NVIDIA GF100 & Friends speculation

[DegustatoR]

Wasn't clear enough I suppose.
480 SPs for GTX480. 250W TDP.
There were many numbers prior to this which may have came from the SP number and final clocks being changed up and down several time. But 480/250 seems to be the final specs for GTX480.

 

[CRoland]

"Most people who claim it has 480 SP claim to know it from their own sources, not to choose the rumor they like best. And lately, those people (Fudo, Hardware-Infos, VR-Zone, I think Neliz too) seem to be more numerous than the ones saying 512."

Not really. Both of them already spread both rumors, changing it almost every day. The TDP as well. It was 295, then 275, then 250...And neliz, well...he's been saying that the GTX 480 does NOT have 480 ALUs.

I don't claim to know the truth, but they at least say they know things:

The number of shaders is 480, something that our colleagues in the press reported a few days ago, but we managed to confirm it just recently.

 

[Ailuros]

195% - gtx480
183% - 5870 1GB
161% - gtx470
153% - 5850

But between the gtx285 and the 5850, difference is high.
Therefore, the tested games a little bit like better Ati.

mao5 said "synthetic" for the record. What that graph is supposed to tell me is that a 5970 is on average by nearly 90% faster than a 1GB/5870 yeahrightsureok.

 

[Silus]

Could you explain in detail, how exactly does this theory of yours work?

I'm in agreement with this:

http://forum.beyond3d.com/showpost.php?p=1410281&postcount=4171

And given that we more or less know that the chip will be clocked modestly, whether it has 480 or 512 ALUs enabled, it will gain more by having more units enabled.

 

[rpg.314]

mao5 said "synthetic" for the record. What that graph is supposed to tell me is that a 5970 is on average by nearly 90% faster than a 1GB/5870 yeahrightsureok.

Looks more like ~50% to me

 

[Silus]

I don't claim to know the truth, but they at least say they know things:

But that's the problem. All those sources (and some more) claim to know things. But then change them again the next day. Why don't we wait for the review days (which are near) and end this silly rumor season ?

 

[psolord]

195% - gtx480
183% - 5870 1GB
161% - gtx470
153% - 5850

But between the gtx285 and the 5850, difference is high.
Therefore, the tested games a little bit like better Ati.

Or maybe they used higher AA settings!

 

[Silus]

Wasn't clear enough I suppose.
480 SPs for GTX480. 250W TDP.
There were many numbers prior to this which may have came from the SP number and final clocks being changed up and down several time. But 480/250 seems to be the final specs for GTX480.

Based on what ? Personal belief or higher number of recent rumors pointing to 480 ?
Because in regards to TDP, there is a higher number that points to > 250 (275, 295).

 

[Ailuros]

Looks more like ~50% to me

The 5870 ends up just a notch below 200 (I estimated 180) while the 5970 seems to be in the 270 ballpark.

 

[rpg.314]

I'm in agreement with this:

http://forum.beyond3d.com/showpost.php?p=1410281&postcount=4171

And given that we more or less know that the chip will be clocked modestly, whether it has 480 or 512 ALUs enabled, it will gain more by having more units enabled.

IMHO, the setup/raster per clock will be the same whether it has 14/15/16 SMs. So I'd say clocks matter more for that particular bit of rendering.

 

[Alexko]

I'm in agreement with this:

http://forum.beyond3d.com/showpost.php?p=1410281&postcount=4171

And given that we more or less know that the chip will be clocked modestly, whether it has 480 or 512 ALUs enabled, it will gain more by having more units enabled.

That shouldn't really matter as long as you don't disable any err... what was it called again? Geometry engine thingy? Anyway, as long as all 4 are enabled, whether 480 or 512 SPs are enabled shouldn't change much.

Also, 10% higher clocks will usually give you a bigger performance boost than 10% more units, regardless of the architecture.

 

[rpg.314]

The 5870 ends up just a notch below 200 (I estimated 180) while the 5970 seems to be in the 270 ballpark.

My guess exactly. Which makes it equal to 270/180 which is 1.5 aka 5970 is 50% faster than 1G5870

 

[CRoland]

The 5870 ends up just a notch below 200 (I estimated 180) while the 5970 seems to be in the 270 ballpark.

180/270 = 1.5, so 50%.

 

[rpg.314]

Also, 10% higher clocks will usually give you a bigger performance boost than 10% more units, regardless of the architecture.

In general,

 

[rpg.314]

180/270 = 1.5

It seems I need to go back to primary school again.

 

[CRoland]

It seems I need to go back to primary school again.

I may have got that one wrong... Unfortunately I cannot edit.

 

[Alexko]

In general,

Assuming you don't add any cache with the additional units, I don't see how you could get more performance out of the same proportional increase in units.

 

[rpg.314]

I may have got that one wrong... Unfortunately I cannot edit.

It's ok. Welcome to B3D

 

[GZ007]

Watts isn't a unit of heat, it's a unit of power.
.

Watt is a unit of energy first. 1 watt is 1 joule of energy per second. U cant create nor destroy energy, just change its form. U can use this unit to different form of energy. Engines are measured in watts too.

 

[rpg.314]

Assuming you don't add any cache with the additional units, I don't see how you could get more performance out of the same proportional increase in units.

This is part of my point. It is in general, at least sometimes, given enough tweaks, possible to increase perf by increasing units even with lower clocks. Which is why I (later on) added a qualification to my response.

Clocks aren't the primary performance determining factor every time.