Yet another bulging-capacitors replacement

Hell yes it matters. I'm going to fix a year old Coolmax 650 watt PC PSU and will be looking for some replacement caps. I like to keep a spare and I need 650 with this new AMD 120 watt quad core PhenomII 3.2 ghz CPU and Asus M4A78E-T mobo. With Asus overclocking friendly special settings I'm able to run it at 4.0 ghz for each core. Makes video encoding on an application supporting multicore encoding really fly. Not unusual to get over 350 frames/sec out of NTSC 740x480 avi's. I can make a high quality

20 chapter DVD with all the bells and whistles in about an hour. Used to take 24 hours on a 2ghz single core AMD :)

--
Live Fast, Die Young and Leave a Pretty Corpse

Just as a matter of interest Meat, what is your preferred brand and type of heatsink goop when working with these very high power processors? I've recently been working with some games machines that have two very powerful processors on the board, and have been having some thermal issues when using 'standard' white silicon grease on them, which appears to be what the manufacturer used originally. I have today reassembled one using some Arctic Silver compound instead, and it seems to be doing a fine job. I have always resisted using this stuff, because it's so messy, and so hard to remove unless you use the complementary cleaner, but if it really is that much more effective, then I might be prepared to live with these shortcomings. Anyone else got any constructive comments on the subject of thermal interfacing of coolers to high power chips ?

Arfa

In a past life, I used to design HF SSB marine radios. The typical transmitter was Class AB 150 watts with about 30% efficiency. That's two devices, dissipating about 125 watts each, over an area of about

70 sq-cm. Oh yes, no fan allowed.

This is quite a bit more dissipation than the average desktop, causing some things to be more critical. In the process of getting it to work, I learned a few things.

1. The less silicon grease used, the better. The idea behind silicon grease is to fill in the gaps, scratches, and gouges in the power transistor base and aluminum heat sink. Cross sectional microphotographs show metal to metal contact on the peaks, but huge gaps, filled with silicon grease, in between. Under ideal circumstances, maximum metal to metal contact, with minimum gaps is the target practice.

1. All heat sinks and transistor bases are NOT flat. I made a dramatic improvement to the measured thermal resistance by polishing flat the base of the xsistor and the face of the heat sink. That meant removing the gold from the copper base, but that's what was necessary. I used a Moire pattern to measure flatness. A mirror finish was best, but difficult to achieve. To prevent corrosion, I plated the exposed copper with electroless tin or silver. For the aluminum heat sink, I just used abrasive polish and a glass polishing plate to obtain a mirror finish and flat surface.

2. Compression pressure is important. None of the standard spring clip CPU heat sink holders come even close to optimum. Compression adjusts for the bends, and also provides some level of galling to provide metal to metal contact. If done correctly, adding silicon grease actually increases the thermal resistance. However, this is difficult to do with a CPU that has components on the bottom side, thus preventing compression. Applying pressure only on the top center of the CPU, will cause the substrate to bend, and eventually break. I have some ideas, but nothing that can be retrofitted to an existing motherboard and CPU socket. This is close, but not optimum:

Note the comments on base finish and flatness.

So, if you want the best head sinking, polish flat the CPU top (removing all the laser scribbled markings, polish the heat sink face, use very very very very little silicon grease, and compress the sandwich until it nearly breaks the CPU.

--
Jeff Liebermann     jeffl@cruzio.com
150 Felker St #D    http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann     AE6KS    831-336-2558

of

Any opinions on silipads? From my limited trials mica sheet and absolute minimal white grease has better thermal transfer. Emphasis again on minimal. Having to decide this week whether to spend out on 6 MJ series TO3 devices in an old Carver amp failed due to one of the driver TO3 being pushed through a heap of white grease , so grease on pins getting inside the TO3 socket housing , so insulating partially and eventually burning up the pin .

