NVIDIA GF100 & Friends speculation

[Mindfury]

GTS450 draws 40W more than HD5770 under load in overclockzone's benchmark.

 

[no-X]

That's performance of HD 5750 with power consumption of HD 5850-5830...

 

[CRoland]

So is GF106 192SP and this their first fully functional Fermi derivative?

 

[Alexko]

So is GF106 192SP and this their first fully functional Fermi derivative?

Yes and yes, apparently. Looks like a half-GF104.

 

[Mindfury]

So is GF106 192SP and this their first fully functional Fermi derivative?

No,fully functional GF106 should have a 192bit bus.

 

[no-X]

GTS455 isn't based on GF104?

 

[neliz]

GTS455 isn't based on GF104?

That would be my question too, Why would GF104 be half a GF100 and GF106 not be half a GF104? (and probably a square die again at that.)

 

[Mindfury]

That would be my question too, Why would GF104 be half a GF100 and GF106 not be half a GF104? (and probably a square die again at that.)

GF106 die is too huge for half of GF104,over 230mm^...

 

[mczak]

http://overclockzone.com/tor_za/year_2010/08/asus_engts450_top/index5.htm

the card is not competitive with HD5770 performance wise , that job is for GTS 455 they say .

Dunno doesn't look too bad depending on the benchmark - I was long saying half a GF104 will have trouble competing with HD5770, unless they get clocks up quite significantly (which they did, but not quite enough). What was the clock of the OC GTS450?

No,fully functional GF106 should have a 192bit bus.

Hmm so what now? The minimum is 192CC, 128bit bus. So is full GF106 really having 192bit bus - and 192 or 240CC? I seriously doubt 192bit bus would be making a whole lot of a difference anyway. Though it has to be either something along these lines or GTS455 is a further crippled GF104 instead of GF106 (which would leave the question why GF106 is so big).

 

[Mindfury]

Dunno doesn't look too bad depending on the benchmark - I was long saying half a GF104 will have trouble competing with HD5770, unless they get clocks up quite significantly (which they did, but not quite enough). What was the clock of the OC GTS450?

asus top 925/1850/4000 instead of 783/1566/3600

Hmm so what now? The minimum is 192CC, 128bit bus. So is full GF106 really having 192bit bus - and 192 or 240CC? I seriously doubt 192bit bus would be making a whole lot of a difference anyway. Though it has to be either something along these lines or GTS455 is a further crippled GF104 instead of GF106 (which would leave the question why GF106 is so big).

I guess GF106 is half of GF100,256SP/192bit.

 

[Jawed]

Big because it has two GPCs?

 

[mczak]

asus top 925/1850/4000 instead of 783/1566/3600

Are those release clocks not further OC? That would be higher than the so far record shader clock of GTS250 (well non-OC that is). It would also explain the somewhat high power consumption, if the review simply clocked this card down to nominal levels (because I'm pretty sure this high clock would need voltage bump).

I guess GF106 is half of GF100,256SP/192bit.

For a straight half GF100 it would be too small, unless other things got chopped out.

 

[Mindfury]

It would also explain the somewhat high power consumption, if the review simply clocked this card down to nominal levels (because I'm pretty sure this high clock would need voltage bump).

GTS450 at default 783Mhz clock draws 40W more than HD5770 under load in overclockzone's benchmark.

 

[mczak]

GTS450 at default 783Mhz clock draws 40W more than HD5770 under load in overclockzone's benchmark.

When OCed,it draws 60W more than HD5770.

Yes but didn't they just simply downclock the factory OC card, hence still using a higher voltage?

 

[Alexko]

GTS450 at default 783Mhz clock draws 40W more than HD5770 under load in overclockzone's benchmark.

I think mczak is saying that they reviewed the Asus TOP and lowered the clocks to match the reference model, but since the TOP model is so heavily overclocked, it probably features higher voltage as well, thus increasing its power consumption even at reference clocks.

Edit: whoops, too slow.

 

[CarstenS]

Big because it has two GPCs?

No, 192 divide by 48 quite nicely for a (as in „one”) full, GF104-style GPC.

 

[trinibwoy]

That would be my question too, Why would GF104 be half a GF100 and GF106 not be half a GF104? (and probably a square die again at that.)

GF104 isn't half a GF100, far from it. GF106 seems much closer to half a GF104, but for the 192-bit bus.

 

[Jawed]

No, 192 divide by 48 quite nicely for a (as in „one”) full, GF104-style GPC.

Can a single GPC provide the requisite geometry/rasterisation throughput?

 

[CarstenS]

Please specify. Prerequisite for what?

 

[trinibwoy]

Not prerequisite, requisite. i.e required, adequate, necessary....