AMD Unveils the Liquid-Cooled, Dual-GPU Radeon R9 295X2 At $1,500

wesbascas (2475022) writes "This morning, AMD unveiled its latest flagship graphics board: the $1,500, liquid-cooled, dual-GPU Radeon R9 295X2. With a pair of Hawaii GPUs that power the company's top-end single-GPU Radeon R9 290X, the new board is sure to make waves at price points that Nvidia currently dominates. In gaming benchmarks, the R9 295X2 performs pretty much in line with a pair of R9 290X cards in CrossFire. However, the R9 295X2 uses specially-binned GPUs which enable the card to run with less power than a duo of the single-GPU cards. Plus, thanks to the closed-loop liquid cooler, the R9 295X doesn't succumb to the nasty throttling issues present on the R9 290X, nor its noisy solution."

Comments

Crypto

[phishen]

And they all sold out instantly and the Litecoin difficulty went up ...

Re:Crypto

[Anonymous Coward]

And then all who bought them realised they should have bought a Viper instead...

Re:Crypto

[fuzzyfuzzyfungus]

And they all sold out instantly and the Litecoin difficulty went up ...

Are there any dual-chip cards that cryptominers actually buy? (honest question, I don't know). Back in my youthful gaming-nut days, dual GPU cards, because of some mixture of worse economies of scale and 'people who absolutely must have the bleeding edge will pay, so why not?', always commanded rather more than twice the price of two equivalent single-GPU cards (and sometimes clocked worse, as well, just to keep the heat down).

Now that everything is PCIe, and the low bandwidth requirements of mining allow you to stuff even x1 slots with GPU cards, I'd have to imagine that your miner would be willing to pay only a very small premium to conserve slots.

Re:Crypto

[drinkypoo]

Now that everything is PCIe, and the low bandwidth requirements of mining allow you to stuff even x1 slots with GPU cards, I'd have to imagine that your miner would be willing to pay only a very small premium to conserve slots.

I'd imagine (I've done as much research as you have, apparently) that it's difficult to come by PCIEx1 cards with meaningful GPUs on them. Only some subset can be hacked from x16 to x1.

Re:Crypto

[ArtForz]

Nowadays? No.
A few years ago? Yes.
Before everyone jumped on the bandwagon, 5970s had similar price/performance to 5850s while being superior in pretty much every other way.
4 cards == 8 GPUs off a $45 board with x1 to x16 riser cables.
Fans that could actually run 24/7 for years without dying.
Lower power usage.

Re:Crypto

[ArcadeMan]

You don't need the memory bandwidth for Scrypt mining, so you can use whatever card you want in a x1 slot, with the help of a riser/cable.

Re:Crypto

[Xmastrspy]

Can someone explain to me why you would use video cards for this type of processing and not get a better CPU? I have read a few articles on this type of processing, but never an explanation why. Is it better suited for this type of processing? Is it just cheaper then CPU?

Re:Crypto

[wagnerrp]

Or a dremel. Just cut off the remaining pins for the extra channels and it works fine. You can even test it out by taping off those pins and mounting it in an x16 slot.

Re:Crypto

[drinkypoo]

You don't need the memory bandwidth for Scrypt mining, so you can use whatever card you want in a x1 slot, with the help of a riser/cable.

Not all cards' BIOS will permit that.

Re:Crypto

[Entropius]

Even a low-ish end GPU is many times faster than the fastest CPU.

CPU's are optimized to make single threads go fast; GPU's are essentially massively parallel processors (hundreds of "cores") optimized to make a collection of threads doing similar things go very, very fast.

The conventional wisdom in my field of computational physics is that one GPU = 30 or 40 CPU cores.

Re:Crypto

[Entropius]

I should qualify: by "low-ish end" I mean "low end compared to what we put in supercomputers". I don't mean Intel integrated graphics here; I mean things like an Nvidia 760 GTX or something -- cards that budget conscious gamers use.

Re:Crypto

[PopeRatzo]

Or a Rage II Pro!

Re:Crypto

[Anonymous Coward]

Or a scrypt ASIC!

