Cooling Technologies & Ideas (Thread Split)

[DoFoT9]

You should in the interim, while you get some paste in hand and get it applied.

argh ive just started uni and everything! i need it as much as anything now! i might just resort to using 'non-intensive' apps, no video converting ok

How are the laws for consumer rights where you are?
That is, do they favor corporations (like they do in the US), or the consumer?
(I'm under the impression the UK has a better outlook for consumers as an example). Unless I totally misread a few posts in other threads.

tbh i havent the slightest idea who they favour (UK is different to Australia just incase you didnt know that haha). its out of warranty by a big lot, so i assume their help would be NULL.

In what sense?

Isn't the chip (since soldered to the board), and the cooler stock?

im confused by your confusion. what do you mean??

 

[nanofrog]

argh ive just started uni and everything! i need it as much as anything now! i might just resort to using 'non-intensive' apps, no video converting ok

Get your hands on some paste quick. 15 - 30min and you're done.

tbh i havent the slightest idea who they favour (UK is different to Australia just incase you didnt know that haha). its out of warranty by a big lot, so i assume their help would be NULL.

You mean it's not next to Ireland!

Seriously, I wasn't sure if the legal system mirrored (somewhat) that of the UK in terms of consumer advocacy. It sucks here in the US. If the DRM of 1998 isn't a clue, then I don't know what is.

im confused by your confusion. what do you mean??

From the comment you made, I thought you had modded the system, and I was wondering how you equated that would make any difference to the CPU and cooler.

Especially since the chip is soldered, and the cooler is custom to the case.

Anyway, I've found an article on cooling that might be interesting. It has a nice flow of information (limits of air, and some directions for solutions). Please note, this is another aerospace/military applications article, but is quite relevant to the OT.

 

[DoFoT9]

Get your hands on some paste quick. 15 - 30min and you're done.

i will! wont be for a few days though, because the way that everything is over here (i.e. lack of IT stores) i will have to order something in.

You mean it's not next to Ireland!

haha no.. sort of a tiny bit further south

Seriously, I wasn't sure if the legal system mirrored (somewhat) that of the UK in terms of consumer advocacy. It sucks here in the US. If the DRM of 1998 isn't a clue, then I don't know what is.

well it used to, some things have changed but many many things are still the same. being IT and the fact that nobody cares im sure that they are identical legislation lol. but im in IT, not law, so i couldnt tell you..

From the comment you made, I thought you had modded the system, and I was wondering how you equated that would make any difference to the CPU and cooler.

Especially since the chip is soldered, and the cooler is custom to the case.

oh right!! sorry sorry no no its ALL custom!! i was just referring to the fact that my CoreDuo is more inferior to most peoples Core2Duo chips because of the differences yada yada, and probably why its THAT much hotter then the C2D (my gf's machine only hits 70°C under load for crying out loud!).

Anyway, I've found an article on cooling that might be interesting. It has a nice flow of information (limits of air, and some directions for solutions). Please note, this is another aerospace/military applications article, but is quite relevant to the OT.

question, would gas cooling be some sort of idea? like an air-con for your laptop! or isnt that very realistic?

 

[nanofrog]

question, would gas cooling be some sort of idea? like an air-con for your laptop! or isnt that very realistic?

Well, the idea of using an A/C system has already been done, and a couple of pics of existing products are a few posts back.

For laptops, it will be much harder, due to the size constraints and power limitations. So if it ever happens, it will take longer to get it small enough (and actually move enough heat), without running down the battery in a few minutes.

I can imagine a world where the power (performance) differences will become more of a gulf again between laptops and desktop systems, unless the total power never exceeds what's already being consumed. This is the current direction of chip makers on the mobile parts. Try to squeeze out increased performance for the same, or prefferably less power.

At some point however, it's going to hit the wall without a major technology change. Optical systems would be the solution, but it's slow in coming. Silicon's going to be milked a bit longer. We're at 45nm now, 32nm next year, and it can go down to 10nm (proof of concept parts work). So figure 1 year for 32nm, 2 for 22nm,... 16nm, down to 10nm for sure, following Intel's recent strategy as historical precedent. A grand total of 7yrs left, if it can't be pushed lower. But the sub 10nm realm is being researched as we speak. And the development cycle of die shrinks could be altered (lengthened) intentionally to buy additional time (assuming there's no technical difficulties implementing each shrink on a massive scale).

 

[DoFoT9]

Well, the idea of using an A/C system has already been done, and a couple of pics of existing products are a few posts back.

ah, its all too technical for me.. im just a dumb nublet.

For laptops, it will be much harder, due to the size constraints and power limitations. So if it ever happens, it will take longer to get it small enough (and actually move enough heat), without running down the battery in a few minutes.

thats alright my MBP only lasts 2hrs at the most anyway

I can imagine a world where the power (performance) differences will become more of a gulf again between laptops and desktop systems, unless the total power never exceeds what's already being consumed. This is the current direction of chip makers on the mobile parts. Try to squeeze out increased performance for the same, or prefferably less power.

i can see it too, but how far away is it?? why would people manufacture these items when making the same speed/powered device can be 50 times bigger but cost 100 times less to manufacture?

