Best Northbridge and GPU waterblocks?

[DNA]

Originally Posted by Jimbo Mahoney

Thanks, but looking on their website, only the socket 754 is metal. The socket 939 is plastic

I think I'm going to go with a Brass topped RBX with nozzle #5 now, a silverprop fusion HL for the gfx and anything really for the nbridge!

The block, which is all metal, is common to all current socket types. The retention bracket is the only difference. You may have been looking at the HP which does have a plastic top.

Link to XP Sockets 754,939,940:
http://www.alphacool.de/perl/shop.pl...9&art_id=12910

Read the description under MOUNTING. This is the only text in the product description that refers to Sockets 939,940. Pretty poor marketing if I may say so. I wonder how many sales they've lost due to that blunder in the product name which implies Socket 754 only.

[phextwin]

Originally Posted by Butcher

GPUs are nearing 100W heat output at stock speeds, neglecting them as "easy to cool" is not a good idea. In many cases the GPU heat output will exceed that of the CPU.

Somehow i doubt anywhere near 100W. Look at the dinky little heatsink on the x800xt. Even the 6800U/x850xt sink is much smaller than a 560P4 for instance, a well known ~100W cpu.

Perhaps the whole card might draw close to 100W, but i doubt just the GPU specifically.

[Cathar]

Originally Posted by phextwin

Somehow i doubt anywhere near 100W. Look at the dinky little heatsink on the x800xt. Even the 6800U/x850xt sink is much smaller than a 560P4 for instance, a well known ~100W cpu.

Perhaps the whole card might draw close to 100W, but i doubt just the GPU specifically.

These 70W figures that various reviewers are coming up with are measured at the wall, as in card installed vs card not installed.

Mind you, the PC's PSU which powers the card is itself only ~80% efficient, so our 70W at the wall is more like 56W coming out of the PSU into the PC.

Then the video card itself has to step down the voltages, as it's being fed 5v and 12v, and the GPU core and memory don't run at these voltages. Around another 10% gets lost here in the voltage step-down, so our 56W being fed to the card is more like 50W actually consumed by the componentry on the card.

Then we get to the peripheral devices on the card, as in everything that is not the GPU and the memory, and we're let's say they all consume 5W together, leaving 45W of power being drawn by the GPU and memory.

A single piece of high speed DDR will be consuming around 1W of power when under load. There's typically 8 memory chips on a high end card, so that leaves us with around 30-35W of power being drawn by the GPU itself.

Now looking at the heatsinks which get fitted to 80-100W CPU's, and the heatsinks which get fitted to video cards, and given how hot the GPU's actually get under load (my finger is still aching from the last time I touched the back of my new stock air-cooled video card behind the core), we can guess that the 30-35W figure should be fairly close to the mark. I'd be leaning more towards 30W myself.

Now I'm not saying that the above figures are all accurate and occur in reality, but what I'm trying to highlight is that there's a pretty damn big difference between how much power is being drawn from the mains wall socket which is where these reviews are taking their measurements, as opposed to how much power/heat a GPU is actually using/generating.

[BillA]

spot on Cathar
these numbers are puffed up all out of proportion
how can our 80W TEC do so well if the GPU was at 100W ?
no one wants to think about reality

[Butcher]

Interesting article here suggesting a 6800U can draw up to 123.5W (that's measured at the card not the wall).

[Butcher]

Additionally, PCI-e can supply 75W via the slot and yet 6800Gts/6800Us require an additional power connector suggesting card consumption is somewhat in excess of 75W.
I would say that GPUs are definitely consuming the high side of 60W, maybe as high as 80W...

[Tempus]

Originally Posted by Butcher

Interesting article here suggesting a 6800U can draw up to 123.5W (that's measured at the card not the wall).

I'd assume that most of the 6800 cards in with PCI-X use a bridge and are basically the same electronics as the AGP - so drop that 123.5W out. Thats just crap.

Second, you still have voltage stepdown issues.

Third, and always fun, you have no idea what else is going on. To make this slightly more accurate you'd need an additional PSU just running the videocard and then do the measurements.


