delay of a month I would think they would have been a little more specific than just "Q2".

 

[zephyrprime]

I'm sure it was #3: some weird bug cropped up. I have no proof but that is the only reasonable thing I can think of. Even if it is not as good as the G80, delaying release to improve it's performance would hurt them more than help them overall in my opinion.

 

[apoppin]

Originally posted by: SexyK
As the days go by I'm starting to believe more firmly that some type of hardware bug was uncovered. I can't really think of any other scenario that would lead to this turn of events (i.e. scheduled launch, plane tickets, hotels, etc... then canceled).

Just because they (presumably) had been testing production silicon and decided to schedule the launch doesn't mean that all possible usage scenarios had been tested. I could see the situation unfolding something like this: ATI/AMD get production silicon to their testing centers and begin polishing drivers and testing popular software. Weeks go by without issue, the chip seems to be performing well, and driver development is on course, so they schedule launch. A few more weeks go by and testing continues. One day someone loads up a new (or old) piece of software to further test compatibility, a specific feature or even a specific piece of shader code and *whoops* it crashes. Or it performs WAY under expectations. Revised drivers don't fix the problem. Uh oh, time to look at the hardware. Small bug (or large bug) is uncovered in the hardware and it's simply impossible to fix the design or code the drivers around it in time for launch.

This doesn't mean the whole architecture is FUBAR by any stretch of the imagination, but it's very possible that some aspect of the design that isn't central to usability wasn't tested sufficiently before scheduling the launch. Even though it could be an obscure usage scenario, they decide not to launch until its resolved on the off chance that the problem is larger or has other unforeseen consequences. Remember the case of the flaw in the original Pentium, which caused no problems for 99.9% of users in 99.9% of usage scenarios, but still caused a major media backlash and precipitated a recall? That's the kind of situation I suspect they're dealing with and trying to avoid.

*STOP Making Sense!*
:Q

no one is listening to us

they would much rather believe:

either

1) AMD is going down the tubes

or

2) Nothing is wrong ... it's a strategy change



i think i's a big issue and they will fix it soon

if not


worst case - r660 in the Fall
- to compete with whatever nvidia throws at it

--right now, it's called a *setback*

 

[nrb]

Originally posted by: apoppin
sure ... they do extensive testing

and expect to fix some bugs as they progress

it appears they found something *unexpected*
--or they would NOT have scheduled the launch in the first place

Precisely. Which means it cannot possibly have been the result of generally poor performance. If the problem is "the chip is too slow compared to G80" they would have known about that long before they booked the flights to Amsterdam: any testing at all would have revealed that.

The fact that the launch was cancelled at the last minute means that it must have been a bug: most likely a hardware bug, possibly a problem with the card design. It must be something that doesn't go wrong often (so it slipped past the testers) but, when it does go wrong, goes wrong so catastrophically that the product simply cannot be released until the problem is rectified.

It cannot be a general performance problem. (It is conceivable, of course, that the bug is performance-related, but only manifests itself in very specific circumstances).

 

[apoppin]

agreed .. that's what i said

a HW "issue"

strange ... x1800 ....x2800

i think they should get out from under the 8-ball


my 'advice' to AMD - forget eights ... stick with 9's

 

[nrb]

Originally posted by: Wreckage
Another month or so delay would not really allow for any major hardware changes.

Who says it's only going to be delayed by a month? All ATI has said is "Q2". That could be anything up to a three-month delay.

 

[TanisHalfElven]

Originally posted by: josh6079
And this is different from the other "R600 Delayed AGAIN!" thread how?

it just started.

 

[jdoggg12]

Im curious... why are some of you so passionate about speculating why a card is delayed? Get out and enjoy some sunshine or something! It's just a video card, these ridiculously long explanations of what 'could' be going on simply amaze me. I come to threads like this hoping to see some info backed by facts... but the 50 or so pages about the R600 in various threads offers nothing substantial.

Is it so you can throw a guess out there and say ' i told you so ' if your speculation hits near the truth some time down the road?

 

[nrb]

Originally posted by: terentenet
I don't seem to find the link and I'm too lazy to search the net for it again, but when the R600 specs were first leaked, ATI even said something about "being faster with the proper software tweaks".

