Monitoring, loop tuning and expansion

[Owenator]

Wonder what the plausibility of making a larger core on the same architecture is...basically increase the surface area so that you can cool more efficiently...

Probably not too high because they are trying to shrink everything to speed it up. At least I think that is why the die (core) shrinks make for faster chips. Something to do with a shorter path. But I'm not very knowledgeable in that area.

What I'd like to see is an integrated top that is made of copper and thick enough to act as a good heat spreader. I wonder if it would be possible to remove the standard one and solder a new one on? But probably easier to just buy higher end water or even a phase change system.

 

[m0r7if3r]

The reason they're making it smaller is to do with transistor characteristics and the rate at which transistors are able to switch (that's my understanding at least). But the idea I was toying with was using the same architecture (and thus, same transistor size) but space it out some so the heat is a)less concentrated and b)more spread for a larger contact area with whatever you're using to remove heat. It would be easier to buy a high end water system, but then what happens when you employ these ideas with that system? You'd still make gains.

I don't think that soldering a new heatspreader would be possible cause I think that it would overheat the cpu. I'm not sure how intel does it, but the fact that the solder they use is removable with only a lighter means it's gotta have a pretty low melting point. I also don't know how they get the solder to stick to the top of the chip (its not metal is it?), but its probably one of those one time things. If you could get a chip that hadn't had an ihs put on yet and some of that solder you might be able to pull it off, but I don't know the practicality of that. If someone has an in with AMD or intel and can offer some sort of insight to the process that would be great...

 

[Elfie]

Is an Intel chip soldered into place? I thought the copper IHS was glued with a form of epoxy.

 

[m0r7if3r]

1366 are all soldered (to my knowledge) its the 32nm architectures that use the paste (and I don't think its all of them). The best thermal contact would be solder though.

 

[QuietIce]

The reason they're making it smaller is to do with transistor characteristics and the rate at which transistors are able to switch (that's my understanding at least). But the idea I was toying with was using the same architecture (and thus, same transistor size) but space it out some so the heat is a)less concentrated and b)more spread for a larger contact area with whatever you're using to remove heat. It would be easier to buy a high end water system, but then what happens when you employ these ideas with that system? You'd still make gains.

You might gain in shedding heat but the manufacturers would loose a LOT from making the die itself. While die shrinks do incorporate the smaller transistor size it also allows more CPUs per wafer, which translates directly into more profit for the manufacturer. Make the die bigger and the manufacturing cost rises. I don't think you'll convince anyone to give up their profit margin for better cooling when the CPUs already run fine with the POS stock coolers - at stock speeds ...

 

[m0r7if3r]

It would be like an eVGA ftw edition card, it would come at a price premium and be targeted at the enthusiast market specifically. They don't need to make many of them and they could probably make more on them than they do on the actual part (have you SEEN the official "What i've killed" thread in the overclocking team's forum? Between them, I wouldn't be surprised if they'd topped $100k of their own money in burned out parts.)...I think if there were a demand for it the product would exist, that's how a market economy works.

 

[QuietIce]

It would be like an eVGA ftw edition card, it would come at a price premium and be targeted at the enthusiast market specifically. They don't need to make many of them and they could probably make more on them than they do on the actual part (have you SEEN the official "What i've killed" thread in the overclocking team's forum? Between them, I wouldn't be surprised if they'd topped $100k of their own money in burned out parts.)...I think if there were a demand for it the product would exist, that's how a market economy works.

I don't know what the FTW is or what makes it special but unless the chip itself is a one-off I don't see the similarity. Even if the transistors are kept the same size you'll still have to design the CPU from scratch, make the photographic plates for it, re-tool as needed to wire the chip into it's base, and all the other little things in between and you'd be doing all this only to increase the heat transfer area of the chip. In the end it would probably be cheaper to go phase-change cooling ...

 

[m0r7if3r]

Yea, but that's not the point of this thread...we're postulating on ways of extracting every last drop of performance from an average watercooling setup. We're talking in practical and impractical terms, theoretical and real world, its about getting ideas out there, collaborating, and just generally trying to get some good ideas of ways to improve the efficiency of a loop. That said, we have focused an awful lot on just getting heat off a chip, we should examine the rest of the loop...maybe go one step down and consider ways of efficiently getting it out of the block. I wonder how a block that had barbs on the bottom and a tower cooler on top would perform...you could run some ultra-low speed fans on it. It'd essentially be a hybrid between air and water targeted primarily at increasing the effectivness of the watercooling...it would be a HUGE pain to mount though...