Fan = 12V @ 0.5Amp or so?

-------

Question: Are the devices on a psu rail wired in parallel or in series?

[freeloadingbum]

X-bit showed the 6800 ultra to use 72.09 watts (under load) as measured at the card (including the agp port, which used minimal power). More interesting was that increasing the gpu from 400 to 440 and the memory from 1100 to 1200 showed a total increase of only 4 watts. This 10% increase sugests that total oc load is then only about 44 watts (gpu and memory combined prior to voltage stepdown)

It also needs to be mentioned that the ultra is 280 sq mm. At 35 watts it has the same watt per sq mm as a amd64 winchester (90 sq mm) does at 11 watts

[Butcher]

Parallel of course. Series wiring would mean everything at a different voltage...

Why would AGP cards using a bridge cause the 123.5W to be rubbish? I can't imagine the bridge uses a vast amount of power (no heatsink). Ok you can knock off a few watts for the bridge, big deal.

Voltage stepdown accounts for 10-20% of power (6800U cards are known to have switching power regulators so they are relatively efficient), not 50+%. Still talking near 100W.

If you check the article you will see he measures the current from the molex connector going into the card. That's not going to be affected by other components, because he's measuring direct input to the card, not total PSU draw.

Unsure on fan power - will check mine later (it's no longer on my card so easy to verify power draw).

[Butcher]

Originally Posted by freeloadingbum

X-bit showed the 6800 ultra to use 72.09 watts (under load) as measured at the card (including the agp port, which used minimal power). More interesting was that increasing the gpu from 400 to 440 and the memory from 1100 to 1200 showed a total increase of only 4 watts. This 10% increase sugests that total oc load is then only about 44 watts (prior to voltage stepdown)

It also needs to be mentioned that the ultra is 280 sq mm. At 35 watts it has the same watt per sq mm as a amd64 winchester (90 sq mm) does at 11 watts

How did they measure current drawn via the slot?

[freeloadingbum]

Originally Posted by Butcher

How did they measure current drawn via the slot?

They taped it off then supplied a direct source

[freeloadingbum]

Here is a link to the original article. http://www.xbitlabs.com/articles/vid...-nv-power.html

unfortunately, I can no longer access any of the important pics or charts so I'll include one I saved earlier.

[Tempus]

Originally Posted by Butcher

Parallel of course. Series wiring would mean everything at a different voltage...

Why would AGP cards using a bridge cause the 123.5W to be rubbish? I can't imagine the bridge uses a vast amount of power (no heatsink). Ok you can knock off a few watts for the bridge, big deal.

Voltage stepdown accounts for 10-20% of power (6800U cards are known to have switching power regulators so they are relatively efficient), not 50+%. Still talking near 100W.

If you check the article you will see he measures the current from the molex connector going into the card. That's not going to be affected by other components, because he's measuring direct input to the card, not total PSU draw.

Unsure on fan power - will check mine later (it's no longer on my card so easy to verify power draw).

Otherway around - they are based on AGP electronics using an AGP -> PCI-X bridge. Ergo, they are wired for agp + aux voltage and then the max rating of the PCI-X bus has no impact.

Second, just because 120ishW is used IF the pci-x bus is full exhusted does not mean that it is (or will be) fully used. Thats the theoretical max load given the measured aux - has nothing to do with reality.
------------------------------------------------------------------
EDIT: Sorry, I'm being dumb. I was confusing this with a different review I just saw covering the power on the PCI-X versions.

The 46W max AGP is still the theoretical max. So I still wouldn't assume its going to hit that. I guess we'd need to hit the bottom of the mobo with the multimeter.

[Tempus]

Originally Posted by Butcher

Parallel of course. Series wiring would mean everything at a different voltage...

Yeah, that pic just looked really odd for a second before I realized how he was testing it. It's been too long since I've taken those electronics classes.

So, please refresh me, because I seem to remember you needed to check against the ground wire -- and he seems to be checking on only the +12V and using the multimeter as the "connection wire" which would seem to imply that its adding the multimeter's own resistance into the circuit and giving a false high reading.

