Next NV High-end

The article that sparked this whole "G72 = R520 killer" line doesn't explicitly state anything of the sort, and in my opinion meshes pretty well with the previous G72-as-NV44-replacement rumour.
 
Another hint towards 512mb 7800 'round Nov. 1. Mem clock will be interesting, as will gpu clock if they decide to bump a bit, or mucho. I would think this would be their shot at an Ultra-like board. I can't imagine they'd do a 512mb release and then another on top of it.

http://www.theinquirer.net/?article=26819
 
geo said:
Another hint towards 512mb 7800 'round Nov. 1. Mem clock will be interesting, as will gpu clock if they decide to bump a bit, or mucho. I would think this would be their shot at an Ultra-like board. I can't imagine they'd do a 512mb release and then another on top of it.
Well, it is unknown how ATi plans to attack the 512MB category, is it not? At first I had the impression that ATi wanted to standardize this memory capacity, but now I am not so sure. I conclude this from the fact that the XLs that were released and benchmarked were of the 256MB variety and not 512MB as hinted at before release. That is not to say that there won't be a 512MB XL, but surely ATi would have submitted 512MB XLs for benchmarking if it was planned as the standard SKU.

The fact that the XT used for reviews was 512MB has not convinced me this will be the prevalent flavor of that particular card either, but maybe another case of "Press Edition". I say this because the XL is supposed to fill out the high end segment and you would really need a standard amount of VRAM throughout that whole segment to make it stick (a la 6800 Ultra, GT and Radeon X800 XT and XL; all at 256MB standard). This has me thinking the 512MB level was launched half-heartedly by ATi.

This also makes me think that Nvidia may simply announce a 512MB version of the GTX, leaving other improvements for another day. A theoretical 7800 Ultra may be price restained if it were only launched in 512MB and if ATi is not going to play that game I don't think Nvidia will either. Looks like it will be a very messy 256MB v 512MB transition, but certainly with more units than the 512MB versions of previous cards.

Those were just some thoughts and I don't like them very much. I was very excited when I read the rumored 512MB standard for the new Radeon lineup as that would have firmly planted a foot in the direction of 1GB VRAM...yum! (512MB will be fun, but it is my personal belief that the next REAL step in 3D graphics won't happen until we are at 1GB and 512MB is a necessary stepping stone...)
 
Assuming the differential roughly holds at Street, I think all you need to know about whether ATI intend the XT 512mb to be the prevalent flava of that card is that there is only a $50 price difference vs 256mb. How many people at this level are going to save that $50 and give up 256mb of high-speed memory? Damn few approaching zero, I think.
 
geo said:
Assuming the differential roughly holds at Street, I think all you need to know about whether ATI intend the XT 512mb to be the prevalent flava of that card is that there is only a $50 price difference vs 256mb. How many people at this level are going to save that $50 and give up 256mb of high-speed memory? Damn few approaching zero, I think.
That, I would agee with, but it doesn't solve the XL conundrum. A single SKU with 512MB is not going to bring about the "512MB Revolution," that's my point. Should we expect a similar $50 differential on a 512MB XL?

I'm just saying that the whole image of a bringing the whole high-end up to 512MB, which I think is necessary for a real impact other than benchmarking advatnages, crumbled when the XL was announced as 256MB. I suppose the "theory" that the R520 XT will be pushed down soon to be replaced by the R580 XT at the top could help this by pushing the XL out of the high-end segment and into the middle, but still...I dunno. It doesn't look like ATi took that decisive step over the 256 line and, therefore, I don't want to assume Nvidia will either.
 
http://www.nvnews.net/vbulletin/showpost.php?p=717357&postcount=23 translated from http://www.hkepc.com/bbs/viewthread.php?tid=486092

According to Samsung's disclosed order manifest, apart from ATi, nVidia has also placed order for the 1.26ns GDDR3 memory chips, therefore we expect to see nVidia release its new 1.26ns GDDR3 featuring product in the near futre, which is believed to be the weapon against the X1800XT. The size of the Samsung K4J52324QC-BJ12, which is currently used on the ATi Radeon X1800XT, is different from the more commonly used GDDR3 memory we've seen, because the packaging used has been changed to 144 Ball FBGA unlead from 136 Ball FBGA, the specs are 2M x 32Bit x 8 Bank,voltage 2.0V. BJ12 means the speed is 1.26ns,the highest Samsung official clockspeed is 1.6GHz, much higher than the 1.6ns on the Geforce 7800GTX. If this memory is to be used, the PCB will have to be modified.

