Graphics Performance List

[S0ULphIRE]

I look at it this way. AMD/ATI releases the 5000 series to compete against nVidia's 300 series and vice versa.
Now the 5000's are out (barring a few more possible x2 cards or so I hear?) and so far they've only managed to put out ONE card that beats nVidia's previous gfx set.

Now I don't think nVidia's been working all this time to release something that's only 'just as good' as their previous cards. When the 300 series is finally released, my prediction is that the majority of the cards will blow the 5000's out of the water in terms of processing power.

 

[veg1992]

I look at it this way. AMD/ATI releases the 5000 series to compete against nVidia's 300 series and vice versa.
Now the 5000's are out (barring a few more possible x2 cards or so I hear?) and so far they've only managed to put out ONE card that beats nVidia's previous gfx set.

Now I don't think nVidia's been working all this time to release something that's only 'just as good' as their previous cards. When the 300 series is finally released, my prediction is that the majority of the cards will blow the 5000's out of the water in terms of processing power.

Agreed!

What I'm waiting for is to see the prices, I'm not expecting good prices though.

 

[yzmxer608]

I like Nvidia (all of my gpu's have been Nvidia, first was the mx 400 ) but their prices are just too high compared to ATI imo.

 

[Apokalipse]

I read that Nvidia are having difficulty reaching their clock targets with G300 (now called GF100)

 

[Puddle Jumper]

Now I don't think nVidia's been working all this time to release something that's only 'just as good' as their previous cards. When the 300 series is finally released, my prediction is that the majority of the cards will blow the 5000's out of the water in terms of processing power.

The GF100 based Tesla cards are supposed to have double precision floating point performance in the 520-630 GFlops range which is quite a bit lower than the 768FFlops some sites had estimated it would have. A HD 5870 gets 544GFlops in double precision so at least as far as floating point is concerned Nvidia is hardly blowing anyone away.

 

[Peter.Cort]

isn't part of it how the card is made though, not just straight power? I thought that was the big downfall of the 3k series from ATI, on paper it should have been more powerful, but in practice it wasn't.

 

[Trotter]

The 3xxx line weren't much more powerful than the 2xxx line, but they used less power and ran MUCH cooler.

 

[Apokalipse]

isn't part of it how the card is made though, not just straight power? I thought that was the big downfall of the 3k series from ATI, on paper it should have been more powerful, but in practice it wasn't.

The 3xxx line weren't much more powerful than the 2xxx line, but they used less power and ran MUCH cooler.

The 3000 series was supposed to give about the same performance as the 2000 series - and it did.
The difference is a massive die shrink, making it way cheaper to make - and yes, a massive reduction in power consumption.
It was the first step in AMD's stragtegy of very cheap, but fast GPU's.

Given TSMC's issues with their 40nm process, the small die strategy is working in AMD's favour in another way. Smaller dies tend to get significantly better yields - and that would be partly why AMD have been able to sell their 5800 series cards, while Nvidia are still struggling.

 

[zmatt]

The 3000 series was supposed to give about the same performance as the 2000 series - and it did.
The difference is a massive die shrink, making it way cheaper to make - and yes, a massive reduction in power consumption.
It was the first step in AMD's stragtegy of very cheap, but fast GPU's.

Given TSMC's issues with their 40nm process, the small die strategy is working in AMD's favour in another way. Smaller dies tend to get significantly better yields - and that would be partly why AMD have been able to sell their 5800 series cards, while Nvidia are still struggling.

yeah, you think ATI is having yield issues, just wait until Nvidia starts producing their massive cards. They will be rarer and more valuable than diamonds. Considering that when they began producing GT200 they had single digit yields per wafer, and now they have a bigger and more complex die on an even faultier process. IMO even when GT300 is released we will have HD5000 as the only practical high end solution for some time. And to compound that I have heard, rumors only, that Nvidia is having some serious issues aside from TSMC. They are having scaling and heat issues which doesn't sound hard to believe given their strategy.

 

[Puddle Jumper]

I was reading a interesting thread on Anandtech where someone estimated the cost of a GF100 die based on it's size, the number of dies that could fit on a wafer, the cost per wafer, and TSMC's current yields. He estimated that you would get ~90 functional Cypress or ~38 functional GF100 dies per wafer with Cypress costing $78 per die and Fermi costing $184 per die.

When you factor in the profit margins ATI and Nvidia would like to make per chip Cypress would sell for $170 while Fermi would cost $370. This is just for the gpu and doesn't include the cost of any of the other components on the card or the OEM and Retailers profit margins.

Here is the source

AnandTech Forums - View Single Post - Fermi possibly delayed til March or April