ATi's roadmap for 2004

[John Reynolds]

What's amusing is that this is just rumors. If Anand truly did know R420 specs he'd clearly be violating his NDA and ATI could then take legal action against him. But since it's being posted on a major hardware site, people take it as legit without thinking it through.

And, IMO, it's very unprofessional of Anand to post these rumors.

 

[vb]

so, if this is true, how will six units work on a quad?

Easy. Vertex shaders don't work on quads, they work on vertices (the points defining the shape of a polygon).

Well VS are programs (you know, bytes and stuff aka software)

I'm talking about VS units (hardware stuff, I could draw the difference but you might be under age)

Anyway, I'm starting to think that ATI maybe mentioned VS units 50% more powerful per clock. IMHO.

 

[Luminescent]

What if Ati improved their dual core pixel pipeline implementation (even R350 does not achieve its max theoretical fillrate) in addition to allowing usage of vertex pipelines for pixel shaders ops?

 

[Guden Oden]

Well VS are programs (you know, bytes and stuff aka software)

Uh, yes? So what?

I'm talking about VS units (hardware stuff, I could draw the difference but you might be under age)

Your point being...?

Six vertex shader (units) or three or four or whatever doesn't matter. Each unit works on one vertice at a time. Remember, early DX8 hardware had just one unit, and that was clearly not a problem.

Anyway, lay off the condescending attitude, that's completely uncalled for.

 

[Reverend]

... it's very unprofessional of Anand to post these rumors.

Actually, I find nothing wrong with "expression of speculation". I only think it's "unprofessional" if such expressions of speculations come out as statements of fact. Which is the way I see it in the article. Which I don't particularly fancy. Even if the speculations are true.

I'm talking about VS units (hardware stuff, I could draw the difference but you might be under age)

Your post count suggests you're relatively new to our forums. I'd appreciate it if you could please learn some manners.

 

[Hanners]

Actually, I find nothing wrong with "expression of speculation". I only think it's "unprofessional" if such expressions of speculations come out as statements of fact. Which is the way I see it in the article. Which I don't particularly fancy. Even if the speculations are true.

Exactly - I was none too keen on the 'this is the way it will be' wording of the article either.

 

[Dave Baumann]

I can guess with a reasonable level of certainty that this will have come from a vendor.

 

[digitalwanderer]

I can guess with a reasonable level of certainty that this will have come from a vendor.

Does that make it legit or FUD though? I'm out of shape at this game...

 

[Xmas]

"will have come"?

 

[digitalwanderer]

"will have come"?

Yeah, Dave's got a time machine and sometimes slips up on his tenses....how else do you think he knows the things he does? :|

 

[Pete]

"will have come"?

"The drinks are on the roof."

 

[WaltC]

The only interesting part from anand's was this:


This could be a smart way of pushing PCIe.

later,
epic

Yes, and expect to see it pushed for all it's worth... It's no small investment to do separate PCBs for the two buses in retail packages, especially when the demand for AGP will probably be 5x-10x greater initally, I would imagine. I would also imagine that ROI is why nVidia's reportedly doing only a bridge initially for PCIe support. A lot of glitzy, hard-hitting promotional effort for PCIe may help in the beginning, and so I expect to see a lot of it, as it will benefit sales for everybody doing something with PCIe, not just 3d IHVs. Plus ATi will benefit peripherally from the backwash of all of the hype that will be generated by Intel and various motherboard OEMs, who will all be pushing PCIe hard this year--probably moreso than nVidia will benefit in that regard if it's true that nVidia's initial efforts for PCIe will be bridge-only. It may well be a wash, really, so nVidia may be a bit too conservative here--we'll just have to see. No question though that PCIe should be be a big boon for integrated 3d solutions, however.

I saw something I thought was kind of bizarre recently: Chaintech reportedly is previewing samples of an upcoming board which will include both an AGPx8 slot & a PCIex16 slot, physically... That's what I read somewhere, and so I don't know if it's true. But if it is, it's going even further than I thought with bus duplication, as I had expected the first PCIe boards to also sport PCI slots--I expect that will be the case for the next couple of years. But I had thought that such boards would offer *either* a PCIex16 or an AGP slot, so it was surprising to see both in a board.

Considering that AGP hangs off the PCI bus anyway, and all of these boards will include PCI in addition to PCIe, and the slot itself isn't that costly in a retail mboard--I guess it's not all that surprising--and this may turn out to be the rule rather than the exception for awhile. This approach would certainly mean that a mboard OEM wouldn't have to keep separate inventories of PCIe vs. AGPx8 as they'd both be included in every board. Hmmm...

 

[Psikotiko]

I saw something I thought was kind of bizarre recently: Chaintech reportedly is previewing samples of an upcoming board which will include both an AGPx8 slot & a PCIex16 slot, physically... That's what I read somewhere, and so I don't know if it's true. But if it is, it's going even further than I thought with bus duplication, as I had expected the first PCIe boards to also sport PCI slots--I expect that will be the case for the next couple of years. But I had thought that such boards would offer *either* a PCIex16 or an AGP slot, so it was surprising to see both in a board.

http://www.hardocp.com/images/news/1076508119iFR0eEdSdP_1_2_l.jpg

 

[Rugor]

Yes, "will have come" a perfectly reasonable phrasing referring to something that we will discover in the future to have happened in the past. It's because the source is something that will never be in the present tense for us. The discovery of the source is in our future, but the release of the info is something that we will discover to be a past event.

Dave's a very good writer.