At some point however, it's going to hit the wall without a major technology change. Optical systems would be the solution, but it's slow in coming. Silicon's going to be milked a bit longer. We're at 45nm now, 32nm next year, and it can go down to 10nm (proof of concept parts work). So figure 1 year for 32nm, 2 for 22nm,... 16nm, down to 10nm for sure, following Intel's recent strategy as historical precedent. A grand total of 7yrs left, if it can't be pushed lower. But the sub 10nm realm is being researched as we speak. And the development cycle of die shrinks could be altered (lengthened) intentionally to buy additional time (assuming there's no technical difficulties implementing each shrink on a massive scale).

are they REALLY going to push silicon to that level?? is it even worth it?? take the advance from 45nm to 32nm... what benefits are there?? 5% more this 10% less that... thats hardly an improvement.. well i dont really see it as one anyway. stupid throw away society!

 

[Tesselator]

Here we go:

http://www.youtube.com/watch?v=ioCZojN4A0g

Seems the guys at IBM are thinking just like us... Is this the 100 monkey nexus at work or what?




That said I wonder if I can install OS X 10.6 server on this puppy: http://www.youtube.com/watch?v=5Iz00612xTM&NR=1

 

[nanofrog]

Here we go:

http://www.youtube.com/watch?v=ioCZojN4A0g

Seems the guys at IMB are thinking just like us... Is this the 100 monkey nexus at work or what?

I'd be interested in seeing how they made the blocks. I'm wondering if they went to the extent of sputtering the layers together directly on die (no TIM). And if so, how'd they manage it cost wise.

That said I wonder if I can install OS X 10.6 server on this puppy: http://www.youtube.com/watch?v=5Iz00612xTM&NR=1

Noisy, but nice.

Of course you can build one, and it might even be easier if you don't need it to be portable. Though it's version of portable seems like a BIG step backwards (bigger than the first suit case portables).

 

[DoFoT9]

im liking that supercomputer what was that component that allowed for low latency communications called? i would like to look into that!

 

[nanofrog]

im liking that supercomputer what was that component that allowed for low latency communications called? i would like to look into that!

Infiniband.

Sweet, but expensive.

 

[DoFoT9]

Infiniband.

Sweet, but expensive.

ahh cool. how does it have such low latency times? probably as simplistic as you can make them right?

 

[nanofrog]

ahh cool. how does it have such low latency times? probably as simplistic as you can make them right?

That's part of it (direct, bi-directional serial links). The main part ends up being you can "bond" the links for additional bandwidth (run links in parallel). Sort of like Port Teaming with NIC's.

Here's the wiki on it, if you're interested.

 

[DoFoT9]

That's part of it (direct, bi-directional serial links). The main part ends up being you can "bond" the links for additional bandwidth (run links in parallel). Sort of like Port Teaming with NIC's.

Here's the wiki on it, if you're interested.

i just read it. very interesting stuff, i never knew such a thing existed and i always wondered how they connected each node in a supercomputer. *drool*

 

[Tesselator]

ahh cool. how does it have such low latency times? probably as simplistic as you can make them right?

Short high-speed component to component connections within the unit, high speed high quality components. Short high speed system level interconnections (motherboard to hub/router to motherboard) - probably on a 10GBASE - Something line (I think Infiniband is copper tho, and not fiber so 10GBASE-CX4 maybe?).

 

[DoFoT9]

Short high-speed component to component connections within the unit, high speed high quality components. Short high speed system level connections (motherboard to hub/router to motherboard) - probably on a 10GBASE - Something line (I think Infiniband is copper tho, and not fiber).

interesting stuff. so many questions, not enough time!! (must...finish...uni....work.... ).

would love to learn more though, as i hope to get into networking etc. its all very interesting.

could this sort of system be implemented in large buildings to eliminate the need for computer nodes at the worker end? i.e. we go back to terminals etc.? most people only run office and stuff these days, hardly uses any CPU time at all.

 

[nanofrog]

Short high-speed component to component connections within the unit, high speed high quality components. Short high speed system level interconnections (motherboard to hub/router to motherboard) - probably on a 10GBASE - Something line (I think Infiniband is copper tho, and not fiber so 10GBASE-CX4 maybe?).

It's a switched fabric like Fibre Channel, but it is copper, not FO cable.

For a 12x connection, it's max is 96Gb/s (12GB/s = Hell YES!).

 

[DoFoT9]

It's a switched fabric like Fibre Channel, but it is copper, not FO cable.

For a 12x connection, it's max is 96Gb/s (12GB/s = Hell YES!).

so that is for all nodes inside the computer right? or just for each CPU to talk to other CPUs? or is it for all, e.g. storage has a max of 12GB/s, CPUs max is 12GB/s, etcetc...

 

[nanofrog]

so that is for all nodes inside the computer right? or just for each CPU to talk to other CPUs? or is it for all, e.g. storage has a max of 12GB/s, CPUs max is 12GB/s, etcetc...

