NVIDIA Fermi: Architecture discussion

Was PhysX being used?

Jawed

 

It can be that while they're using "égalerait" (will reach, will be on par), maybe they meant "will rival", implying that a vga can rival a card that has lower performances.

 

What do you base that on? What makes you think the x2 version would be 300w? You seem adamant to not let the 300w TDP stand...

 

Very apparent around 1' before the end, when they throw some (chickens?) on it and destroy the engine.

Not that impressive though, any software physics engine could do that.

 

that's how I would read "Il se susurre d’ailleurs que la plus puissante des Fermi égalerait une Radeon HD 5870 mais pas la 5970" - that it would "match" the former but not the latter; ie. that it's performance will still surpass that of the 5870. Why else would you bring in the 5970 into that phrase if all you wanted to say was that fermi would only be the equal of the 5870? A range is implied - it is able to "equal" the lesser card, but not the greater.

however, the 300 watt claim, if true, would be horrible, even for the top-end variant, unless perhaps nvidia is heavily overclocking the cards for purpose of the demo, which would still be quite worrisome.

 

TDP isn't the same thing as power consumption

The GTX 280 also has a 8 pin and 6 pin connector with a TDP of 300 watts never used that much though.

 

Yeah, exactly. The same thing happens on a daily basis to owners of graphics cards (drivers and hardware from the same manuf not playing nice).

Well for some reason people aren't paying attention to GT200 when making comparisons. Parity with HD5870 would be a complete and utter failure given that it's only 30-40% faster than the GTX285.

The whole thing is PhysX based. The engine is made up of individual components that exert forces on each other and the sled. The demo has a wireframe mode showing all the force vectors as well. There's also a shack that disintegrates when the sled passes by and some stone structure that collapses onto the track too. All PhysX presumably.

 

Pricing it at the level of 5870 will be a financial fail.

 

Parity with HD5870 would make it 5-10% faster than a GTX295 in AA/AF environments, or on par with a 295 without AA/AF:
http://www.computerbase.de/artikel/hardware/grafikkarten/2009/test_grafikkarten_2009/18/

 

Right. TDP means how much heat it generates and is required to dissipate.

 

The TDP of GTX280 is 236 Watt. I think Fermi will stay in the same range. But never ever would they go nearly to 300 watt with this cooling solution.

They are using two D3D11 and 8 D3D10 games for the performance rating. So Fermi will be also faster with d3d11.

http://www.tomshardware.com/news/Nvidia-3D-Surround,9394.html

Looks like, they will go the quadro nvs way.

 

I'm not saying that that's what the x2 will have. I did say that a TDP as high as 300w is expected for the X2 card and not the single chip version, as seems to be implied by the french site. Single chip version should be on the same level of GT200 in terms of TDP: 236w
If GT200 @ 65 nm, 576 mm2 and 1.4 billion transistors had a TDP of 236w, how can 3 billion transistors @ 40 nm, < 576 mm2 have a TDP of 300w ?

 

Gotcha.

Jawed

 

Perhaps by the virtue of the "hurry up guys, we're lagging behind !" effect...

It's the only thing which could explain why GF100 would ever need that much power : bad performance. GT200 didn't have a performance target, GF100 has one.

As you said, there are 3 billions transistors in there, twice as much as inside the GT200, and power doesn't scale linearly with processes as already seen with previous references (G92b and to a lesser extent GT200b), but it scales more or less linearly with transistor count on a given node for a given design, and again with traces lengths, which increase with transistor count.

Oh, and don't forget GTX295 exceeded its TDP even with "non-powervirus" shaders according to some reviews, since HD5970 draws no less than 25 watt less although it's specced 5 watt higher.

 

I'll be straight forward, it doesn't need that much power and its performance, drivers are raw and its performing well.......

 

I take you don't believe Nvidia when they say it uses PhysX? Do you have reason to believe otherwise or is everything Nvidia says simply assumed to be a lie?

Is that supposed to be impressive? The 295 is only 35% faster than a 285. I sense a repeating theme here.

 

Nope. I'm still going for the "it's either 30-35% faster than the HD 5870 or it's fail".

 

Not really, they survived the GT200 versus RV770 battle. If performance lead isnt significant (>30%) then they should price it at 5870 levels. If performance is greater than that grin then they can claim a price premium.

 

I'll ask you again, was PhysX being used? Nothing I can discern in the video or audio indicates as such. First you state it as fact, then you say presumably, then you say it's a fact, again

Jawed