Fermi has 480 shaders, 675MHz+ clock

[mattyfez]

err... how can they run a dx 11 bennch with cards that dont support it?

 

[RavenXXX2]

ASUS Fermi Voltage Tweak

Brave man who turns up the volts on the 480

LOL, 50% faster, and it's gonna need it by the looks of things..muhahaha

http://translate.google.gr/translat...n/304-000-106-109/2010-03-19/19241285024.html

 

[bhavv]

err... how can they run a dx 11 bennch with cards that dont support it?

SSAO works in DX9 and DX10, it is just handled a lot more efficiently in DX11.

Battleforge doesnt use DX11 to improve visual quality in anyway. It uses the same graphics under both DX10 and DX11 mode.

Running under DX11 mode provides the same graphical quality with much improved FPS.

 

[TaKeN]

Brave man who turns up the volts on the 480

LOL, 50% faster, and it's gonna need it by the looks of things..muhahaha

What are they offering on the Voltage Tweak?... 2 day warranty? lmao

 

[bhavv]

Ummmm....

'Communication GTX470 card official recommended price of £ 299!'

No idea how true that is, but thats what the translated link says.

 

[Orangey]

That ASUS box must be an early mockup.

 

[RavenXXX2]

Nah, that's they way Asus do them, shot of the 5850 box of the Asus site.

 

[Orangey]

Yeah, I have the ASUS 5850 myself.

Anyway I expect there to be a tiny "*" next to the voltage tweak graphic somewhere.

 

[Tom|Nbk]

this is ocuk for the past few months

Lmao!

 

[555BUK]

Here are some very unqualified assumptions about 4x0 overclocking. This is based purely on the fact that both 57xx, 58xx & 4x0 use the same 40nm process.

57xx is half the size of 58xx, but both GPU will overclock to ~1000MHz with voltage increases (1.25 to 1.35v). After 1000MHz, heat seems to be the limiting factor. Several people with custom (water) cooling can hit 1100MHz.

Assuming Fermi has the same voltage tweak functionality, and assuming the reference cooling is sufficient (big assumptions), it is not unreasonable to expect ~900MHz out of good cores. We also need to consider the memory. NVidia use very low clocks for their GDDR5. If they use the same IC's as ATI (Samsung & Hynix), we should see "easy" 20-30% headroom for overclocks.

So, theoretically Fermi may be a complete monster when overclocked. Perhaps the reason NVidia have released the 480 with only 480 SP's is because they know they have a monster, and are holding the 512 SP version back for ATI's response.

Of course, this could all be complete rubbish and Fermn may be a lemon.

 

[Rroff]

You'd deff. get ~900MHz on good cores... the problem is the number of good cores is likely to be a very very low proportion.

 

[Orangey]

Will overclocking the mem matter on Fermi or will it be like 5800s...?

 

[Rimsy]

this is ocuk for the past few months

lmao... its funny because its true.

 

[Lightnix]

You'd deff. get ~900MHz on good cores... the problem is the number of good cores is likely to be a very very low proportion.

I'm just afraid Nvidia might have gone the wrong way with their cooling solution. Y'know, having plastic fins on their shroud and all.

 

[jigger]

Here are some very unqualified assumptions about 4x0 overclocking. This is based purely on the fact that both 57xx, 58xx & 4x0 use the same 40nm process.

57xx is half the size of 58xx, but both GPU will overclock to ~1000MHz with voltage increases (1.25 to 1.35v). After 1000MHz, heat seems to be the limiting factor. Several people with custom (water) cooling can hit 1100MHz.

Assuming Fermi has the same voltage tweak functionality, and assuming the reference cooling is sufficient (big assumptions), it is not unreasonable to expect ~900MHz out of good cores. We also need to consider the memory. NVidia use very low clocks for their GDDR5. If they use the same IC's as ATI (Samsung & Hynix), we should see "easy" 20-30% headroom for overclocks.

So, theoretically Fermi may be a complete monster when overclocked. Perhaps the reason NVidia have released the 480 with only 480 SP's is because they know they have a monster, and are holding the 512 SP version back for ATI's response.

Of course, this could all be complete rubbish and Fermn may be a lemon.

I doubt Nvidia will use the same spec RAM as ATI, as it would add quite a lot of unnecessary expense to a card thats already going lose Nvidia a lot of money on each sale. I think the 5870 has the .4ns Samsung chips and they are still quite expensive, plus its typhoon season in Taiwan and that usually bumps up the IC prices quite a bit.

 

[jigger]

You'd deff. get ~900MHz on good cores... the problem is the number of good cores is likely to be a very very low proportion.

Maybe with a cherry picked core ( 1 out of hundreds of wafers supposedly ) and something like a 500watt pelter. If the TDP we are hearing is accurate its very unlikely Fermi will hit much over stock speeds without some extream cooling.

 

[bru]

one thing that i have thought of .......considering we are expecting limited stock of these cards there seems to be an awful lot of different companies showing the boxes for them.

 

[bee]

rather large heatsink

 

[bhavv]

the larger the heatsink, the better

 

[Broken Hope]

What £30 for some realistic physics effects?, sounds a reasonable deal to me - or effectively nothing extra for me as I already have a decent nVidia card setup.

Once devs start using multi-threaded CPU PhysX properly we shouldn't really need a nvidia card present to get decent performance.

http://physxinfo.com/news/2390/new-physx-fluidmark-1-2-first-tests/

Of course in a game you wouldn't be able to use 100% of a CPU for PhysX but most games barely use more than 2 cores, which on an i7 or hex core system leaves plenty of available CPU power to run multi-threaded physX too.