Separating a video card from a Ninja

Cooling Processors quietly

Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee

Post Reply
nightmorph
Posts: 316
Joined: Thu Aug 10, 2006 11:07 am

Separating a video card from a Ninja

Post by nightmorph » Fri Nov 17, 2006 9:55 pm

I'm using a Gigabyte 7600GT SilentPipeII inside my Solo case. I'm also using a Scythe Ninja on my CPU. Now, when I first built the system, there was maybe a millimeter or two of clearance between the fins on the upper video card HS, and the fins on the Ninja. Now that I've disassembled it to reapply thermal paste and reassembled it, I'm pretty positive the fins of both heatsinks are actually just barely touching.

I figure this can't be good, because all the thermal energy of the video card now has a direct heat path to the CPU heatsink, and I don't want the CPU/GPU to start heating each other up until the magic smoke comes out.

What are my options here? Should I just bend the heatpipe of the video card and lower the upper HS some, or can I just take an insulated thin piece of plastic (one of those metal-cored garbage bag ties, actually) and wedge that in between the heatsinks -- so there would still be a path for heat to flow, but one of higher resistance. . .at least there wouldn't be metal-on-metal contact.

Or should I not worry about it at all -- maybe I'm just misunderstanding the physics of the situation?

Gxcad
Patron of SPCR
Posts: 429
Joined: Sun Aug 11, 2002 3:26 pm
Location: San Francisco, CA
Contact:

Post by Gxcad » Fri Nov 17, 2006 10:50 pm

I personally don't see why you'd need to worry from a thermal standpoint. If the heatsinks were connected, yes there would be heat from the videocard going to the cpu heatsink at a much greater efficiency, but remember if they were touching the size of the heatsink also increases. Well, not exactly but I hope you get where I'm going here... For example, if you have a population (thermal load) and a plot of land (heatsink), and a second population"2" and a second plot of land"2", and they decide to chop down a wall that was between the 2 plots of land, population2 might cross over to land1 and vice versa. Now there is twice as much population (heat) but there is also twice as much land (heatsink), so the population density (thermal load) is still the same. The total population (heat) nor the total amount of land available (heatsink) does not change. Now if one population or one plot of land was bigger or smaller than the other, and one group of the population had lower tolerances for population density, then one population might die, but the GPU and CPU max temperature tolerances and heat created are similar enough that this won't matter.

Lets take heat spreaders as another example (on DIMMs). Essentially, all they are doing is heatsinking 8 sources of heat per side with one heatsink, so that the thermal load is more balanced. RAMBUS use heatspreaders for this reason - one chip tends to get hotter than others, so one part of the heat spreader gets hot, but the other stay cool. Therfore, the heat SPREADS out to the cooler part of the metal, thus leaving less heat concentrated in one spot. Sure, the other cooler chips would run warmer, but as a result the one hot chip will now run cooler. If the heatsinks were seperate, one chip would get very hot and the other chips would stay cool.

My point is, the fact that the heatsinks touch each other doesn't change the total amount of heat they have to get out of the case, nor do they change the amount of metal that is being used to evacuate the heat.

Now going back to your GPU and CPU heatsinks, if they were to be touching and lets assume the heat transfer was very efficient, it would simply be dividing the thermal load (lets also ignore that heat rises). If one got hotter, that would mean the other is running cooler because of it.

In the end, I don't think its anything to worry about, from a thermal standpoint.

Which gets me to my next point...is something bending? Thats more of a concern for me than your thermal situation.

Sorry I went a little overboard there and might have explained it more complicated than necessary and might have ended up confusing more than enlightening. I actually lost myself in my explanations too.

-Ken

nightmorph
Posts: 316
Joined: Thu Aug 10, 2006 11:07 am

Post by nightmorph » Fri Nov 17, 2006 11:14 pm

Er, I think I actually got all that. However, in this case, I think that significantly raising the CPU temp by letting the heatsinks touch metal-to-metal (which would certainly happen, since the video card idles at about 48C) would have more adverse effects than just keeping 'em wedged slightly apart.

Although....the Gigabyte is designed to pull in cool air from the back of the case on the lower PCI slot (double slot card), transfer it via heatpipe to the heatsink on the upper side of the card, and from there to the fins on that top side, and THAT heat is supposed to be carried up and out the back of the case by the exhaust and PSU fans....so maybe that would just be doing an even quicker job of it, if they touched? :D

I still don't like the idea of getting my CPU any warmer than it has to be, though.