As a complete aside, six hardware VS engines does imply two triangles, assuming each shader works on a single vertex at a time.

 

[WaltC]

What's amusing is that this is just rumors. If Anand truly did know R420 specs he'd clearly be violating his NDA and ATI could then take legal action against him. But since it's being posted on a major hardware site, people take it as legit without thinking it through.

And, IMO, it's very unprofessional of Anand to post these rumors.

The only positive thing I saw was his remark that AGPx8 3d cards would be sold all year long in addition to PCIex16 3d cards--as the last so-called "PCIex16" rumor mongering (in the form of an unattributed "official" article on another web site) I read stated flatly that after Q2 '04 nobody would be able to buy a high-end AGP x8 3d card any longer. I did appreciate Anand weighing with that molecule of common sense, as there seems to be so little of it around these days...

I think Anand's problem with 3d is the same as it's always been: he's not terribly swift on the basic concepts involved, and so when he says in his text that the "pixel fill rate is double that of R350" he doesn't understand how that comment misaligns with his table stats which show that the R350/60 & his hypothetical R4xx both share 8 pixel pipelines. Or, it could be that it's his multiplication that's weak and so he doesn't see that ~500MHz does not equal 2x380 MHz, or something similar. Curiously absent is any mention of texturing units that I could see, as well as any mention of the sources for his comments.

"Coming across a roadmap," could mean any number of things--like finding it crumpled at the bottom of a trash can in the restroom of the local Shell petrol station while hunting for change to access the prophylactic dispenser, to receiving it embedded in an anonymous email from "SantaClaus," to actually getting the information from a contact within ATi. But if he doesn't tell his readers how it was that he "came across" the info--then any of those scenarios is as likely as the other, isn't it?...

My sincere hope is that now that Anand has been admitted to a college program, that he will avail himself of some basic journalism courses in which he will discover the word "attribution" and divine how that might apply to reporting the things he represents as facts on his web site. I hope he will discover that phrases like "a little bird told me," and "from what I hear," and "we stumbled/came across the following info" are not valid forms of journalistic attribution.

 

[Xmas]

Yes, "will have come" a perfectly reasonable phrasing referring to something that we will discover in the future to have happened in the past. It's because the source is something that will never be in the present tense for us. The discovery of the source is in our future, but the release of the info is something that we will discover to be a past event.

Dave's a very good writer.

As a complete aside, six hardware VS engines does imply two triangles, assuming each shader works on a single vertex at a time.

To me "will have come" sounds more like "in the future someone will claim it came from a vendor"...

btw, six VS engines does imply nothing at all wrt triangles.

 

[WaltC]

http://www.hardocp.com/images/news/1076508119iFR0eEdSdP_1_2_l.jpg

Thanks! Couldn't recall whether I'd imagined it or what...

 

[Catcher]

Something disturbing about this release, assuming that the information is vaguely correct, is how ATI seems to be "ghettoizing" the value conscious. It began with a limited 9500 Pro run, then a "hard floor" on 9700s, then 9800NPs disappeared rather quickly, then 9600 XTs only had the core speed bumped, then the next generation of mid-range cards will be PCIe only and STILL just a speed-bumped 9600 Pro (grain of salt excepted, Dave ). Anyone see a pattern here? It seems that ATI is determined to enforce a vast gulf between high-end and mid-range. :? That's really bad since either a .13 micron Low-k card with either more pipes or a wider memory bus would be less expensive than the 9500 Pro and still be an appropriate card for medium-level gaming. I'll hold on to my cash and my salt for now, but doesn't anyone want the "bang-for-the-buck" gamers' money?

 

[WaltC]

Something disturbing about this release, assuming that the information is vaguely correct, is how ATI seems to be "ghettoizing" the value conscious. It began with a limited 9500 Pro run, then a "hard floor" on 9700s, then 9800NPs disappeared rather quickly, then 9600 XTs only had the core speed bumped, then the next generation of mid-range cards will be PCIe only and STILL just a speed-bumped 9600 Pro (grain of salt excepted, Dave ). Anyone see a pattern here? It seems that ATI is determined to enforce a vast gulf between high-end and mid-range. :? That's really bad since either a .13 micron Low-k card with either more pipes or a wider memory bus would be less expensive than the 9500 Pro and still be an appropriate card for medium-level gaming. I'll hold on to my cash and my salt for now, but doesn't anyone want the "bang-for-the-buck" gamers' money?

Just to play devil's advocate, it's always seemed to me that in relation to "bang for the buck" in 3d the high end is the place to be... As well, everybody wants a Rolls/Jag/Mercedes for the price of a Ford or Chevy, don't they?

Contrast the ~$200 9600XT with the ~$350-$399 9800P--you get better than 2x as much with the 9800P, aside from clockrate, of all of the things that matter much more than clockrate in 3d, it seems to me. Or contrast the ~ $449-499 9800XT with the $350-$399 9800P, and the extra $100 buys you a faster core clock & 2x the amount of onboard ram. But contrast the 9600XT with the 9800XT and it seems to me that "bang for the buck" is pretty darn close in both cases.

I think the common mistake is that most people think "bang for the buck" goes out of the window on the high end, but it doesn't--bang per buck should be fairly consistent, it's just the amount of bucks, and therefore bang, that differs. As long as they stick to the rule of "you get what you pay for" I'm happy...

 

[DerekBaker]

"will have come" sounds just right to me.

Both Dave and I are in the UK.

Are the people to whom it sounds incorrect from outside the UK?



Derek