On games PC's i've gone to water cooling, blissfully quiet (apart from the fishtank type noise!), CPU and GPU a few degrees above room temp even at full load. Never go back to jet engine graphics card fans. JC

You answered your own question. The AMD heatsink / quad core PhenomII 955 Black Edition package comes with Artic Silver already applied. I'm using an Antec server case that has a hole and tube in the side cover. The end of the tube fits directly over the CPU heat sink so it draws air directly from the outside. In back is a pair of 120mm fans controlled by the mainboard. If the CPU temp goes up all three fans increase according to the temp. Or you can set them to run at full speed all the time. The 650 watt PSU also has a temp sensing 120mm fan. So the box is really quiet most of the time. But when rendering video and the CPU usage hovers around 50% fan speed increases slightly. Video rendering with an application that takes advantage of multi-core processors seem to use the most CPU percentage. I've never seen it go over 50%. Most of the time it doesn't go over 10%.

--
Live Fast, Die Young and Leave a Pretty Corpse

I've seen pads used in a lot of high power amps. Soundcraftsmen, BGW, Carver, Peavey, Crown all used them at some time or another. If the devices are torqued down properly I don't have a problem with them. You are talking about the rubberized pads with embedded compound right?

--
Live Fast, Die Young and Leave a Pretty Corpse

Do the math. Silicon impregnated rubber TO-3 pads have a thermal resistance of about 0.4K/watt. 0.0002" Mica, with silicon grease smeared on both sides is about 0.6K/watt. Depending on your total power dissipation, that's hardly any difference. However, if your heat sink is undersized, buried inside a cabinet, or located in a place where there's no air flow, it might make a difference.

See Page 6-8. The example shown is for a TO-3 packaged device.

Keep it stock. However, if this is a push-pull type of amp, where the thermal balance of the xsistors has an effect on the bias point, crossover distortion, and possibly linearity, I would make sure that whatever you do to one side, the same gets done to the other.

--
Jeff Liebermann     jeffl@cruzio.com
150 Felker St #D    http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann     AE6KS    831-336-2558

Thanks for the insights Jeff. All interesting stuff. These are dedicated games machines, not based on a PC in any way. The power supply is specced to deliver 12v at 23 amps, yes, that's twenty three amps ...

Almost all of this is potentially going into these two processors, so not far off 300 watts between them. No mean task shifting the heat off them !

Arfa

Yes, seems to be 'busy' video rendering that causes all the problems on the machines I am working on. Q & D calcs show that at max chat, the two processors are potentially using close to 300 watts of input power between them, and the heat that this generates in them, takes some shifting ...

Arfa

"Arfa Daily" wrote in news:dgZdo.79946$Pa3.38201@hurricane:

what processors(microprocessors?) run at 12V? ISTR that today's uPs run mostly on 3.3V

Most other digital logic runs at 5V,I believe.

I think you wil find that most of your power is going into the video drive (or LCD backlight) circuitry.

--
Jim Yanik
jyanik
at
localnet
dot com

There was quite a bit of "Learn By Destroying(tm)" involved. Measuring flatness and thermal resistance were a major exercise, but settled all kinds of lab arguments.

Incidentally, you might be interested in how Arctic Silver works.

Silver has a much higher thermal conductivity (410 W/m*K) as compared to zinc oxide (21 W/m*K) and aluminum oxide (30 W/m*K) which is what's in common thermal compound.

However, if your shove an ohms-guesser into a puddle of Arctic Silver, it's not conductive. That's because the particles of silver are so far and few, that the bulk of the solution is polyolester or mineral oil, which insulates the particles from each other, preventing mutual contact. However, if you tear apart a CPU/heatsink that's been used for a while, you'll notice that the Arctic Silver is a thick and dense paste which is conductive. What has happened is that the polyol ester mixture has evaporated sufficiently to provide contact between particles. Since thermal conductivity is best through the silver particles, the result is a superior thermal connection, with the sliver particles filling the voids. You could do the same thing with silver dust, but it would difficult to handle and apply. Meanwhile, ordinary silicon grease does the same thing, but there's a difference. The oil does not evaporate as easily, and the ceramic particles are much larger and less compressible than the silver particles. Fewer points of contact and lower thermal conductivity of ceramic, means a worse thermal connection.