In additon,there appears to be a new G70 definition in the recently released Forceware 81.84 beta driver. This could well be the 1.26ns GDDR3 Geforce 7 part. If the 1.26GDDR3 is used along with an increase in core clockspeed, it's not difficult to surpass the ATi Radeon X1800XT in performance.
 
Ah ha! Interesting times ahead. :D

I begin to suspect that from a performance aspect, that X1800XT is going to look merely "okay" in retrospect a little down the line, unless they pull a signficant increase out of the drivers in optimizing the memory bus (which certainly may happen). I like the features a lot, but I look at how much space that whatever-it-is in the left-middle of the die is taking for "infrastructure" and I come to the conclusion that ATI is, to a degree, "taking their medicine" on architecture shift on this part, and reaping the rewards later down the line as that big sucker gets relatively smaller from a percentage pov on later parts.

Which is the long way around to "I really want R580", but that really isn't news at this point, is it? :LOL:
 
geo said:
I begin to suspect that from a performance aspect, that X1800XT is going to look merely "okay" in retrospect a little down the line
Begin? Heh. And yes, R580 does seem quite kickass, although remember that there are a number of things it just doesn't improve. Things that NVIDIA will improve in its G70 refresh. So a clear superiorityis unlikely. What matters now, financially-speaking, is mindshare. And margins. Anything else is just "details" nowadays. Let the fans and OEMs speak.

Uttar
 
Last edited by a moderator:
Uttar said:
, although remember that there are a number of things it just doesn't improve. Things that NVIDIA will improve in its G70 refresh.

Oh, do tell! ;)

So a clear superiorityis unlikely. What matters now, financially-speaking, is mindshare. And margins. Anything else is just "details" nowadays. Let the fans and OEMs speak.

Sure. Tho I think ATI did enuf to skinny over the bar for now. Tho some of the OEM stuff has yet to shake out from the previous delay rather than from here forward. It's a bit of a trailing indicator, yes? On margins, I'm thinking it is about a wash --ATI gets 90nm advantages from now, but NV had pricing power for several months with which to squeeze blood out of the rock (that would be us. :LOL: ).
 
geo said:
Ah ha! Interesting times ahead. :D
I did start a thread on this subject:

http://www.beyond3d.com/forum/showthread.php?t=23738

And it turns out that the 7800GTX uses 144-ball memory chips and the new memory chips are 136-ball. So the boards need to be re-designed (I suspected as much when I started the thread, though I was a bit wobbly on the technical details...)

It turns out the HKEPC article has got 136- and 144-ball the wrong way round... Or at least the translation has.

I begin to suspect that from a performance aspect, that X1800XT is going to look merely "okay" in retrospect a little down the line, unless they pull a signficant increase out of the drivers in optimizing the memory bus (which certainly may happen).
It seems all three, RV515, RV530 and R520 are going to look like "version 1.0", which you don't buy because v1.1, which'll be along shortly, will make them look a bit silly.

At least don't pay MSRP for them, for goodness sake.

I'd love to be optimistic about there being 20%+ latent driver performance increases in games across the board with these cards, but I just don't see it happening. [OT ramble deleted]

Which is the long way around to "I really want R580", but that really isn't news at this point, is it? :LOL:
And RV540, RV560.

Jawed
 
Jawed said:

I saw it. But now we have reports of actual NV orders. :smile:


I'd love to be optimistic about there being 20%+ latent driver performance increases in games across the board with these cards, but I just don't see it happening. [OT ramble deleted]

I would like to think I'm not being hopelessly pollyanna-ish about this, but events certainly may prove you correct. I'm relying on history here. . .in this case, NV's history --the signficant mid-life boost they got from their last major memory controller upgrade after they learned how to tweak it.
 
trinibwoy said:
I had the impression that the "bugged' libraries were something that ATi licensed from a third party for their own design. No where did I see it being referred to as a TSMC problem.

