The first "proper" Kepler news Fri 17th Feb?

Duff-Man · 21 Mar 2012 at 20:09

Gibbo said:
For example, when I ran Heaven earlier my GPU clock increased too 1201MHz, yet when I was playing Quake III it stayed at 1006MHz. Yet I had not changed any of the settings in afterburner.

Quake 3 will undoubtedly be CPU-limited. The 'boost' auto-overclock will be a function of current GPU load, as well as current power draw and framerate.

If GPU load is sitting at well below 100%, there is little reason to increase core clockspeed.

Macro · 21 Mar 2012 at 20:10

Sorry, just messing with the fanboys... I apologise before there's a page full of rant

StonedPenguin · 21 Mar 2012 at 20:12

xsistor said:
Yeah cos clocking to 1201MHz is so very important in Quake III

People still play that :confused:

xsistor · 21 Mar 2012 at 20:13

Gibbo does apparently.

ALXAndy · 21 Mar 2012 at 20:15

Mate of mine was playing the web based one. I got sick of it on the mac back in 1999 lol.

Sin_Chase · 21 Mar 2012 at 20:16

The launch video graph shows boosted clocks of 1.1Ghz wonder what headroom there is above that

ALXAndy · 21 Mar 2012 at 20:19

Sin_Chase said:
The launch video graph shows boosted clocks of 1.1Ghz wonder what headroom there is above that

Gibbo said it does 1200 in Heaven. I've heard rumours of 1500 but that may be false.

Duff-Man · 21 Mar 2012 at 20:19

StonedPenguin said:
People still play that

Damn good game

I still play it on occasion...

I wouldn't use it as a GPU test platform though! It will run quite nicely on modern mobile-phone GPUs...

Sin_Chase · 21 Mar 2012 at 20:26

What kind of PSU ratings do people reckon to give the 680 enough juice considering it's boosting capacity?

My Corsair 520Watter supports my 285GTX that seems to have a similar overall system power consumption figure to the ones leaked from Toms. That's on older hardware also. Reckon it should be fine?

ALXAndy · 21 Mar 2012 at 20:35

285 is a bit of a piggy IIRC so you should be good.

djxput · 21 Mar 2012 at 20:39

maybe I shouldnt be bringing this up but ...
what was that whole thing with alxandy about? ... did you feel nvda wasnt coming out with a quality product and then so this started a flame war?

and how are you feeling now in regards to nvda's product now that you learned more (suppose you will learn more tomorrow).

muon · 21 Mar 2012 at 20:49

So $499 will undercut the 7970 then. Seems good.

Jakus · 21 Mar 2012 at 21:50

Sin_Chase said:
What kind of PSU ratings do people reckon to give the 680 enough juice considering it's boosting capacity?

My Corsair 520Watter supports my 285GTX that seems to have a similar overall system power consumption figure to the ones leaked from Toms. That's on older hardware also. Reckon it should be fine?

Yes, that will be fine !

Gregster · 21 Mar 2012 at 22:09

Geeeez this thread got nasty again

1pm and I shall be ordering

I am like a kid in a sweet shop.

Jakus · 21 Mar 2012 at 22:10

Duff-Man said:
Quake 3 will undoubtedly be CPU-limited. The 'boost' auto-overclock will be a function of current GPU load, as well as current power draw and framerate.

If GPU load is sitting at well below 100%, there is little reason to increase core clockspeed.

Clock changing like that always worries me as there is always some situation that catches it out.
I have always found power draw to be directly proportional to GPU load not clock speed, essentially higher clocks don't heat the core or draw more current they just allow the potential for higher load related current draw.
It's more than likely just another power state in the bios, Fermi had 3.

It would surprise me if this is benchmark related function, enthusiasts are smart enough to pick this up and Nvidia would get a real slating.

xsistor · 21 Mar 2012 at 22:13

JAKUS said:
Clock changing like that always worries me as there is always some situation that catches it out.
I have always found power draw to be directly proportional to GPU load not clock speed, essentially higher clocks don't heat the core or draw more current they just allow the potential for higher load related current draw.
It's more than likely just another power state in the bios, Fermi had 3.

It would surprise me if this is benchmark related function, enthusiasts are smart enough to pick this up and Nvidia would get a real slating.

The reason power draw appears to be a function of GPU load is because of power gating circuitry on the chip. Power draw IS a function of frequency, however. The faster these transistors switch the more it draws because MOSFETs in CMOS design are such that they dissipate most of their power during switching and very little other wise.

Jakus · 21 Mar 2012 at 22:55

xsistor said:
The reason power draw appears to be a function of GPU load is because of power gating circuitry on the chip. Power draw IS a function of frequency, however. The faster these transistors switch the more it draws because MOSFETs in CMOS design are such that they dissipate most of their power during switching and very little other wise.

Far from me to argue with you as you are far more current with modern digital electronics than me...but sorry that's not my findings, Just put a power meter on the PC and run a GFX bench clocks and volts will stay in (p0) state for fermi but power will change massively.

You know what, ignore that, I just realised I don't know what is really happening inside the GPU between light and heavy rendering being totally honest, maybe you could humour my ignorance on that ?

xsistor · 21 Mar 2012 at 23:02

JAKUS said:
Far from me to argue with you as you are far more current with modern digital electronics than me...but sorry that's not my findings, Just put a power meter on the PC and run a GFX bench clocks and volts will stay in (p0) state for fermi but power will change massively.

You know what, ignore that, I just realised I don't know what is really happening inside the GPU between light and heavy rendering being totally honest, maybe you could humour my ignorance on that ?

I realize now that my post may not have been very clear. What I meant was, yeah, when you increase load you will see higher power draw. The reason for this is because there are more circuits active. Modern CMOS circuits, especially the modular ones like processors, use transistors to switch off chunk of circuitry at low loads.

But for a given circuit the power draw varies with frequency. What happens when you increase the load is more circuit elements switch on, so while you had 10 % of the chip drawing power before, you may now have 50% of it drawing power.

So that's a different cause altogether from what happens when frequency is varied. But for any given circuit in the on state (i.e. not power-gated or switched off), increasing frequency will increase power draw.

trickycurve · 21 Mar 2012 at 23:11

Apologies if its been mentioned before, but can this auto-overclock feature be turned off in favour of a static clock speed (overclocked or otherwise)?

xsistor · 21 Mar 2012 at 23:16

trickycurve said:
Apologies if its been mentioned before, but can this auto-overclock feature be turned off in favour of a static clock speed (overclocked or otherwise)?

You can overclock in spite of the auto-OC. So I don't know why you would turn it off. Maybe you can... Idk tbh.