No, they didn't. That was a quote from The Inquirer in an article written by someone who is not a native English speaker. God alone knows what he meant, or whether what he meant bore any resemblance to what his source originally said, or whether his source was actually telling the truth anyway. or, indeed, whether there really was a source at all.

Originally posted by: terentenet
ATI's specs mentioned about 64 unified shaders with the capability to process 4 instructions/clock. That sounds like the old Pentium 4 HyperThreading. You only got benefit from it if the application was written for it. Otherwise, you got worse performance with HT ON. .

No. Absolutely nothing like hyperthreading: this is an almost entirely invalid analogy. I say "almost" - there is a small element of truth in that it is difficult to keep vec4 shaders busy all the time. It's easier to keep scalar shaders (such as are found in G80) busy.

G80 shaders are also unified, incidentally - that has nothing to do with whether they are vector or scalar. "Unified" means that there is no longer a separation between pixel shaders and vertex shaders.

If R600 has 64 Vec4 shaders clocked at 750Mhz or so, that would put the performance close to G80. G80 has twice has many shaders running at roughly double the clock-speed, but each of R600's shaders can do four times as much work in one clock cycle. It would probably give G80 a small edge, because (as I said) keeping scalar shaders busy is easier.

More recent rumours suggest that R600 will have 128 shaders - however, they don't specify vector or scalar.

Originally posted by: terentenet
When did we ever see something like this from Nvidia? The GT(S) version was allways lower clocked and had less shaders than GTX. With a GTS having the same number of shaders and just a lower clock speed, it would mean that no GTX will sell. Everybody will get the GTS and overclock it to GTX level.

Well, at least some of those figures are clearly b*llocks, but exactly which ones are and which ones aren't is anybody's guess. It clearly makes no sense at all to have a 320-bit memory bus on the GTS (as compared with 384-bits on the GTX) but the same number of shaders on both.

 

[nrb]

Originally posted by: jdoggg12
Is it so you can throw a guess out there and say ' i told you so ' if your speculation hits near the truth some time down the road?

Some of us just have a low tolerance for speculation that is quite so obviously wrong.

 

[Matthias99]

Just because they (presumably) had been testing production silicon and decided to schedule the launch doesn't mean that all possible usage scenarios had been tested. I could see the situation unfolding something like this: ATI/AMD get production silicon to their testing centers and begin polishing drivers and testing popular software. Weeks go by without issue, the chip seems to be performing well, and driver development is on course, so they schedule launch. A few more weeks go by and testing continues. One day someone loads up a new (or old) piece of software to further test compatibility, a specific feature or even a specific piece of shader code and *whoops* it crashes. Or it performs WAY under expectations. Revised drivers don't fix the problem. Uh oh, time to look at the hardware. Small bug (or large bug) is uncovered in the hardware and it's simply impossible to fix the design or code the drivers around it in time for launch.

Also, with the speed at which this kind of hardware is being developed and rolled out -- if they've been doing multiple respins, it's very likely that nobody has had final "production" silicon for more than a few weeks. It's not like they have been sitting on the finished product for six months, and just now found some absolutely fatal flaw with it.

 

[SickBeast]

Originally posted by: jdoggg12
Im curious... why are some of you so passionate about speculating why a card is delayed? Get out and enjoy some sunshine or something! It's just a video card, these ridiculously long explanations of what 'could' be going on simply amaze me. I come to threads like this hoping to see some info backed by facts... but the 50 or so pages about the R600 in various threads offers nothing substantial.

Is it so you can throw a guess out there and say ' i told you so ' if your speculation hits near the truth some time down the road?

We don't have much 'sunshine' in Canada ATM.

Being correct publicly on a hunch *is* part of the reason for this thread's existance. Life is sometimes about pride and glory.

I agree with you on the lack of info. Hopefully this thread will help with that. :thumbsup:

 

[SickBeast]

Originally posted by: nrb

Originally posted by: apoppin
sure ... they do extensive testing

and expect to fix some bugs as they progress

it appears they found something *unexpected*
--or they would NOT have scheduled the launch in the first place

Precisely. Which means it cannot possibly have been the result of generally poor performance. If the problem is "the chip is too slow compared to G80" they would have known about that long before they booked the flights to Amsterdam: any testing at all would have revealed that.

