AMD: R7xx Speculation

[w0mbat]

So its 256mm².

 

[AlphaWolf]

So its 256mm².

~33% larger than rv670, would suggest suggest 890m transistors. (assuming the same process)

 

[Razor1]

think its on the performance process same node though, but in anycase shouldn't matter.

 

[trinibwoy]

So its 256mm².

Impressive. A nice round power of 2 to boot.

 

[Lukfi]

=>Razor1: It sure should, because the performance process will allow RV770 to clock ~200 MHz higher than RV670.

 

[Bouncing Zabaglione Bros.]

Well, that bolded part is what I was after. Why use a big, expensive, mid-range chip and cut it down when it's not necessary? That's why you've got low-end chips which are a lot smaller and cheaper to produce. Cutting down part of a perfectly functioning chip wouldn't seem like a sound decision to me. Especially in AMD's case. They don't have the high-end market, so they've got to find a way to appeal to consumers. So why cut functionality away from a chip? Wouldn't it be smarter to either lower the price of the whole product range, so consumers get a better perf/$ ratio? That's bound to get you more attention from press and consumers, hopefully resulting in more sales, a better name and something like an halo effect. I think AMD would benifit more from that path than from artificially creating new SKU.

Firstly, making chips is expensive. If you look at all the different SKU's, they are not going to make a dozen different chips. They are just going to make three or four and change speed and numbers of processing units.

Secondly, you don't want to cannibalise your high end sales by making your midrange cards too close in performance to your top range cards. You want mid-range cards to be better than your competitors cards at the same price. You don't want them to be half the price of your top range cards yet give 95 percent of the performance of your top range cards.

Thirdly, you want to manage your customer's expectations. If you sell them a fantastic card at $100, not only will they not buy your more expensive cards, they will not buy the upgrade for more money because you will have trained them to expect top performance for a mid-price. If yields go down at some point in the future due to some new product/process, you'd have somewhere to sell those faulty chips. Train your customers to only expect a top range chip, you're stuck with trashing all those chips that are partially faulty.

You need to have presence in these markets because (I as said above), it's better to sell a chip at a lower price/spec than not sell anything at all. It doesn't make any cash sitting on a warehouse shelf as a potential sale for more money.

 

[CarstenS]

~33% larger than rv670, would suggest suggest 890m transistors. (assuming the same process)

If they didn't have to add transistors to get clean signalling across the die, then I'd assume that number to be even larger, as transistor density tends to increase with larger chips. Maybe some nice round 900M?

 

[v_rr]

~33% larger than rv670, would suggest suggest 890m transistors. (assuming the same process)

wrong math. Is 34.7% ~35% larger then rv670 = 900M transistor in the same node as RV670.

But RV670 uses the first 55nm TSMC node and there is a better one.
RV770 should take the newer one.

All in all it can be close to 1B transistor chip with new 55nm process node and the removal of redundant parts of RV670.

Many changes can be made on that increase of space. Expecting 40-50% performance increase over RV670 is a nice value to bet and win.

 

[XMAN26]

wrong math. Is 34.7% ~35% larger then rv670 = 900M transistor in the same node as RV670.

But RV670 uses the first 55nm TSMC node and there is a better one.
RV770 should take the newer one.

All in all it can be close to 1B transistor chip with new 55nm process node and the removal of redundant parts of RV670.

Many changes can be made on that increase of space. Expecting 40-50% performance increase over RV670 is a nice value to bet and win.

40-50% better than RV670 for a single chip will put it on equal footing to the G80 GTX/Ultra cards. I sure hope for ATIs sake they do better than that.

 

[Slyne]

Impressive. A nice round power of 2 to boot.

666M transistors here, 256mm^2 there.... If that latter fact turns out true I'll start suspecting someone at ATI of number fetishism.

