The New and Improved "G80 Rumours Thread" *DailyTech specs at #802*

This is basically what nVidia's been saying all along.

 

http://forums.vr-zone.com/showthread.php?t=94319

 

You're giving them a lot more credit than I am I'm just speculating as to where these numbers could have possibly come from. A whisper about a 1.5Ghz memory clock can easily turn into a rumour about a 1.5Ghz core clock as the whisper is passed along.

 

If a current part is doing 1.0Ghz I don't think 1.5Ghz is too unbelievable for the maximum speed of a next-generation GDDR4 memory controller. G8x is going to last quite a while.

It's not really a jump either for GDDR4 itself - 1.25Ghz was done and dusted last year and 1.6Ghz is on the way. http://www.xbitlabs.com/news/memory/display/20060214062714.html

 

In other news, a reliable source told me the ATI lineup will be the X2800XTX/XT/GT, the X2600XT/XL and the X2300XT/Pro.
Look at me, my info is reliable!!1!1!!oneoneone...... sigh.

Uttar

 

I was actually thinking about the first high end release of the G8x family; that 1.25GHz you're mentioning equals roughly >50% more bandwidth than on G71. Cost is always a consideration, as well as keeping reserves for refreshes, when prices there drop even more.

Currently the 64GB/sec of the R580+ plays out mostly in ultra high end resolutions (something like 2048 and beyond) and I doubt that for the next 6-8 months more than 80GB/s bandwidth are actually an absolute necessity. Personally I'd be glad if a G80/R600 can play coming games like UT2k7 with all bells and whistles on at 16*12; if anything should keep those from reaching playbility in higher resolutions I have severe doubts that the primary bottleneck will be bandwidth.

 

Yep.

Jawed

 

Sure, 64-80GB/s is a very good possibility for the first G80 SKU's. Which is even more reason to believe that the memory controller would happily handle speeds higher than 1.25Ghz. And we don't know what's coming in terms of AA - maybe there are some cool things there that can sap up any extra bandwidth if things are shader bound. However, I'd still wager a lot on things continuing to be bandwidth bound next generation with HDR+AA and other fancy framebuffer stuff going on.

 

Well, considering that nVidia will probably be making use of the G8x architecture for the next 2-3 years, 1.5GHz isn't that outlandish an upper limit on clock speeds for the architecture. But it won't be seen any time soon, so it seems odd to have that number in any list of specs.

 

Depends on how you look at it. Intel made a big fuss of however Netbust was supposed to scale to 10GHz. ATI made a big fuss about how the memory controller in R5xx was supposed to scale to GDDR4. PR PR PR.

Anyway, what we've seen are hardly specs, more like a bunch of rumours / made-up-stuff flying in close formation.

 

The memory controller does and did scale to GDDR4....The damn thing was BUILT for GDDR4 because they thought thats what the cards would be using at the time the R520 was planned for.

I suppose what you should have said was that ATI touted some hidden performance increase when the memory controller was configured for GDDR4 which, frankly, the X1950 failed to show completely. Its given some nice perks, namely the OpenGL ones, but no difference between the memory can be seen.

 

It would and i really don´t see anything that could hinder it to scale even higher.

First, there isn´t any reason why the memory controller would suddenly just stop working if you would fit the board with GDDR4 memory chips that are >1.5GHz, because, if the specs are met, you´ll just have to provide sufficient voltage and your PCB needs to cope with it. Second (they were refering to core frequency, after all) there is simply no way to estimate clock-improvements over time, counting in process maturity, new process-nodes, maybe SOI etc., so that number can only be derived from some obscure misunderstanding, or it was just put in there to make it look like it´s genuine, which is kinda funny, because it just looks like the exact opposite.

Concerning BW, R600 will have >80GB/s and i highly doubt that NV will be far below that, even on their parts which are based on the first production run of G80. They will only get even more ambitious, and G80 was designed for >2 years in mind.

 

Nvidia GeForce 8800 Specs Posted by Web-Site - http://www.xbitlabs.com/web/display/20060919075610.html

...
The whole Nvidia G80 specifications list by VR-Zone looks as follows:
* Unified Shader Architecture;
* Support FP16 HDR+MSAA;
* Support GDDR4 memories;
* Close to 700M transistors (G71 - 278M / G70 - 302M);
* New AA mode: VCAA;
* Core clock scalable up to 1.5GHz;
* Shader peformance: 2x Pixel/12x Vertex over G71;
* 8 TCPs & 128 stream processors;
* Much more efficient than traditional architecture;
* 384-bit memory interface (256-bit+128-bit);
* 768MB memory size (512MB+256MB)
...
* GeForce 8800GTX: 7 TCPs chip, 384-bit memory interface, hybrid water/fan cooler, water cooling for overclocking. $649;
* GeForce 8800GT: 6 TCPs chip, 320-bit memory interface, fan cooler. US$449-499;
...

 

Well, what I meant was that ATI make a big play of their forward-looking design, their future-proofing of their memory controller, and so on. This was positive spin on the design which, at the time it was announced, was a feature which was totally useless as far as the people buying the boards then was concerned, because it wasn't exploited. But it was good PR, because it made ATI look like a company with "vision"; made the best of a bad situation maybe.

So you could judge "scales to 1.5 GHz" either as as specification (which it clearly isn't). Or as the same sort of positive PR spin. Or as complete nonsense.

 

I can't believe even you're bungling supar sekret info, U. That last "oneone" should obviously be "eleventy" for anyone who knows anything about 3D. Hang your head!

I guess bdotobdot2's unnec'y repetition is as good a way to bookend this craziness as any. Now, let's all take Rys' recommended cold salt shower and move on ... to the leaked R600 specs! Like salt, they also contain zero calories and are likely to lead to hypertension: good stuff. Oh, right, geo, wrong thread.

 

did these guys piece together that xbox 360

 

Actually from what I've seen Nvidia's pricing is surprisingly UN-aggressive.

Basically what they're doing is being competitive in price/performance at every segment, but only that. In fact a case can be made ATI leads in performance in many price segments. The X1900GT vs the 7950 GT, for example. Or considering the X1900XT can be had for nearly only $300 these days.

But Nvidia is competitive in every segment with MUCH smaller dies, ~80% smaller.

So what that tells me is, unless ATI is somehow getting hugely better yields which is highly unlikely, what Nvidia is really doing is maximizing profit rather than market share (although their market share is certainly not suffering either). They are selling tiny dies at the same prices ATI is selling big dies. They are pushing that 40% profit upwards, while ATI is what, 20-30%?

The one exception to this price/performance analysis is the 7600GT, which while ATI owns performance in the $200+ segments, ATI has absolutely no reasonable answer for.

Also I would like to note the latest INQ G80 article states in a end footnote that Fuad thinks the leaked G80 specs unreasonable for this process generation. A quick thought exercise proves him wrong imo, 700 million transitors is just above double G71, while ATI is already selling ~80% larger dies at willy nilly sub $300 prices! Sure, G80 would be the largest and most expensive chip yet, but it's NOT unreasonably larger than R580.

Although I just realized he may also have meant the clockspeed, ah well I leave my comment.