GF100 evaluation thread

Maybe Harison meant a loss compensation through it's software/compute/Tesla/Quadro department? In a way dumping development costs in GeForce and recouping those costs through higer priced professional gear.

Selling 3D peripheral devices also starts looking more like it, although I have no clue about margins on shutter glasses and the Samsung screens.

 

How much larger is the TSMC capacity for applying the upper metal layers compared to the silicon layers? (Note that the number of metal layers is variable in the first place, so apart from parked wafers they need quite a bit of "over" capacity there to begin with.) If they have enough spare resources to do that at say 10x the speed of new wafer starts, and without having to compete with other companies, then NVIDIA might have build up quite a bit of stock on a hail Mary prayer.

Also market presence is a reason to go to small scale production when they know yields will be too low to earn a profit ... whether they are in full production or small scale production is up in the air until we hear how many they shipped, probably through conference calls, and since they conveniently start shipping at the start of a quarter that will likely take quite some time.

 

Why would they build up inventory at all when the yields are extremely low?

 

Misplaced confidence in being able to get it up through metal work combined with a wish to have something on the market place even if it ends up costing them a couple 10s of million?

 

Is it normal for reviewers to have to send back video cards after they have done reviews? Or do Nvidia really not have that many cards available? Because apparently Techpowerup has already sent their card back, and they didn't get a GTX 470 and none of the Nvidia partners have any available for reviewers either.

 

There was a story from Charlie that apparently at around the A2 tape in timeframe, nvidia put in a large order for base layer wafers on the assumption that when A2 came back, all would be good. The number quoted was in the range of 50 mil or ~10-15k wafers.

 

yields were extremely low for 512CC 700+Mhz parts, 480/600 seems to do quite a lot better.

 

It's impressive that no matter how much wrongs the guy gets, for you and others, he's always mostly right...

No tessellation hardware = wrong
Slower than Cypress initially and later only 5% faster overall = wrong
Clocks, general specs and TDP = wrong
GF100 would tank performance in DX11, while Cypress won't = wrong

The only things he got right were the dates...great facts you have there...

 

The GTX 480's GPU is clocked @ 700 Mhz.

 

Tell us what YOU have predicted accurately ?

 

Depends. It can vary by company, product, and website but my understanding is that generally for established sites they generally don't get asked for them back.

 

What did you think ? That reviewers always kept the cards they were sent, free of charge ?

 

I'm pretty sure for a fair few websites that is the case yes

 

Multiple people have confirmed a late change to both alus and clocks and that prior to that change the clocks were as specified. This hit a lot of people, I'm sure XMAN can fill you in.

As for TDP, yeah no one expected the stock 480 to have over a 300W TDP!

 

And is there a confimation to this ? Maybe Rys or anyone from the staff of B3D can confirm or deny that that is or has been the case for them.

 

Even if that's true, I guess the rest isn't too important correct ?

 

Absolutely nothing, but does that make the other predictions, despite being wrong, better ?

 

Without being 100% acurate on the numbers, Charlie was right on the overall picture.

 

It's common. You will often read that the site didn't do XYZ because they ran out of time before they had to return the card. It's not just graphics cards either - lots of product reviews happen where the product is shipped back afterward.