I would not want to put too much weight on this, but Orton referred to is as a problem between engineering and the Fab. . .I wonder if these guys go out for their own libraries, or license them thru the Fab. . .
 
nagus said:
http://www.digitimes.com/news/a20051005A7033.html

Nvidia schedules G72 GPU release in early 2006
Charles Chou, Taipei; Jessie Shen, DigiTimes.com [Wednesday 5 October 2005]

Nvidia has scheduled its G72 graphics processing unit (GPU), which will be manufactured using 90nm process technology, to be introduced in early 2006, according to sources at Taiwan graphics card makers. The G72, a 90nm version of the G70 GPU, will compete with the 90nm R-series from ATI Technologies, which will announce the launch of its long-awaited R520, RV515 and RV530 GPUs on October 5, U.S. time, indicated the sources.

The 90nm G72 will have a much smaller size than the 0.11-micron based G70 allowing for multiple GPUs to be utilized on one graphics card through SLI technology, providing an effective and attractive alternative to those who do not want to fork out money for two graphics cards, the sources claimed.

Graphics card makers are ready to ship R520- and RV515-based products, following ATI’s official release, according to sources at the makers. RV530-based graphics cards will be volume shipped at the end of this month, indicated the sources.

Damn, I must have been in full R520 orgy mode and this went right past me that it was Digitimes (and, thus, "respect must be paid").

Oh woe. "G72". Ick.

What about that SLI thingie? Back around G70 release, you may recall, there was much hinting (and misuse of terminology before someone slams me --not that it will stop them ;) ) at "dual-core" from NV, and "almost like 512-bit bus" kind of comments with a wink from a few in-the-know. I think even Tom's hinted in that direction (Oh, Lars. . .where are you now? :p ).

Could we be on the verge of seeing an official SLI-on-one-board reference model from NV and sold as a regular model across the full line of AIB instead of just the occasional AIB-on-a-lark thing?
 
geo said:
Damn, I must have been in full R520 orgy mode and this went right past me that it was Digitimes (and, thus, "respect must be paid").

Oh woe. "G72". Ick.

What about that SLI thingie? Back around G70 release, you may recall, there was much hinting (and misuse of terminology before someone slams me --not that it will stop them ;) ) at "dual-core" from NV, and "almost like 512-bit bus" kind of comments with a wink from a few in-the-know. I think even Tom's hinted in that direction (Oh, Lars. . .where are you now? :p ).

Could we be on the verge of seeing an official SLI-on-one-board reference model from NV and sold as a regular model across the full line of AIB instead of just the occasional AIB-on-a-lark thing?

I still think that G72 is a lower-end part, meant to compete against the X1300 series, atleast from what I know. ;)
 
I know, I know. . .he's been known to "learn better" too on occasion! I hate them calling it "G72" anyway.
 
Fodder said:
That article doesn't say it's a high end part, so they may indeed be referring to G72 as we know it.

You are being even more Jesuitical than I can be at times. It says twice it's the 90nm version of G70.
 
geo said:
You are being even more Jesuitical than I can be at times. It says twice it's the 90nm version of G70.

It also suggests that it will compete with the R-series as in "X1800, X1600 or X1300" also. ;)

Nvidia has scheduled its G72 graphics processing unit (GPU), which will be manufactured using 90nm process technology, to be introduced in early 2006, according to sources at Taiwan graphics card makers. The G72, a 90nm version of the G70 GPU, will compete with the 90nm R-series from ATI Technologies, which will announce the launch of its long-awaited R520, RV515 and RV530 GPUs on October 5, U.S. time, indicated the sources.
I know I being mr. skeptical here but I dont want to get any hopes up. :| Also Digitimes has been wrong before on multiple occasions. Remember they were also propagating the 24 pp R520 theory. ;)
 
Back
Top