I don't believe it. The winner of the power hogging consumer CPU contest was the DEC/Intel Alpha 21364 (EV79):

which burned 155 watts. Itanium II came close with 130 watts (per core). I had an Alpha CPU machine to play with for a while, which would burn my hand from the hot air coming out the back.

If you have a power line wattmeter or a Kill-A-Watt meter, I think a measurement would be helpful.

--
Jeff Liebermann     jeffl@cruzio.com
150 Felker St #D    http://www.LearnByDestroying.com
Santa Cruz CA 95060 http://802.11junk.com
Skype: JeffLiebermann     AE6KS    831-336-2558

Er no. There are no backlights. Or display processor. These are X-Box / Playstation type boxes. Apart from some support circuitry in IC form - which admittedly does gobble enough power to make it run hot enough that a degree of heatsinking to the pcb shielding via thermal pads is required - everything goes on in a pair of very large BGA processors, one of which is the data processing engine, and the other of which is the graphics processing engine. It is they which make use of the 12v, and they which gobble the amps from it ...

The power supply does have other outputs, but these are all at very low current availabilities, so will be for support logic and maybe some core supplies for the two processors. Trust me when I say that the two big chips is where all the power is going, and generating heat that needs shifting :-)

Arfa

Again, more interesting stuff Jeff. As to the power consumption of these chips, see my reply to Jim above. Also, it is split between two chips, it's actually not quite as much as 300 watts, and will of course be an 'on demand' thing, depending on how hard the chips are being asked to work by the processing task that's happening at the time, so the 23 amps is only a worst case potential input current. However, that said, these two chips do produce *very* considerable heat even when idling to produce nothing more than the splash screen.

Arfa

"Arfa Daily" wrote in news:O24eo.36157$r24.2988@hurricane:

Odd that BGA processors are using 12V instead of logic level voltages.

I'm surprised they don't use some sort of liquid or heat-pipe plumbing to remove all that heat. Wasn't it the CRAYs that used liquid Freon to flood the processor cabinet to dissipate al the heat built up?

--
Jim Yanik
jyanik
at
localnet
dot com

Then they would need around 100A at 3.3 volts. The voltage drop would be a big problem. I'm sure there is a DC to DC converter near the chip, like used on computer motherboards.

--
Politicians should only get paid if the budget is balanced, and there is
enough left over to pay them.

Most run at 12. The core at 1.6. Both AMD and Intel boards have a 4 pin Molex plug near the CPU for direct 12v from the PSU.

--
Live Fast, Die Young and Leave a Pretty Corpse

Yes, I was amazed when I read the PSU specs. One of the versions is actually specced 12v at 32 amps !! The PSU plugs directly onto the board via a pair of brass pins as thick as those on a power cord for a kettle. The heat is removed via a pair of flat plates that are connected to a network of sealed copper pipes, a bit like you see on some Technics amps. I've no idea what is inside those pipes. This whole assembly is cooled by a centrifugal fan that idles at a very low speed. The heatsinks have to get up to blisteringly hot before the processor thinks that it might be a good idea to ramp up the speed of the fan a bit. I guess they have done this to try and keep the thing quiet, but personally, I think it is a really poor bit of design. I am looking at ways to make the fan idle faster, without compromising the auto ramp up beyond that, when the processor deems it necessary, but unfortunately, it's not quite as easy as a simple 'analogue OR' function, because the fan is controlled digitally. It is supplied with a constant 12v, but a third wire has a PWM signal placed on it by the CPU, and the fan's internal electronics respond to that to control the speed.

Yes, it was the Cray. It had a central octagonal bus backplane as I recall. The met office here in the UK used to use one for weather data number crunching, but I think it has been replaced now.

Arfa

Yes indeed Michael. There are in fact 6 of them. Three on each side of the board ...

Arfa