The fact that the launch was cancelled at the last minute means that it must have been a bug: most likely a hardware bug, possibly a problem with the card design. It must be something that doesn't go wrong often (so it slipped past the testers) but, when it does go wrong, goes wrong so catastrophically that the product simply cannot be released until the problem is rectified.

It cannot be a general performance problem. (It is conceivable, of course, that the bug is performance-related, but only manifests itself in very specific circumstances).

I'm starting to think that the card somehow catastrophically melts down during prolonged heavy usage. Perhaps all that power going to such a massive core generates so much heat that it causes unforeseen problems. There's also the 512-bit memory interface that could be causing issues. GDDR4 too. There's so much that's new on the card.

 

[SickBeast]

Originally posted by: apoppin
maybe i should start another thread giving the *real reasons*



:roll:

:thumbsup:

:beer:

:Q

 

[Greenman]

Originally posted by: jdoggg12
Im curious... why are some of you so passionate about speculating why a card is delayed? Get out and enjoy some sunshine or something! It's just a video card, these ridiculously long explanations of what 'could' be going on simply amaze me. I come to threads like this hoping to see some info backed by facts... but the 50 or so pages about the R600 in various threads offers nothing substantial.

Is it so you can throw a guess out there and say ' i told you so ' if your speculation hits near the truth some time down the road?

You assume too much. I don't buy high end video cards as I have other things to waste my money on, so there isn't a chance in he1l I'll be buying an R600, nor will I buy an 8800.
I'm just a bit let down that AMD/ATI seems to be on the wrong track. And when I say a bit let down, I mean about as let down as I'd be if I walked outside and noticed one of the tires on my truck needed air.

 

[thilanliyan]

Originally posted by: SickBeast
So: R600 has been in the making for probably 4-5 years, which is longer than the 2-3 year development cycle of the G80.

I thought I read before that G80 has been in development for 4 years.
Here's a link (this is not where I read it first by the way):
http://tweaksrus.com/forums/lofiversion/index.php/t6949.html

 

[lopri]

Where are you guys seeing this 25% more shader processors from the leaked 8900 specs?

All I'm seeing from 8900 is a die-shrink (therefore clock speed boost) and GDDR4. There is no difference from architectural specifications. Basically this is very similary to 7800 GTX (or even 7800 GTX 512) -> 7900 GTX refresh. And GX2 following shortly thereafter.

 

[Matt2]

Yeah I dont know where they are getting 25% more shaders from either. Last I checked 8800GTX has 128 scalar shaders, ditto for 8900 via leaked specs.

Maybe they meant 25% faster per clock?

 

[dreddfunk]

I'm not really overly passionate about this issue. ATI/AMD know what they're doing, or they don't. They'll deliver a competitive product, or they won't. I don't turn my hardware over fast enough to care about it one way or another. If they go completely belly-up then I would be worried--but only in the short run. Someone would eventually give nVidia a run for their money.

I'm more bemused and amused by the idea that people actually believe that their own speculation is the only *logical* conclusion.

A lot of philosophers on this board, and not many scientists.

Ironic and somewhat odd given the topic.

 

[imported_thefonz]

me thinks they want to release, new cpus, chip sets, and vid cards at same time,

thats all the major($) silicon in a computer comming from AMD

as long as it is competitive with the current nvidia/intel solution, even 5-10% better performance would make me buy their platform, gotta upgrade the whole comp too.

 

[5150Joker]

Although this R600 delay doesn't affect me since I'm waiting on DX10 games, I've had to recommend the 8800 GTX to a family member and friend that are in the process of building a new PC. AMD needs to get their act together and get the R600 out by May.

 

[DeathReborn]

Originally posted by: lopri
Where are you guys seeing this 25% more shader processors from the leaked 8900 specs?

All I'm seeing from 8900 is a die-shrink (therefore clock speed boost) and GDDR4. There is no difference from architectural specifications. Basically this is very similary to 7800 GTX (or even 7800 GTX 512) -> 7900 GTX refresh. And GX2 following shortly thereafter.

There was a "leak/rumour" saying 160 Shaders in 8900GTX. Technically, nVidia are capable of doing it but it's very unlikely to happen.