Re:Crypto

[Ralph Wiggam]

That can't be a good idea. Litecoin has steadily dropped from $25 to $11 during 2014, after briefly touching $44 late last year. I don't see any reason why the downward trend in bitcoin or litecoin price will reverse.

I can't imagine that a $1500 card plus electricity costs are ever going to pay for themselves.

Re:Crypto

[Xmastrspy]

Wow thanks! I am going to have to do some research into this!

Re:Crypto

[Anonymous Coward]

Graphics processors spend a lot of time applying the same math operations to a large set of numbers, e.g. multiplying a bunch of vectors by a matrix to rotate a model, interpolating a bunch of points in a texture, doing dot products between vectors to do lighting, etc. Instead of doing things in a strict order intended for basic 3d graphics, this over time evolved to allow the operations to be applied in essentially arbitrary programmable order, giving a lot more flexibility for graphics methods and ultimate allowing use on non-graphics data.

The result is a GPU becomes very good at doing a preset list of operations in parallel very fast. If you want to do the same thing to a long list of numbers, the GPU can essentially do hundreds of those numbers at the same time. It is far from a universal replacement of a normal CPU though. There are limits on the program complexity and memory access of the program. If your program has a lot of conditional paths, cross talk between results of other numbers, overlapping memory access, then the operations slow down a lot or might not even be possible on a GPU depending on the situation.

I felt a great disturbance in the Force...

[b0r0din]

as if millions of Litecoins suddenly cried out in terror and were suddenly silenced.

Re:I felt a great disturbance in the Force...

[Anonymous Coward]

Yeah, right... Litecoin and the other scrypt coins are done with custom ASIC's these days. A fraction of the power usage for insane hash rates.

Yes the scrypt ASIC's are real and yes you can get them now. Although personally I'm sticking with my GPU's because they can handle any algorithm changes.

here's how stupid this is

[slashmydots]

Thermal conductivity of water: approx 0.58
Thermal conductivity of copper: approx 401
The only reason to have water cooling in anything is to brag to your friends that you have water cooling. In reality, metal cooling works better.

Re:here's how stupid this is

[Anonymous Coward]

Thermal conductivity of:
Air, athmosphere (gas) 0.024
Air, elevation 10000 m 0.020

HERP DERP

Re:here's how stupid this is

[Dolda2000]

The point with liquid cooling isn't to replace the metal in contact with the chip, you know. It's to replace the air that is normally cooling the metal.

Re:here's how stupid this is

[serviscope_minor]

I really hope this post is a joke.

Not sure: my sarcasm detector is on the fritz today.

Re:here's how stupid this is

[Dancindan84]

It's not so much the thermal conductivity of the GPU->water vs. GPU->copper heatsink that's the direct benefit. It's using the water to carry the heat to a much larger radiator rather than having to have the heatsink directly on the GPU (which greatly limits its size).

Re:here's how stupid this is

[bulled]

Parent is mad because when he tried to dump cold water over his GPU it never ran the same afterwards...

Re:here's how stupid this is

[wisnoskij]

Also, it is LIQUID cooled, not water cooled. I am sure they find better liquids than water, bonus points for them if it does not conduct electricity and therefore will not fry your computer if you spill it all over it, and if they can make it evaporate without leaving a residue.

Re:here's how stupid this is

[troon]

Oh dear.

Water flows, due to being liquid. Copper, on the other hand, is a solid at any temperature you're going to have at home.

You circulate the water between the heat-producing surface and a heat-dissipating radiator.

Re:here's how stupid this is

[Anonymous Coward]

AFAIK, liquid coolers use some kind of alcohol.

Re:here's how stupid this is

[Anonymous Coward]

I detect no sarcasm in the original post, only another s-word.

Re:here's how stupid this is

[Chrisq]

Thermal conductivity of water: approx 0.58 Thermal conductivity of copper: approx 401 The only reason to have water cooling in anything is to brag to your friends that you have water cooling. In reality, metal cooling works better.

