AMD: R7xx Speculation

Only drivers and power consumption under full load

EDIT: 4870X2 power consumption runnning idle, no complains from me.

 

Sometime november 2006, everybody was all raving about physics processing on a GPU. Now it's finally there and people are complaining about Nvidia cooking the books? It's not exactly their fault that FutureMark chose to go with PhysX (long before the acquisition of Ageia by Nvidia I would assume).

As for your general statement: upon release R600 was neck and neck with an 8800. We all know how well that reflected real world performance.

 

That is exactly the case here: http://forum.beyond3d.com/showpost.php?p=1179096&postcount=43

 

Because your crystal ball briefed you about what exactly the card will be and how it will behave, obviously.

 

I heard that a physical count of the die showed 180 units so that makes 900SPs? Was this confirmed by anyone else?

 

I never said R700 sucks, it just need time to mature. "R700 is very good, but it will get better overtime".

 

And you know this how?

 

True, it's not their fault that FM botched up and added that without a way to test if it will work simultaneously on one card.

From what I've heard so far the only way to take advantage of this in games currently is to have 2 video cards. One for the video rendering and one for physics.

Vantage however tests each seperately and then composites the scores (as I understand it). Leading people to think they can have GPU accelerated 3d graphics along with GPU accelerated physics on 1 card.

Then again I could be wrong about not being able to currently render 3D and Ageia game physics at the same time on one card. I haven't been following it too closely lately.

Regards,
SB

 

it's supposed to be possible to get both things on a single cars (as nvidia claims)

the argument of nvidia is totally wrong if you have to buy a second card to get physx, and nvidia know that

remember the time when nvidia was yelling that their cards could do physx even better tha the physx card, without pushing that hard

i think physx built-in driver have just been released
they might have some test on the internet

 

You can see on some benefit (working as physx and gpu) on custom UT3 maps, but there's not a lot of places where its working atm. Right now I'd put this in the future feature category, buying a card right now based on potential physx support may lead to disappointment.

 

You have an API that defines PhysX. You have a piece of SW that looks for the presence of the API, and when it's there it uses it. You have an implementation of that API that uses the GPU.

Exactly what do you suggest Nvidia should have done? Do app detection and disable PhysX? Of course, not, after all people have claiming for years that app detection is evil. Or is that only true when it serves your case?

The blame on the scores being skewed rests 100% on FM: they are the ones who decided that PhysX is part of the final results. They knew that PhysX would be implemented on the GPU before Vantage was released. They had a chance to take it out of the equation, but they didn't. Their mistake. Case closed.

Personally, I'm excited to finally see a reasonable chance of HW accelerated physics that doesn't suffer from the chicken and egg problem. You can't blame Nvidia for AMD not (out-)bidding on Ageia.

 

I'm perfectly able to run 3D stuff and CUDA on my little 8600GT: there are examples that do this as part of the SDK. My understanding is that PhysX is based on CUDA: at least that's what Nvidia has been telling the world. I have no reasons to doubt that.

If the inability to run PhysX and 3D at the same time is the core of the argument, then "from what I've heard" is not good enough.

 

Partially from logic!

 

Care to explain your logic?

 

Agreed. That's why I didn't suggest they add together all 3 separate scores, like they do now.

I find it rather pedantic that FM do not allow spare video cards to be used for physics even though Nvidia has shown it's perfectly capable of doing so in Vantage. As a user, I'd prefer the ability of running the equipment the way I want to. If a game doesn't support SLI/Crossfire or doesn't scale enough with it, then re-allocating a GPU to physics should be perfectly acceptable.

 

I don't think anyone disagrees with you. Even when you accused people of being angry at NVidia, it looks like everyone is pissed at FutureMark.

Any time a technology goes exclusive, it's silly to include it as part of a universal benchmark unless it is present in the majority of games.

 

EDIT: http://www.asus.com/news_show.aspx?id=11871

 

How about calling it pragmatic instead of pedantic?

It's hard to program for something if there's no way to test it.

 

Well, my bad then for completely misunderstanding the intention of CJ's post and those linking to it. So he meant it was FM 'cheat'ing then?

 

Tom's have the first look on 4870:

http://www.tomshardware.com/cn/398,news-398.html