disabling pipes - does it affect power consumption/heat output?

[neologan]

Let's take R580 as an example, an 8/48/16 part. The x1900 GT has one quad disabled making it an 8/36/12 part. How will this affect power consumption and heat output?

With one quad disabled can you essentially say the transistor count is now less than a full R580 with all quads enabled?

 

[Tim Murray]

Let's take R580 as an example, an 8/48/16 part. The x1900 GT has one quad disabled making it an 8/36/12 part. How will this affect power consumption and heat output?

With one quad disabled can you essentially say the transistor count is now less than a full R580 with all quads enabled?

Uh, what 8 are you talking about? It's not 8 ROPs; 16 full pipes, 16 TMUs, 48 ALUs.

Also, judging from coolers on various parts, yes, power consumption and heat should be reduced somewhat. You're not going to have a full 25% reduction by disabling one of four quads due to the rest of the board, but it's not insignificant.

 

[neologan]

was talking about vertex pipes, though to be honest, it was irrelevent to mention those in this question.

 

[Pete]

An uneducated guess

We'll know soon enough, once sites get their mitts on the X1900GT. But it's hard to deduce from current cards, as most differ in clock speeds as well as pipes (X800P/XT, 6800/GT, 7800GT/GTX) and possibly--perhaps more importantly--in core/mem voltages. I suppose comparing the X800XL and X800GTO gives us as good a chance as any to detect differences, as both of those cards are based on the same R430 chip and are clocked the same (and so may have the same voltages, despite disabled pipes). There's a big difference in "peak 2D" consumption but practically none in peak 3D. That only serves to confuse me more.

I'm with Baron thinking that it should reduce power draw somewhat (good point about using the coolers as clues), but I don't know how close it gets to the proportion of the GPU that's bypassed. Many of these "disabled" quads are done via the BIOS, so there's no physical separation. I don't know how similar or sophistiicated this is to dual-core CPUs "shutting down" an unused core (which I believe the Core Duos do).

 

[ERK]

Most modern processing chips (to my knowledge) have the capability to "power down" sections when not in use. I would be very surprised if these GPU chips did not do so as well, especially since we know that power is very important to the application, and that the capability of de-selecting quads is designed into the chips. It would be silly not to do it.
ERK

 

[ChrisRay]

It really wont make a difference on NVidia hardware. If you disable vertex/pixel pipelines the tempatures remain just about the same as if you enable them. Back when the 6800 NU's were modding this was a pretty hot topic as the cards didnt change in tempature output.

 

[ERK]

Heh. I'm surprised.

Although I guess most of my experience (my company, for instance) is with low power designs. After all, it does take a few extra transistors. *shrug*

 

[soylent]

Does ATi list a different recommended amps on 12v line(s)/total wattage for x1900gt than for x1900xt?

 

[stepz]

Even if they would power down the defective quads, the effect is small compared to ramping down the clock and voltage. Assuming the chips are running near their reasonable limits, I'd say that power could more than halved by reducing clock 25-33%.

 

[ERK]

Even if they would power down the defective quads, the effect is small compared to ramping down the clock and voltage. Assuming the chips are running near their reasonable limits, I'd say that power could more than halved by reducing clock 25-33%.

Only if the voltage could be reduced a little at the same time, which is usually possible.

 

[roadie]

I would really like to know exactly how much power the x1900xt uses. I know that in Dave's review the test system only drew a little over 300W but I am having problems with a 400w Enermax Liberty and a 480W Hiper. The Enermax can supply 30A combined over the 12V rails and the Hiper 29A.

I am having a real problem with system stability and have 3 suspects: the PSU(s), RAM and the video card.

For some reason I get many more crashes when using the Enermax instead of the Hiper, which I bought on the basis of its strong 12V rails, cool running and the fact that my system wasn't going to need as much power as Daves test sytem (I believe). I have tested the Hiper using a multimeter and the voltages are high and stable. I still need to test the Enermax with it.

I also have downloaded memtest to test my RAM and rthdribl to test the video card. However, I am pretty sure that rthdribl will cause it to crash too. So hopefully once memtest has run I can strike something else from the list.