Both. It works for processor nodes and high performance I/O.

 

[DoFoT9]

Both. It works for processor nodes and high performance I/O.

ahh ok thats great

are they common in most businesses? or is it more for the super-computer top500 sorta area only?

 

[nanofrog]

are they common in most businesses? or is it more for the super-computer top500 sorta area only?

No, only where such throughputs are needed, so typically very specialized, and well funded operations.

Think Clusters used for real time research data (NASA, Pharmacuticals,...), Telecommunications, Banks (especially something like the US Federal Reserve),....

Definitely not your SOHO/SMB operations.

 

[DoFoT9]

No, only where such throughputs are needed, so typically very specialized, and well funded operations.

Think Clusters used for real time research data (NASA, Pharmacuticals,...), Telecommunications, Banks (especially something like the US Federal Reserve),....

fair enough really. i can imagine that it costs a good penny (or a squillion)

Definitely not your SOHO/SMB operations.

hmm i see. fair enough point though normal clusters would be enough for most businesses, just not the high end stuff.

wish i had me one lol

 

[nanofrog]

fair enough really. i can imagine that it costs a good penny (or a squillion)



hmm i see. fair enough point though normal clusters would be enough for most businesses, just not the high end stuff.

wish i had me one lol

If you don't need that much speed, or want to scale over time, then Fibre Channel is a good option (1, 2, 4, 8, 10, 20Gb/s options). You can run into an issue with 10 & 20Gb/s (derivative of 10) though, as it's not backwards compatible with 8 & down.

 

[Tesselator]

interesting stuff. so many questions, not enough time!! (must...finish...uni....work.... ).

would love to learn more though, as i hope to get into networking etc. its all very interesting.

could this sort of system be implemented in large buildings to eliminate the need for computer nodes at the worker end? i.e. we go back to terminals etc.? most people only run office and stuff these days, hardly uses any CPU time at all.

If I understand your question right that is exactly the purpose of heavy metal. (main-frame blade servers or like this mini on rollers I linked.) The basic idea is to put your blades in a cluster and then divide the total into virtual allocations.

Of course I want to allocate ALL of the resources to just one virtual machine - all for me...

 

[DoFoT9]

If you don't need that much speed, or want to scale over time, then Fibre Channel is a good option (1, 2, 4, 8, 10, 20Gb/s options). You can run into an issue with 10 & 20Gb/s (derivative of 10) though, as it's not backwards compatible with 8 & down.

hmm i see. but not having the Infiniband means that latency will greatly be increased (plus a number of other factors such as the whole parallel thing). anyway, im sure both is faster then what i have now

If I understand your question right that is exactly the purpose of heavy metal. (main-frame blade servers or like this mini on rollers I linked.) The basic idea is to put your blades in a cluster and then divide the total into virtual allocations.

yes thats correct. is virtual the only way to go? could the users "remote desktop" into the machine (is it possible to do that?) or would they still need to be virtual machines? (this would make the machines fast, but the images depend on the link speed) that would be very cool to do. i wonder if i could try it at home

Of course I want to allocate ALL of the resources to just one virtual machine - all for me...

all for Q3e, right?

 

[Tesselator]

yes thats correct. is virtual the only way to go? could the users "remote desktop" into the machine (is it possible to do that?) or would they still need to be virtual machines? (this would make the machines fast, but the images depend on the link speed) that would be very cool to do. i wonder if i could try it at home

Yeah, there are many many ways to configure or operate such a system. For multi-user in your office building example that's probably what you would wanna do.

all for Q3e, right?

Hells yeah!






.

 

[nanofrog]

If I understand your question right that is exactly the purpose of heavy metal. (main-frame blade servers or like this mini on rollers I linked.) The basic idea is to put your blades in a cluster and then divide the total into virtual allocations.

Of course I want to allocate ALL of the resources to just one virtual machine - all for me...

That's how to make a proper T-CAD system. <drool>

hmm i see. but not having the Infiniband means that latency will greatly be increased (plus a number of other factors such as the whole parallel thing). anyway, im sure both is faster then what i have now

Latency isn't that bad. It's still a switched fabric implementation (FC-SW), not hierarchal, like Ethernet. That helps reduce latency, and it's certainly faster than anything you've got now.

The real question is, can you actually utilize it.

yes thats correct. is virtual the only way to go? could the users "remote desktop" into the machine (is it possible to do that?) or would they still need to be virtual machines? (this would make the machines fast, but the images depend on the link speed) that would be very cool to do. i wonder if i could try it at home

Virtual is the primary method intended, as that's what's done for Cloud Computing (cheap, and aimed at consumers, and businesses alike). It's not my favorite though, as I don't like the idea of not having "possession" of my data.

You could remote into it, if it's setup as a powerful VPN server. But you'd have to own all the equipment, and need adequate bandwidth. As it's currently limited from anything close to getting a single access point even close to full utilization. Especially on the consumer end. The ISP infrastructure just isn't there yet.

all for Q3e, right?

What a waste!