AMD: R7xx Speculation

[Wirmish]

AMD = more dies = more testing & more packaging = more cost ... no ?


EDIT: Sorry, now I understand...
The yields of RV770 must be in the range of 60% ~ 80% and I use the same yields for both chips.

wafer = $5000 (let's say)
GT200: 89 dies with 40% yields = 35 good dies -> $5000 / 35 = $143 per die + testing/packaging.
RV770: 221 dies with 70% yields = 154 good dies -> $5000 / 154 = $33 per die + (2.4 x testing/packaging)

Wouldn't those GPU dies need to be tested anyway?

Good point... but RV770 = 2.4 x dies than GT200 -> 2.4 x more testing.

 

[v_rr]

AMD = more dies = more testing & more packaging = more cost ... no ?

It´s also more cards on the market and more money, no?

 

[Foodman]

AMD = more dies = more testing & more packaging = more cost ... no ?

Wouldn't those GPU dies need to be tested anyway?

 

[DegustatoR]

4850 will be a bit more expensive than 8800GT, but significantly faster. 4870 will be a bit more expensive than 9800GTX but significantly faster. That means ATI is actually competing with Nvidia on price/performance for the first time since R580. Why wouldn't you be optimistic about that?

I've said "overly" optimistic. Cautiously optimistic i am =)
At the very least RV770 isn't half-year late to market as it was with R520 and R600.
But what you have to understand here is that 8800GT EOLing soon and 9800GTX price is a result of another market structure.
RV770 undoubtedly will provoke some price correction and i have my doubts that in the end of it 4850 will compete with 8800GT and 4870 -- with 9800GTX.

R700 is still an unknown quantity at this point. If it does actually have a shared memory architecture, it may do a much better job of AFR than any previous product. If not... well, in games where AFR works well, it should quite significantly outperform GTX280 for about the same money. How many games that is remains to be seen.

Any AFR solution is a waste of money from my POV, sorry.
And NuMA rumour seems to be the same as 480 SPs rumour =(

And by that time comes RV770 in 40nm, and them we will see who catch who

By that time we'll just start seeing 4870 in normal quantities i'm afraid.

 

[AlexV]

By that time we'll just start seeing 4870 in normal quantities i'm afraid.

Really?

 

[Wirmish]

Dealer here say 480 SP.

 

[AlphaWolf]

Dealer here say 480 SP.

I'd be inclined to believe my mailman as some retailer.

 

[Pete]

It's XS, what do you expect?

Something more extreme!

 

[Arun]

Let's not be ridiculous, unless GT200b taped-out substantially longer ago than anyone thinks it did and it's an A11, I don't see how 4870 wouldn't be available in volume before then.
BTW, when do I get my cookie for sticking to the 800 SPs theory? Or for being the first person to publicly state the possibility that GT200 had 240 SPs, fwiw...

 

[Shtal]

When we presented you with a slide saying the RV770 core sports 800 shaders and not 480 like most people believed, it was met with a certain amount of skepticism, which is very understandable as many of the larger news sites have been reporting 480 shaders, although without presenting any proof for it. We know now that the rumor of 480 shader processors was planted by AMD and that RV770 indeed has 800 shader processors. We also know that AMD is aware that the RV770 as a single core is not enough to match the GT200 core from NVIDIA. Not surprisingly, AMD is going to play its trump of price/performance, but that doesn't mean that there's no raw performance to play with.

RV670 and the Radeon HD 3800 series offered a decent price/performance ratio, but it was rather weak compared to NVIDIA's best cards. RV770 is a mainstream chip, but unlike RV670, RV770 is going to play in the upper mainstream and performance segments and is a much more worthy opponent to NVIDIA's high-end GPUs. The slide below also shows AMD comparing the performance of Radeon HD 4870 and GeForce GTX 260 measured in FLOPS;
After NVIDIA's decision to cut prices, Radeon HD 4870 and GeForce GTX 260 will be more evenly priced. However, you will still be able to get two Radeon HD 4850 for the price of one GeForce GTX 260 and looking at the recent CrossFire results posted over at Chinese forum Chiphell, 4850 CrossFire scores better. http://www.nordichardware.com/news,7860.html

Interesting....

I wonder how well HD4850X2, HD4870X2 will perform in real games compare to Nvidia GT200.

 

[AlphaWolf]

Another zeroed graph, I think I might faint. What's going on at AMD?

 

[jamis]

BTW, when do I get my cookie for sticking to the 800 SPs theory?

Well it's not actually confirmed is it? It still could be 480 with clock domains, that rumour was first.

edit: Too late it seems to be out...

 

[Jawed]

Another sloppy AMD slide. Not only is the green bar too short, but so is the red one

EDIT: The back wall of the graph is spaced back, so it does make sense after accounting for that - and I checked by counting pixels...

Jawed

 

[Jawed]

BTW, when do I get my cookie for sticking to the 800 SPs theory?

I'm afraid to say nullified by your insistence that this would be a significantly larger chip, competing at the same level as GT200 and that there'd be no X2

Jawed

 

[nAo]

Let's not be ridiculous, unless GT200b taped-out substantially longer ago than anyone thinks it did and it's an A11, I don't see how 4870 wouldn't be available in volume before then.
BTW, when do I get my cookie for sticking to the 800 SPs theory? Or for being the first person to publicly state the possibility that GT200 had 240 SPs, fwiw...

You don't get any cookie until you explain how they put all that stuff in 260 mm2.

 

[Shtal]

Another zeroed graph, I think I might faint. What's going on at AMD?

For some reason I don't see Texture unit graph made against comparing competition. "How well - fast / extreme edition TMU's does by ATI/AMD"

 

[pso]

You don't get any cookie until you explain how they put all that stuff in 260 mm2.

ATi uses engineering wizardry.

 

[Wirmish]

So...

HD 4850 = 1042 MHz (clock domains) * 2 FLOP (MADD) * 480 SP = 1.0 TFLOPs
HD 4870 = 1250 MHz (clock domains) * 2 FLOP (MADD) * 480 SP = 1.2 TFLOPs

or

HD 4850 = 625 MHz (core) * 2 FLOP (MADD) * 800 SP = 1.0 TFLOPs
HD 4870 = 750 MHz (core) * 2 FLOP (MADD) * 800 SP = 1.2 TFLOPs

?

 

[Arun]

I'm afraid to say nullified by your insistence that this would be a significantly larger chip, competing at the same level as GT200 and that there'd be no X2

I am still not convinced at all by the current die size rumours, although it does look like I was probably wrong and it'd still be small enough to be used in a X2 if clocked low enough and properly binned to keep the TDP in check... We'll see.

 

[coredump]

ATi uses engineering wizardry.

It's a Tardis GPU