The LAST R600 Rumours & Speculation Thread

[Jawed]

It's entirely possible that demand was significantly higher than AMD was expecting. If Dell placed a large order they very well could be struggling to get enough chips for the hard launch which they promised.

There's a degree of plausibility here I guess.

The 12" card is OEM only, and there's been talk that amongst others it's for Apple. Has Apple got any new models coming in spring?

Dell might be a factor too. NVidia is incumbent with 8800GTX at the high-end of Dell's range and it seems unlikely to me that NVidia's lost that, since the non-WHQL'd Vista charade is over: so theoretically the pressure's off NVidia now.

I suppose there's a slim chance Dell got so cheesed-off at the whole situation that AMD got a look-in...

But, overall, the "out of the blueness" of this reeks of damage-limitation rather than "being swamped by demand". If the demand story is there, surely there'd be a way to spin that into the story in a gently allusive manner.

Jawed

 

[Shtal]

Did you see anyone claim so far that R600 will be a power saving wonder? Au contraire all speculative newsblurbs so far suggest a higher power consumption than G80.

I never said it will be power saving card, you did not understood me.
When I said that If R580 had 24 ROP's & 24 texture units, you reply that it will need more transistors and will draw more power, which I agree, all I said that R600 will be even greater of the transistors and power requirement....

 

[trinibwoy]

But, overall, the "out of the blueness" of this reeks of damage-limitation rather than "being swamped by demand". If the demand story is there, surely there'd be a way to spin that into the story in a gently allusive manner.

Yeah, I'm taking a pretty cynical view of some of the positive spin. There is no way that IHV's move high-end parts at launch (or ever?) through the OEM channel faster than at retail so that explanation holds zero probability IMO. Why is it so hard to accept that something obviously upset AMD's launch plans and that that something was most definitely unwelcome. There is no possible gain to be had by not selling a product. The CPU thing makes a little sense - they wouldn't want R600's awsomeness to benefit Intel's Core2 sales but that is a consequence they would have included in their strategy a long time ago - it's not like they woke up yesterday and realized that Intel has the enthusiast market on lock.

 

[Jawed]

There is no way that IHV's move high-end parts at launch (or ever?) through the OEM channel faster than at retail so that explanation holds zero probability IMO.

Wasn't that the fate of X800XTPE? Grandpa Joe bought the top-end Dell back in the summer of 2004 and found it equipped with an X800XTPE?

Jawed

 

[trinibwoy]

No that's what we were told in order to whitewash the poor availability. Until I see actual numbers I'm not going to buy that there are more joe blows willing to purchase a whole new high-end machine than there are gamers/enthusiasts willing to upgrade their existing machines with a new high-end card.

 

[Jawed]

No that's what we were told

It happened, there was no "telling" involved.

People were buying the Dell, stripping out the X800XTPE and selling it on eBay.

Jawed

 

[trinibwoy]

How does that anecdotal evidence indicate in any way that ATi sold more through Dell than they would have at retail? It's a simple bundling scenario - there will always be more people willing to buy the cheaper unbundled product (card) than the full shebang (card+machine). That's why there's bundling!!

 

[Voltron]

http://www.extremetech.com/article2/0,1697,2097750,00.asp

Perhaps the most reputable confirmation yet, along with a tidbit of speculation. Not that there was any doubt of the delay. Speculation of a respin for some reason seems fairly likely to me. For one, silicon that was hoped to be production silicon would probably just be starting to come back from fab about now. Yields or clocks maybe didn't pan out? They have had a long a long time to work on boards and if memory was a problem for the time being you think they would still go ahead with editors day. Canceling editors day is sign that it wont be released for a few more months, at least, which is the length of a respin.

I know Silent Guy wants to talk about this!!

 

[Jawed]

How does that anecdotal evidence indicate in any way that ATi sold more through Dell than they would have at retail? It's a simple bundling scenario - there will always be more people willing to buy the cheaper unbundled product (card) than the full shebang (card+machine). That's why there's bundling!!