 

[Slyne]

An example could be NVIDIA's 8800GT. It was hyped because it gave near 8800GTX like performance for half it's price. Perhaps NVIDIA could have sold it at a higher price point, or created two SKUs out of it, perhaps with less SP's or something like that. They didn't however, and the result was an enormous amount of attention and praise, essentially drowning AMD's 3870 in the proces. The 8800GT may have had lower margins than possible, but it made sure everyone talked about NVIDIA. Who knows how many extra sales this gave them, just from the halo effect and the brand awareness? And I'm not just talking about 8800GTs, I'm also talking about other cards in NVIDIA's lineup.

Not sure you picked the best example to support your case. You think 8800GTS was a different design than 8800GT? 'Cause I definitely see a 16ALU/8TMU block disabled in the latter SKU.

 

[v_rr]

40-50% better than RV670 for a single chip will put it on equal footing to the G80 GTX/Ultra cards. I sure hope for ATIs sake they do better than that.

40-50% improvement can put it in the way of 9800GTX. With good pricing strategy it can be amazing to recover ATI.

But we need to wait to see more clear numbers.
Improvements of 100% in performance i think that whon´t happen but I hope that they surprise us

 

[Berek]

This article summarizes nicely the release schedule of RV770 and Nvidias as well. Looks like it will be GT200 first... a welcome surprise for those looking to buy the highest performing parts of the two (given the same card type... x2 vs. x2, etc.):

http://www.tgdaily.com/content/view/37453/135/

"highest" if assumptions going around are correct, anyway.

 

[Jawed]

Our sources explained to us that using a PCIe Gen1 controller 3870 X2 was a mistake, since the board was hungry for data and didn't sync well with this interface.

Hmm.

Jawed

 

[ChronoReverse]

I thought GPU-z is just a database. How on earth can the die-size reported by it be considered real information before the product comes out?

 

[XMAN26]

Not sure you picked the best example to support your case. You think 8800GTS was a different design than 8800GT? 'Cause I definitely see a 16ALU/8TMU block disabled in the latter SKU.

The 8800GTS 640/320 is different than the 8800GT. However the 8800GT and 8800GTS512 are the same chip.

 

[XMAN26]

40-50% improvement can put it in the way of 9800GTX. With good pricing strategy it can be amazing to recover ATI.

But we need to wait to see more clear numbers.
Improvements of 100% in performance i think that whon´t happen but I hope that they surprise us

Given the 8800GTX/Ultra is still better than that in high resolutions and high settings, meeeting the 9800GTX still in my view isn't good enough price with standing.

 

[Pete]

I thought GPU-z is just a database. How on earth can the die-size reported by it be considered real information before the product comes out?

My thoughts exactly. Are we sure Wizzard [sic?} knows the die size, or is he just guesstimating? It's not as earth-shattering as knowing ALUs or texture units, so it may not be as closely guarded a sekret, but are people accepting it b/c it's reasonable (in the expected ballpark) or b/c it corroborates their own source(s)?

 

[AlphaWolf]

wrong math. Is 34.7% ~35% larger then rv670 = 900M transistor in the same node as RV670.

Wrong math. 34.7% would be 897 million.

It's really just silly trying to get that accurate with a transistor number when the area is basically an estimate.

 

[Rangers]

This article summarizes nicely the release schedule of RV770 and Nvidias as well. Looks like it will be GT200 first... a welcome surprise for those looking to buy the highest performing parts of the two (given the same card type... x2 vs. x2, etc.):

http://www.tgdaily.com/content/view/37453/135/

"highest" if assumptions going around are correct, anyway.

The pricing sounds nice..the performance claim, "on par or better" than 3870X2 is pretty meaningless since with SLI issues that board scores all over the map, anywhere from trouncing any Nvidia single card solution to deplorable.

The fact 4870 supposedly not until July, and the 4800 series after GT200, are both pretty ughh. There's a period in there (after GT200 before 4800) where AMD is just going to be (even more) utterly nonexistent in market share.

I wonder what the "delay" was about this time.