 

[evolvedpc]

Actually, smaller chips are not only cheaper, they theoretically run cooler - that is part of the point. I'm not sure exactly why, but pretty much every tech news article quotes less power consumption (and therefore less heat) as a key advantage of smaller die sizes.

Of course, chip makers take that extra TDP headroom and use it up by cramming more transistors onto their chips and then design the spreaders with stock clocks in mind, which don't perform very well when you push the chip beyond its design limits. I suspect that it is the IHS, rather than the smaller die size, that is our problem.

I originally started this thread to discuss ways that we could use monitoring data to guide us in tweaking our loops. It has become somewhat sidetracked. (Not that there is anything wrong with that.) However, IMHO this thread would be more useful if we concentrated on stuff that we have some hope of actually achieving. Popping the top on a CPU and soldering the block directly to the chip is doable, but I doubt that we will convince chip makers to change their ways. :-(

 

[m0r7if3r]

Dieshrinks occur because of smaller transistors, if you could take the same smaller transistors, with the same thermal footprint, and spread them out, your w/in would decrease and you would have more surface area that you could dissipate the heat over...or that's how i figure at least.

 

[Bobnova]

Smaller transistors on a large slab of silicon would almost certainly run cooler, but silicon is brutally expensive, a larger area means more money per die, and may or may not make the product more attractive.

For my benching i'd buy the cheap normal part, i expect that all the other dry ice/ln2 people would as well. Once you hit that level the thermal properties of the chip itself aren't especially relevant, with a good pot design and plenty of ln2 you can get anything cold.
The vast majority of stuff like the FTW motherboard is aimed at people who want the snazzy stuff, but aren't extreme benchers (If i remember correctly the P55 FTW kills cpus at extreme temps anyway). All the "binned for overclocking" GPUs you see available are binned for air OCing, most don't work very well at all for extreme measures.

Now for air OCing, it'd likely be great.

 

[evolvedpc]

I'm not so sure that's how it really works. Spreading out the transistors means further for the current to travel. This is a limiting factor for processor speed, these days. It also requires more power to push the electrons down a longer path. More power, more heat. Back to square one. :-(

ADK

 

[m0r7if3r]

electricity travels at the speed of light, and while you are correct that the connective paths would offer more resistance (since they aren't superconductors), ime, this added resistance is so slight that it can be completely disregarded. That said, I've never done anything on the scale of a processor, but I don't imagine that a connective path would dissipate more heat than a transistor, since the efficiency is better, so while your total thermal load might increase slightly, you would still be decreasing your w/in^2

 

[QuietIce]

I just saw the Indigo Xtreme installation video on YouTube and remembered this post:

What we really need is a way to increase the spread of the heat from the CPU to the waterblock. I read about a company that was making what looked like a flat copper plate but was actually hollow and full of fluid. The idea beign that the heat spread very fast from athe cpu die to the entire surface making it easier to cool. It works because water vaporizing removes heat very fast. Then it moves around to spread the heat. Quite a "cool" idea. Not sure if they sell them yet. http://www.novelconceptsinc.com/heat-spreaders.htm

Is it possible the Indigo Xtreme is their implementation? It looks/sounds kind of like the surface/pad idea you linked ...

 

[m0r7if3r]

I think the surface pad idea is as a replacement, not an auxillary, that was the vibe i got from it at least...

 

[Conumdrum]

I just saw the Indigo Xtreme installation video on YouTube and remembered this post:

Is it possible the Indigo Xtreme is their implementation? It looks/sounds kind of like the surface/pad idea you linked ...

Not at all. Skinnee used MX-2 for ALL tests, so probability he used the same for Vregs.

And Indego is sized specifically for a CPU ONLY. 775 OR 1366 chips.

 

[QuietIce]

?????

 

[m0r7if3r]

I can only assume something went wrong in the forum...

 

[Conumdrum]

Oopsie! Sorry, got mixed up on posts.

Indego is good stuff!