SPCR

This discussion of a review of Crossfire 4670s raised the following question for me:

Presuming you have a certain amount of power dissipation, is it easier to disperse that heat if it is in one place or distributed across multiple devices?

Two 4670s in Crossfire draw approximately the same power under load as one 4850 (and much less in idle, btw). So if one would compare one (passive) 4850 against two (passive) 4670s, let's say inside the P182 - where would you expect lower system temperatures? What would you think would happen if one added the middle (drive cage) fan?

(Note: right now it's not economically feasible to buy two 4670s. But with the introduction of the 4770 between 4850 and 4670, ATI will have to make some price adjustments - and I do not expect the 4850 to go up.)

The question in your header and the much more detailed query in your post are not the same.

In general, the same amount of heat across multiple locations is easier to cooler quietly; low heat density is almost always easier to deal with. However, depending on the kind of heatsink you can install on the GPUs you mention, the room available for size of fan, etc, the answer for your post could go either way.

Hmm, right. Slightly misleading... better now?
However, the question is not necessarily specific to GPUs - another example would be CPUs for applications with ideal multiprocessor scaling.

And I just have the feeling that there is a simple answer to that question. All I could come up with are complicated case studies (see the following).

Regarding the heatsink, let's presume manufacturer-installed passive heatsinks. those will probably be dimensioned so that the 4670 and the 4850 run at approximately the same (chipset) temperature under load, as the manufacturers will only install what does the job.

However, as less power is distributed locally, the ambience of each card (VRam, PCB, ...) should run cooler, right? Even if we take into account that some air will rise up from the second card, this should still add less heat (locally) than if the heat was generated right there. Overall system temperature "far away" (CPU) should stay approximately the same (or less as more heat will escape through the open PCI grills on the back?).

Adding a fan seems to complicate matters. If I have a single heat source, it should be easier to get rid of the heat from a system standpoint, if we manage to transport the hot air out of the case (while raising temperature locally in the airpath).
Now if I could just get rid of the nagging voice in my head that tells me that there is a fallacy right there...

To be specific, a single 4850 is easy to cool silently with the right heatsink/fan. There's no advantage to running 2 4670s, just greater complexity.

MikeC wrote:There's no advantage to running 2 4670s, just greater complexity.

Er... but there is a advantage to running two 4670s?

• 32W less power draw in idle & 2D
• It's easier to get passive 4670s for those who do want warranties
• Probably less heat in the immediate vicinity of the cards
• Redundancy in the event of failure^^

Or did you just mean heat-wise?

K.Murx wrote:... is it easier to disperse that heat if it is in one place or distributed across multiple devices?

It is much easier to dissipate a specific amount of heat at a single location, because that location will get hotter which helps removing the heat.
If you make the dissipating area small enough, black body radiation will eventually make sure that the correct amount of heat is dissipated.

All this is of course based on the assumption that there's no requirement for actual cooling involved...

/Olle

Olle P wrote:

K.Murx wrote:... is it easier to disperse that heat if it is in one place or distributed across multiple devices?

It is much easier to dissipate a specific amount of heat at a single location, because that location will get hotter which helps removing the heat.
If you make the dissipating area small enough, black body radiation will eventually make sure that the correct amount of heat is dissipated.

All this is of course based on the assumption that there's no requirement for actual cooling involved...

/Olle

I'm sure if we do the maths on this one, it will turn out that the CMOS chip will have long since melted/failed before the correct temperature is achieved to dissipate all heat via black body radiation.

If we are simply talking about ease of cooling, then yes, having all the heat on one card is easier, because you can just stick a massive 4000rpm fan on it and job is done. however the noise will be incredible. as Mike says, cooling a 4850 quietly is no problem with an aftermarket heatsink, however if you wish not to void the warranty will probably be easier and quieter, as the heat density is lower (therefore fans don't have to spin as fast).

It depends. Single source outside my window is much easier than multiple sources in my pc.

But in a crossfire vs single card, I would basically always prefer the single card. My understanding of the algorithms used for multi-gpu real time graphics is that usually alternate frames are rendered on separate cards, which means that while the perceived fps improves a lot, the perceived latency does not and hence you won't get accurate headshots

And that is not taking into account the fact that a single card works in all situations, but sometimes the multi card approach doesn't.

How about getting the new card you mention, when that one comes out?

jaganath wrote:... the CMOS chip will have long since melted/failed before the correct temperature is achieved to dissipate all heat via black body radiation...

You mean it can't handle a couple of thousand degrees?
Go figure...

/Olle

Actually, the heat would never be totally dissipated by Black Body Radiation unless the GPU was perfectly insulated (sealed in a vacuum chamber with power and signals supplied/returned by electromagnetic radiation), no matter what the temperature. And if the radiating surface was large enough, it needn't get that hot.

In general a single heat source is easier to deal with.
Which do you prefer a single large heatsink plus large fan or 5 heatsinks and 5 fans

lm wrote: But in a crossfire vs single card, I would basically always prefer the single card. My understanding of the algorithms used for multi-gpu real time graphics is that usually alternate frames are rendered on separate cards, which means that while the perceived fps improves a lot, the perceived latency does not and hence you won't get accurate headshots

Can you explain that a bit further? I would think the reaction time towards my input depends on the processor, not the GPU - in my naive world, everything what (gameplay wise) happens in a frame is determined well befor it is send off to the GPU, and which GPU renders that frame is irrelevant?

lm wrote: And that is not taking into account the fact that a single card works in all situations, but sometimes the multi card approach doesn't.

Well, those benchmarks in the linked review seem to be quite convincing to me! And anything too old to be supported should not be a problem for a single 4670.

lm wrote:How about getting the new card you mention, when that one comes out?

That is another option. But that depends on whether ATI does get the idle consumption fixed for the 4770.


P.S.: Blackbody radiation? That was a long time ago, in a land far away Now, where are those undergrads we keep to deal with stuff like that?

hi do the et's presume manufacturer-installed passive heatsinks