I dunno. I'm sure I'm wrong. Any EE guys out here want to explain it to me?

[Butcher]

That's how ammeters work - you put them in series with the load. A multimeter in current mode typically has a resistance well below 1 ohm.

BTW fan draw - the fan off my XFX 6800GT read 155mA at 12V, so 1.86W.

Also Tempus, please don't use PCI-X for PCI-express; PCI-X is a separate standard and can cause confusion when used interchangably with PCI-express.

[Metzen]

Originally Posted by Butcher

That's how ammeters work - you put them in series with the load. A multimeter in current mode typically has a resistance well below 1 ohm.

BTW fan draw - the fan off my XFX 6800GT read 155mA at 12V, so 1.86W.

Also Tempus, please don't use PCI-X for PCI-express; PCI-X is a separate standard and can cause confusion when used interchangably with PCI-express.

The 6800 Ultra cooling system has a TDP (Thermal Design Power) of 60W. I'm sure they would have given themselves lots of head room...

[Tempus]

Originally Posted by Butcher

Also Tempus, please don't use PCI-X for PCI-express; PCI-X is a separate standard and can cause confusion when used interchangably with PCI-express.

Yeah, sorry about that.

[Jimbo Mahoney]

Originally Posted by DNA

The block, which is all metal, is common to all current socket types. The retention bracket is the only difference. You may have been looking at the HP which does have a plastic top.

Link to XP Sockets 754,939,940:
http://www.alphacool.de/perl/shop.pl...9&art_id=12910

Read the description under MOUNTING. This is the only text in the product description that refers to Sockets 939,940. Pretty poor marketing if I may say so. I wonder how many sales they've lost due to that blunder in the product name which implies Socket 754 only.

Awesome, thanks dude!

OK, now back to NexXxos XP then! LOL

I've e-mailed them suggesting updating the product name on the webbie.

[Butcher]

Originally Posted by Metzen

The 6800 Ultra cooling system has a TDP (Thermal Design Power) of 60W. I'm sure they would have given themselves lots of head room...

TDP is usually the approximate heat dissipation of the device, at least if CPUs are anything to go by (they can actually draw more power than the TDP in some cases). They don't add much on, because lower TDP means the OEMs can design cheaper cooling.

Curious as to where you got that number from?

[Ruiner]

Regarding Cathar's post regarding estimated GPU heat output...
what is the point of watercooling the gpu if you aren't volt-modding?
It's the overvolting that make watercooling so worthwhile on the CPU end, overclocking wise, especially given cooling capacity/dB.
Since even power users rarely voltmod their vid cards, what's the point?

Decent aircooling (i.e. Artic cooling vga silencer) should do the trick, and quietly enough.

Just throwing that out there.
My question assumes you haven't built a remotely placed radbox.

[Butcher]

It's very easy to up the voltage on the latest nvidia cards since it can be done with a simple bios flash.

[freeloadingbum]

Originally Posted by Butcher

It's very easy to up the voltage on the latest nvidia cards since it can be done with a simple bios flash.

How high can you increase it?

[Jimbo Mahoney]

Originally Posted by Ruiner

what is the point of watercooling the gpu if you aren't volt-modding?

Since even power users rarely voltmod their vid cards, what's the point?

Hey dude!

A watercooled GPU can still be overclocked further (than air-cooled) even if it isn't volt modded.

[Ruiner]

Originally Posted by Butcher

It's very easy to up the voltage on the latest nvidia cards since it can be done with a simple bios flash.

No kidding. Excellent [Mr. Burns grin].

I was assuming it still took soldering the pcb.

[Ruiner]

Originally Posted by Jimbo Mahoney

Hey dude!

A watercooled GPU can still be overclocked further (than air-cooled) even if it isn't volt modded.

Yeah maybe, but what, 10-15 MHz over (good) air cooling? Even if we say 20MHz, that's only a 5% bump on a typical 400MHz gpu.
I'm sure you'll notice the extra 1.5 fps.

Now, bump that gpu Vcore up a notch or two and water has its chance to shine. Heat output goes up much faster with increasing voltage than clock speed.