What was pissing people off back then was Dell PCs with XTPEs were shipping while regular sites had no stock.

I've just spent an hour searching Rage3D, but the search over there is awful, so I haven't found the evidence

Now that the hour's up and the torrent's in, it's bedtime...

Jawed

 

[Anarchist4000]

[If I remember correctly its somewhere in the order of 12 weeks from production to start to completed wafer in your hands.

I think that's the time from when TSMC first receives the design. Assuming they already started production and were just increasing the order I would think the turnaround would be considerably faster.

Dell might be a factor too. NVidia is incumbent with 8800GTX at the high-end of Dell's range and it seems unlikely to me that NVidia's lost that, since the non-WHQL'd Vista charade is over: so theoretically the pressure's off NVidia now.

I suppose there's a slim chance Dell got so cheesed-off at the whole situation that AMD got a look-in...

The only Dell model I've seen that actually offers an 8800 is the H2C with extreme cooling or something along those lines. The XPS line with quad core intel chips offer 1950 Crossfire or 7950GX2 SLI options.

Here's another interesting comment from their site.

Windows VistaTM Coming Soon
While Dell continues to validate the compatibility of Windows Vista with popular gaming software & accessories, we will continue to offer XPS 710 customers Windows® XP operating systems.

That tells me they're wanting a DX10 part with good quality drivers and haven't found one yet. ATI drivers are arguably good enough for running DX9 on Vista while Nvidia had some early problems. 1950s and 7950gx2s should be pleanty fast for DX9 acceleration on Vista. Which leads me to believe they're wanting DX10 and aren't making 8800s available for whatever reason, probably drivers or stability. I mean seriously, PhysX cards, quad cores, 2TB drive setups and no 8800's?

I'm going off the assumption that people don't know what they're buying. They go to Dell and order the best gaming machine they can come up with or their parents place the order. Did we ever have any breakdown of high end sales for OEM vs retail? That might be an interesting statistic to see. Plus with the introduction of Vista OEMs will likely be looking at more powerful GPUs to run the systems. I would have thought it would be a shift from low to mid based volumes and not necessarily affect the high end as much. Especially if there is a growing market of HD where more powerful GPUs are required to properly accelerate it.

I would definitely agree that something stinks with them calling off editor's day but with the Vista release, Nvidia driver problems, and new architecture I wouldn't call this a typical release. It could likely be a combination of many of them. But like mentioned before, we haven't heard any yield issues, the power envelope is fairly close to that of G80 and it's using a smaller process. For performance to be worse with more power, on a smaller process, and with possibly twice the potential bandwidth(1.4GHz GDDR4), they'd need to have some really horrendous clockspeed issues. And if clockspeed is the issue I think you can safely throw Vec style ALUs out the door as well as start thinking about a 65nm process being used. If they used Vec style ALUs I don't know how they would take an architecture similar to what they've used in the past and mess it up so badly.

Unless of course someone forgot to connect the pins again or the out of control annual TSMC frisbee fight someone else mentioned occurred.

 

[swaaye]

http://www.extremetech.com/article2/0,1697,2097750,00.asp

Perhaps the most reputable confirmation yet, along with a tidbit of speculation. Not that there was any doubt of the delay. Speculation of a respin for some reason seems fairly likely to me. For one, silicon that was hoped to be production silicon would probably just be starting to come back from fab about now. Yields or clocks maybe didn't pan out? They have had a long a long time to work on boards and if memory was a problem for the time being you think they would still go ahead with editors day. Canceling editors day is sign that it wont be released for a few more months, at least, which is the length of a respin.

I know Silent Guy wants to talk about this!!

Well, that's some scary news. There's no reason to not believe Loyd Case. Another respin? Talk about deja vu.... Here we come October. heh...

Though, "...have required another respin" sounds like it has already happened and is just taking longer than they expected. Perhaps...