Easy solution, Run your CPU at over 1,085 C and use molten copper as a coolant

Re:here's how stupid this is

[Hamsterdan]

No it's not

1- Noise. I'm running a pair of Noctua NF-12 fans at medium speeds on the radiator of my Corsair H70 (the PSU fan makes more noise)
2- Capacity of liquid to move heat *much* faster than air
3- Weight on the motherboard. The little pump is much lighter than a 2 pound heatsink

Before installing that cooler, I couldn't get my E7200 (2.53) beyond 3.2 without insane temps. Now, I'm stable at 3.8 with 75C MAX, 4.1 will need a beefier radiator.

Besides, it's not that more expensive than high-end air

Re:here's how stupid this is

[Tukz]

Most closed loop cooling systems (at least for PC) use deionized water, with added fluorescents for colour in case of transparent tubing.
AFAIK, it's also common in custom builds.

Re:here's how stupid this is

[mujadaddy]

Please share your newsletter link.

Re:here's how stupid this is

[TheDarkMaster]

The point of entry-level sealed-loop liquid cooling is to have the required cooling capacity without having to hold almost 1kg of metal directly above the processor, and in most cases into a dangerous motherboard-bending position.

Re:here's how stupid this is

[Khyber]

Guess where that heat ultimately dissipates any ways?

THE AIR.

It's like you failed thermodynamics in high school.

Re:here's how stupid this is

[Khyber]

Only to still replace it with air cooling further down the line. What a waste.

I design cooling systems for high-heat semiconductors. Guess what? Liquid cooling SUCKS as you're still limited by how fast you can transfer the heat to the air ultimately.

Direct metal cooling, or phase change. Liquid/water cooling sucks.

Re:here's how stupid this is

[Khyber]

>You circulate the water between the heat-producing surface and a heat-dissipating radiator.

Which means you still need air contact and air flow to cool off the radiator.

So you just waste more power on an already inefficient cooling process.

Might as well be a fool and use a Peltier inside the computer at that pint.

Re:here's how stupid this is

[Anonymous Coward]

Liquid cooling sucks. That's why we all have air cooled engines rated at 25hp in our cars nowadays. Oh wait. No we don't.

Re:here's how stupid this is

[Rockoon]

I design cooling systems for high-heat semiconductors.

So we can expect you to be pretty knowledgeable about stuff, not overlooking important details, and understanding the various limitations imposed by each method...

Guess what? Liquid cooling SUCKS as you're still limited by how fast you can transfer the heat to the air ultimately.

...except what we find out is that actually you only have just enough knowledge to look stupid because you overlooked important details such as the obvious limitations imposed by having your radiator mounted directly on the device being cooled.

Re:here's how stupid this is

[Joce640k]

Guess what? Liquid cooling SUCKS as you're still limited by how fast you can transfer the heat to the air ultimately.

...unless your only goal is to overcome the limitations of working in a space the size of a graphics card.

In which case it ROCKS!

Re:here's how stupid this is

[sharknado]

Yes, copper heat sinks get hot fast because they have high thermal conductivity. But once they are hot, they don't cool down very fast. The limiting factor is the copper-to-air interface, not the CPU-to-copper interface. Why do you heat sinks are designed to maximize surface area? Why do you think case fans are so important? Because these improve the heat transfer from the metal heat sink to air. Copper's thermal conductivity obviously isn't the limiting factor. Water cooling is far superior because water has a very high thermal capacity, so it stays cool over a much longer period than a metal heat sink. In addition, the warm water is pumped out and replaced with cool water constantly, so the thermal differential stays high (where in a metal heat sink, once the metal heats up to 70 or 80 degrees, it loses its ability to cool the CPU).

Re:here's how stupid this is

[fuzzyfuzzyfungus]

Thermal conductivity of water: approx 0.58 Thermal conductivity of copper: approx 401 The only reason to have water cooling in anything is to brag to your friends that you have water cooling. In reality, metal cooling works better.

The water (.58) replaces the air (.024) in the cooling setup, not the metal. Watercooling systems tend to have markedly smaller heatsinks, since they just don't need the same surface area(at the low end, just a copper plate with a flow chamber on top of it, some more complex designs use something more or less similar to a small air-cooled heatsink; but sealed for water to flow through.