Gxcad
Patron of SPCR
Posts: 429
Joined: Sun Aug 11, 2002 3:26 pm
Location: San Francisco, CA
Contact:

Post by Gxcad » Fri Nov 17, 2006 11:20 pm

The other thing about them touching is that the heat transfer WON'T be that efficient. although they will now be physically touching, there is no flat surfaces or thermal compound to assist in the transfer. In the end the difference will be negligible, if even that.

At least I think so.

Hey I got an idea. Try both and measure temps! (seriously, it'll be interesting).

-Ken

nightmorph
Posts: 316
Joined: Thu Aug 10, 2006 11:07 am

Post by nightmorph » Sat Nov 18, 2006 12:07 am

Hmm, measuring could be interesting. But here's another problem: what about electrical conductivity between the heatsinks? I'm more worried about shorting stuff out than raising temps, when it comes down to it!

If I can be assured I wouldn't zap my system, then I'd have no problem testing with each solution.

Edit: It boots! I've got the little twist tie in there right now, so here goes...

Alright, idle temps with C'n'Q enabled, operating at 1Ghz (lowest):
CPU: 37.5-38.5C
Sys: 44C
Graphics: 48C

No joy. Looks like redoing the thermal grease didn't do anything, and/or once again, the HS just slid around too much when I was trying to clamp it down, ruining any nice, thin, even layer of grease there may have been. I nudged the vcore down to 1.200V in BIOS (down from 1.250), but that doesn't seem to have had any noticeable effect. Ah, well, I never expected to do much undervolting anyway.

jaganath
Posts: 5085
Joined: Tue Sep 20, 2005 6:55 am
Location: UK

Post by jaganath » Sat Nov 18, 2006 1:48 am

My point is, the fact that the heatsinks touch each other doesn't change the total amount of heat they have to get out of the case, nor do they change the amount of metal that is being used to evacuate the heat.
Yes, but metal-on-metal contact is infinitely more thermally conductive than metal-air-metal, and heat moves from the higher-temperature source (GPU) to lower-temperature sink (CPU). So you are effectively heating up your CPU. I don't think the population and land analogy was that helpful, actually.
what about electrical conductivity between the heatsinks? I'm more worried about shorting stuff out than raising temps, when it comes down to it!
The heatsinks are not live; most TIM is deliberately a poor (electrical) conductor.

c10000
Posts: 13
Joined: Fri Aug 20, 2004 6:44 pm

Post by c10000 » Tue Nov 21, 2006 9:48 am

Hi, nightmorph,

What mobo and CPU do you have?

I am planning to build a system based on the same case and video card. The rest of the system will be P5B-E, E6400, and Ninja. So, I am wondering if I'll get into similar situations.

After all this, do you still recommend the same combination (Ninja + Gigabyte 7600GT)?

Charles

nightmorph
Posts: 316
Joined: Thu Aug 10, 2006 11:07 am

Post by nightmorph » Tue Nov 21, 2006 10:17 am

Look around on these forums for the P5B thread; apparently that motherboard might be incompatible with the Ninja just because of its other features. Otherwise, if it fits, you shouldn't have the same spacing problems I do. I think this is more of a socket AM2 problem, just because of where the processor tends to be placed in relation to the first PCIe x16 slot.

I definitely recommend both the Ninja and the Gigabyte card. However, my system is built around an MSI K9N Platinum Ultra 'board with an Athlon64 X2 CPU, so . . . :)

c10000
Posts: 13
Joined: Fri Aug 20, 2004 6:44 pm

Post by c10000 » Tue Nov 21, 2006 12:06 pm

Thank you for the information.

Yes, I am aware of those threads about the Ninja Rev B + P5B issue. But someone also suggested using Sythe's Universal Retention Kit with Ninja Rev B, which may make it a better fit on the P5B.

I am just making sure that there is not yet another issue with Ninja too close to the video card.

nightmorph
Posts: 316
Joined: Thu Aug 10, 2006 11:07 am

Post by nightmorph » Tue Nov 21, 2006 12:27 pm

Intel motherboards tend to place rather more space between the CPU socket and the first of the expansion slots. Check around the forums for pictures of your board with the Ninja mounted on it. Unless there's only about 1 PCIe slots's space between the heatsink and your graphics card slot, you'll be fine. On my board, the CPU socket is placed relatively low, and the first PCIe x16 slot is pretty high up. Normal cards would be just fine, but it's the wraparound heatsink on the Gigabyte card that causes the problems. The upper heatsink effectively makes it take up three slots.

Post Reply