 

[Thorburn]

That tells me they're wanting a DX10 part with good quality drivers and haven't found one yet. ATI drivers are arguably good enough for running DX9 on Vista while Nvidia had some early problems. 1950s and 7950gx2s should be pleanty fast for DX9 acceleration on Vista. Which leads me to believe they're wanting DX10 and aren't making 8800s available for whatever reason, probably drivers or stability. I mean seriously, PhysX cards, quad cores, 2TB drive setups and no 8800's?.

The answers in the marketing money.

If you have a nice Microsoft Windows logo on your advert they'll contribute towards a percentage cost of the advert, but to qualify to include this logo you need a system that passes certain criteria, one of them being that all the drivers need WHQL certification.

Since NVIDIA have been late to the party with these for the 8800 cards (I believe there are WHQL 97.xx drivers but obviously these don't support the 8800) it means Dell haven't been able to put the machine forward to qualify for the marketing money.

Ergo keep it with Windows XP and get the advert contribution paid rather than sticking on Vista and have a months worth of adverts solely coming out your own pocket.

 

[silent_guy]

http://www.extremetech.com/article2/0,1697,2097750,00.asp

Perhaps the most reputable confirmation yet, along with a tidbit of speculation. Not that there was any doubt of the delay. Speculation of a respin for some reason seems fairly likely to me. For one, silicon that was hoped to be production silicon would probably just be starting to come back from fab about now. Yields or clocks maybe didn't pan out? They have had a long a long time to work on boards and if memory was a problem for the time being you think they would still go ahead with editors day. Canceling editors day is sign that it wont be released for a few more months, at least, which is the length of a respin.

I know Silent Guy wants to talk about this!!

I'm happy to oblige.

I doubt it would be speed related: real timing bugs are usually found quickly. They don't sneak up on you at the last minute. And when those are solved, there is not too much you can do it increase clock speed without compromising yield.
As for yield: the only fixable factor that can drastically improve yield are bugs in the custom memories. Similar story here: you'd notice that on the first silicon. There are probably secondary effects you can do after that, but I don't think you can expect miracles.

If they find a problem this late, it must be some kind of functional corner case bug that didn't trip until recently. I would assume a lot of testing happens on real games, so there's only DX9 for that. Maybe they ran into some DX10 specific stuff?
Anyway, if it's something that can be fixed with a few gates (as is surprisingly often the case): 1 week to create an ECO. 4 weeks to respin. After that, it's guessing: a whole bunch of requalification (1 weeks seems like a minimum, ours can take months, but we're in a high-reliability sector.) Don't know how long it takes to build up a decent stock of new boards. 3 weeks?
So let's say 2 months in total, starting now. We're looking at mid to end April then.

Since they already had booked everything for the press launch, I can't believe it was a deliberate marketing driven decision. This smells like an OMG NOT THIS event that will make some engineer sleep very badly for a while. (You always feel for them but you're also always happy it wasn't you.)

 

[Geo]

Apples to Oranges. The NV3x was much more difficult to write a shader compiler for. The R300 practically didn't need one, as its performance on unoptimized translation was already much better than the NV30. The NV3x had tons of gotchas where an instruction stream could lead to pathologically bad performance.

Not that I want to bring up NV30 <-> R600 comparisons, because undoubtedly, the R600 won't be fubared performance wise like the NV3x, but from the PR standpoint, I'm noticing a bizarre deja vu here. When NV30 was delayed, oversized and hot, expensive, and arrived as not the performance killer everyone expected, discussion in forums shifted to other "benefits" of NVidia, like UDA, high quality drivers, OpenGL performance, away from the HW itself. Basically, if you can't win on HW, you shift the PR discussion to other issues.

Now that ATI has no new product shipping and it looks much delayed and pricey, and possibly not the world-performance-beater, lo-and-behold, all the discussion is centered around ATI's release of Vista drivers vs NVidia's as if consumers should wait six months for an unknown R600 product because darnit, the R300-R580 Vista drivers were good and NVidia has bugs.

And in the summer of 2005 they talked about a Crossfire product they didn't really have yet, and Catalyst drivers (again) while marking time waiting for R520. Cats seem to be a favorite subject to talk about when they don't have to talk about what they'd really like to talk about.