Now, I think that there have been a few nutty-and-exotic liquid-metal cooling systems; but those are hampered by the fact that they just aren't better-enough than water for the money (the delta T of the CPU's package and the waterblock is still the same), pumping the (substantially denser) liquid metal is more energy intensive, and most candidates are either unpleasant or expensive, or both.

If you want something that won't go all hazmat on you if the system leaks; but won't harden in the cooler parts of the loop, 'Galinstan' is probably the best bet; but you sure don't make things cheap by making them ~ 20% indium.

If you are...aggressively risk tolerant... a nice Cesium/Potassium/Sodium alloy will stay liquid to almost -80 (celsius); but, um, not a good plan, OK. Straight Mercury works fine down to almost -60; but that stuff is dense and not particularly pleasant(plus, it amalgamates with a number of metals quite readily. You did check your waterblock, radiator, pump, and all other contact surfaces for compatibility, right?). If you aren't the kind of coward whose dishonor makes him cry about things like "my cooling system catching fire on exposure to air or water vapor", NaK is a lovely coolant.

Basically, for something that is such a pain in the ass, you'd better be getting results substantially superior to normal air or water cooling, which you'll only get with active heat pumps that can actually pull the CPU below room temperature. At that point, the rather low freezing points of any available metal alloy become an insurmountable problem. Other materials don't have quite the same thermal conductivity; but they'll be happy enough keeping things well below the -100.

Re:here's how stupid this is

[fuzzyfuzzyfungus]

A generous dash of alcohol(generally the denatured stuff, no reason to not drink the other kind) can be useful to keep the lifeforms at bay. Damn microorganisms appear to be capable of living in/on anything they want.

Re:here's how stupid this is

[slashmydots]

You do know that liquid coolers are air cooled, right?

Re:here's how stupid this is

[Anonymous Coward]

Since you've posted twice it is now obvious that you are serious.

And we are shocked.

But not that much.

Re:here's how stupid this is

[Baloroth]

Only to still replace it with air cooling further down the line.

Honest question: how would you build a consumer system that doesn't rely on air cooling eventually? Even if you use phase change, you still need to dump that heat somewhere, so unless you use evaporative cooling or have access to a practically infinite heat sink such as a river or geothermal exchange system (both of which are completely impractical for consumer level equipment), air cooling is literally the only option. Hell, even most (and by most I mean practically all) air conditioning systems use air cooling, ultimately. Probably 99% of all cooling systems everywhere end up using air.

Re:here's how stupid this is

[slashmydots]

Okay, genius, copper cooling = air blowing across metal fins. Water cooling = air blowing across metal fins.

Re:here's how stupid this is

[Anonymous Coward]

Yeah, I was like "WTF?" any mechanical engineer worth his salt appreciates the benefits of being able to relocate the heat exchanger.

Re:here's how stupid this is

[slashmydots]

Except the new card doesn't do that.

Re:here's how stupid this is

[NIK282000]

That's why cars and trucks are all directly air cooled right? Liquid cooling lets you increase the surface area of your radiating element while moving it away from the important parts of your process, without using heat pipes (also liquid cooling) you can't get anywhere near the same surface area as a radiator on a CPU/GPU cooler. The only time you see direct air cooling is to cut costs or save space or weight.

Re:here's how stupid this is

[fuzzyfuzzyfungus]

The main point of enthusiasm is that liquid cooling is easier to do than phase change cooling with hobbyist fabrication (heatpipes do pretty much dominate all but the smallest heatsinks, and have substantially increased the size of air-coolers since the days of solid-copper by making more distant fins actually do something vaguely useful); but building custom heatpipes at home, while doable, isn't wildly pleasant.

Liquid is substantially less elegant; but it's just simple plumbing to set up and it still allows you to use an impractically gigantic radiator, of a size you couldn't bolt onto the CPU.

Now, as for the idiots who introduce the cost and complexity of a liquid loop, all so they can use a radiator no larger than a simple heatpipe-based aircooler... Those guys are just idiots.

Re:here's how stupid this is

[slashmydots]

'Yes, copper heat sinks get hot fast because they have high thermal conductivity. But once they are hot, they don't cool down very fast"
So they violate the laws of physics? Amazing! Thermal conductivity is exactly the same in both directions.

