AMD: R7xx Speculation

Status
Not open for further replies.
wide memory busses come with a not-insignificant associated board cost, too. They require more layers on the PCB, more complex routing, and produce a greater number of "bad boards" that need to be scrapped. It also can require a greater number of memory chips at a given total RAM amount.

So if ATI's 48xx series stuff has a 256-bit memory interface, and the GTX 200 based cards have wider memory busses, the difference in costs can be quite a bit greater than the difference in chip costs.
 
ATi uses engineering wizardry.

Gandalf the Grey
AMD employee since december 2004
Academic training:
· Research Master's degree in Computer Sciences specilised in Systems and Languages
· Ph.D. Thesis in Computer Sciences
· Ph.D. Electrical Engineering and Computer Science
· HDR – qualified Senior Researcher diploma
Experience/Skills:
· Microprocessor, chipset and memory design
· System and component-level validation
· Lean Manufacturing and process development.

Voldemort, Lord
AMD employee since may 2005
Academic training:
· Chalmers, Dept. of Computer Science, Goteborg, Sweden.
· Master of Science with a degree in Computer Science and Engineering.
Experience/Skills:
· Chipset layout
· Component design
· Digital/analog/RF circuit design

Dr. Strange
AMD employee since june 2005
Academic training:
· Ph.D. Computer Engineering
· M.Eng. Electrical Engineering
Experience/Skills:
· Failure analysis, isolation and diagnosis
· Verification strategies
· VLSI/ASIC/logic

:oops:
 
wide memory busses come with a not-insignificant associated board cost, too. They require more layers on the PCB, more complex routing, and produce a greater number of "bad boards" that need to be scrapped. It also can require a greater number of memory chips at a given total RAM amount.

So if ATI's 48xx series stuff has a 256-bit memory interface, and the GTX 200 based cards have wider memory busses, the difference in costs can be quite a bit greater than the difference in chip costs.

The difference in price between GDDR3 and GDDR5 on the other hand counts against ATi.
 
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

?

That's pretty much the math, but unless GPU-Z wasn't able to detect shader clocks on the 4850's (CCC doesn't show it unless its completely locked w/ core), and there is no shader domain, then it has to be 800SP's
 
Anyways, I don't know if any dealer has any concrete info on what the Sp count is unless they went to the press event.

From what I've seen and heard so far, they've been very very tight lipped on their specs. In fact, if you look at the shots of the RETAIL boxes of the 4850 that are out on the internet now (and being sold in various countries...), they don't even list the SP count when you would think they'd be marketing it.

So ATI has been holding these cards specs VERY close to them
 
The difference in price between GDDR3 and GDDR5 on the other hand counts against ATi.
It'll be interesting to see if GT200 still suffers from the high-res MSAA performance cliff caused by running out of memory - while ATI GPUs at the same res have no problem (of course ATI GPUs have this cliff too, but it takes a higher-res/MSAA setting). If GT200 is so afflicted then GT260, for example, needs more memory to compete against HD4870 512MB.

But I agree, it seems unlikely 512MB of GDDR5 costs less than 896MB of GDDR3.

Jawed
 
AMD don't put the number of SP on the boxes because it's not fixed yet.
AMD will add some SP, when needed, with his latest teleportation technology.
 
ATI:
GDDR5 price

nVidia:
Dies per wafer
Bad yields
More PCB layers
Routing complexity
More of RAM chips
Heatsink cost

Winner ?

Comparing cost of parts between a 4870 and GTX 280/260 is like asking what costs more to build, a G35 or an M6?

If there is equal performance, then this is all a huge advantage for ATI. Since it appears that GTX 280/260 will perform significantly better than the 4870 with games, then it's a different story because the higher performance parts command higher prices.
 
Last edited by a moderator:
Comparing cost of parts between a 4870 and GTX 280/260 is like asking what costs more to build, a BMW 335i or a 545i?

If there is equal performance, then this is all a huge advantage for ATI. Since it appears that GTX 280/260 will perform significantly better than the 4870 with games, then it's a different story because the higher performance parts command higher prices.

And command also a very very very smal market share = few cards sold and lost terrain on the big market 100-300$.

Higher performance is god, but at what price? 400-650 is insame. That market share is <2-5% of global.
And global economy is not doing well also, so less expensive cards take the lead.
 
Last edited by a moderator:
GTX 280 is significantly faster, I know that.

Number of -250$ graphic cards buyers ?
Number of +600$ graphic cards buyers ?

(In "real world" I mean... not on B3D or XS)

So in the end... who's the winner for this summer ?
 
I think declaring a winner before the cards are even out is pretty premature.
 
Actually, NVIO is more likely to reduce the overall bill of materials than increase it for a chip that size.
 
Actually, NVIO is more likely to reduce the overall bill of materials than increase it for a chip that size.

Agreed. If NVIO was a good idea for G80 size, it's an even better idea for a bigger size. You'd cry real tears losing a good die over a bad spot in the NVIO area if it was integrated.
 
Status
Not open for further replies.
Back
Top