Surely it's not news that PR orgs will always try to point at their strengths vs the competition at any given point in time? But whos fault is it that pointing at Vista drivers is a competitive advantage for them right now? 8800 has been out for 3.5 months now, never mind the 4 years of development and 3 years they worked on Vista drivers (or whatever it was they said).

Edit: Oh yeah, and I think if you look back to last year you will find plenty of indicators that ATI was putting a lot of effort into Vista drivers and seeing them as a potential competitive advantage. That would be before R600 was late.

 

[Shtal]

That's a side-effect you get with all delays in the GPU market; the longer the delay the higher the frequency increase needed to compete.

Are you saying that it may exactly what happening to R600 ? just like NV30 had to be @ 500MHz clock frequency to compete with R300 @ 325MHz frequency.

 

[Geo]

I think if I was going to make a small bet, it'd be towards A13 being production silicon. The Extremetech report has got a bit of schizophrenia going on. Either that or they're daintily suggesting that AMD/ATI is fibbing.

 

[SugarCoat]

Are you saying that it may exactly what happening to R600 ? just like NV30 had to be @ 500MHz clock frequency to compete with R300 @ 325MHz frequency.

actually, i think he ment because 8800s will continue to mature and with AIBs releasing OC versions by the time the damn thing launches price wise its going to be competing with multiple models, not just ones running at reference, not to mention it wont be long after launch we'll be seeing the 8900 taking form so how long can any lead really last?

 

[ben6]

I think if I was going to make a small bet, it'd be towards A13 being production silicon. The Extremetech report has got a bit of schizophrenia going on. Either that or they're daintily suggesting that AMD/ATI is fibbing.

The first thought when I heard the Editor's Day was cancelled without a hard new date (this is what concerns me most about the cancellation) was exactly what Extremetech thought, a respin. NVIDIA and ATI don't book events inviting mainstream press, finanical analysts, hardware websites, book hotels, airline tickets a day before an event. They plan everything weeks if not a month or more in advance.

 

[icecold1983]

u rly have to take a hard look at management when a company manages to screw up 3 out of their last 4 product launches.

 

[zealotonous]

There could be a number of reasons why they decided to delay R600, I just don't see that they want to align the launch schedule of their high end and mid range parts as one of them. I took a quick look back at some product launches over the last two years and very few of them (if any) featured a simultaneous launch of mid and high end parts. Anyone, please feel free to prove me wrong on this point. From a technical standpoint, i think the R600 had a couple of things going against it from the start.

The R600 was going to be big, hot, and potentially noisy with some extemely high power requirements. This is not meant to be a Nvidia vs ATI post by any stretch, but credit should be given to Nvidia for producing the 8800 gtx and achieving lower total power draw than an ati x1950xtx. i have not heard about any power saving technology in R600 and I expect that it was going to draw substantially more power than both the 8800 and 1950. All of this leads to question the practicality of such a chip in real world use.

If it wasn't power and heat issues, it could have been performance issues. ATI may have looked at the performance compared to the competition and found that it was either lacking or barely up to par . I don't believe ATI would have delayed this product just because it didn't obliterate the 8800. If it had higher or comparable performance than the 8800 without power, heat, and noise issues, I think it would have been released to much fanfare.

If it wasn't power, heat, noise, or performance issues, it could have been availability. ATI took the more aggressive approach and went to DDR4 memory and a 512 bit bus. DDR4 may not be as widely available as DDR3 and the 512 bit bus presents a greater challenge for the AIBs as it is a more complex board design.

If it wasn't any of the above, it could have just been broke. Either way, this surely is not the result of a change in marketing strategy. Changes in time to market only occur when there is little deviation in product roadmap or timeline. I think Orton (I don't think Hector wants to take credit for this) will give us an explanation prior to the next public conference call. If the investors hear they delayed an already late flagship product due to a change in marketing strategy, there may be a bloodbath