Re:here's how stupid this is

[tompaulco]

unless your only goal is to overcome the limitations of working in a space the size of a graphics card. In which case it ROCKS!

Unless the water cooling system is built on the graphics card itself, in which it case, it sucks.

Re:here's how stupid this is

[slashmydots]

Oh and by the way, thermal capacity doesn't make an ounce of difference. So it reaches max heat in 2 minutes instead of 30 seconds. So what? I'm gaming for 2 hours.

Re:here's how stupid this is

[Dolda2000]

Only to still replace it with air cooling further down the line.

Sure, admittedly so, of course. The point, though, is clearly to be able to use a larger or otherwise better air cooler in the end, which I can certainly see being the point in this case, seeing is how the PCI specification gives too little room for a proper cooler on the card itself, especially if it's going to fit in only two slots.

Re:here's how stupid this is

[Dolda2000]

Yes, of course liquid cooling not magically going to make the heat disappear or otherwise violate the laws of thermodynamics. But it can transfer it to a place where you can have larger metal fins and/or better fans.

Re:here's how stupid this is

[RivenAleem]

Umm why stop there? Why not run it at 2562 C and then you can use both liquid AND gas cooling, as the copper flows, evaporates and condenses.

Re:here's how stupid this is

[Xicor]

you may replace it with air cooling down the line, but that air cooling is much better than the small thing they would have on the gpu. you can get a much larger cooling unit if you dont have to put it directly onto the gpu.

Re:here's how stupid this is

[Anonymous Coward]

Do you own a car? Is it an original beetle? No? Then chances are it's water cooled. Would you like to guess why?

Hint: It isn't because people buy a car due to the cooling system being used. In fact, today, most manufacturers cover the engine bay with plastic covers so ignorant owners can literally not give a shit.

Pro-tip: It might be because the water gets the heat away from the parts that will be damaged really quickly and transfers it to a very large surface where efficient air cooling can happen.

Re:here's how stupid this is

[gnick]

Exactly - Pretty much everything is air-cooled. Even the frion in your fridge. It's just that it's easier to circulate some things than others.

Re:here's how stupid this is

[Khyber]

".except what we find out is that actually you only have just enough knowledge to look stupid because you overlooked important details such as the obvious limitations imposed by having your radiator mounted directly on the device being cooled."

Yet you'll gladly stuff that radiator INSIDE THE COMPUTER CASE and cause premature heat failure of other components inside, like your motherboard.

BRILLIANT!

Re:here's how stupid this is

[K. S. Kyosuke]

That's because you're probably forced to cool high absolute powers. This seems to be more about increasing heat flow through a comparatively small cross-section. Granted, the water-air interface sucks, but I believe the point is to squeeze performance out of a chip that's very high on the diminishing returns curve regarding frequencies and sizes. Also, those high-heat semiconductors don't happen to be low-voltage silicon U/VLSI circuits, do they? I'd imagine power thyristors or rectifiers or something like that.

Re:here's how stupid this is

[wonderingponderer]

I'm replying here because I'm smarter than you are. The question is a matter of surface area. When you combine a small die with limited area to an external HSF the external device will have more surface area. So yes, you still dissipate to air but you can do so to more air via an external device.

Re:here's how stupid this is

[Anonymous Coward]

Unless only the part of the cooling system that gets warmed up is on the graphics card itself, and the part which gets cooled down is (just using TFA as an example) built into a case fan, in which case it ROCKS!

Re:here's how stupid this is

[Khyber]

"Honest question: how would you build a consumer system that doesn't rely on air cooling eventually?"

You don't, pretty much, excepting what you've listed (heat sinking to earth, evaporative cooling.)

Re:here's how stupid this is

[Anonymous Coward]

Water cooling = air blowing across metal fins elsewhere, where it can happen more effectively.

Re:here's how stupid this is

[Khyber]

Plenty of graphite-based heat sink solutions that work better than copper. I've got a graphite-core aluminum fin LGA775 heat sink. Tears up my solid copper Itanium II cooler.

Re:here's how stupid this is

[Anonymous Coward]

All lifeforms, except humans.

Re:here's how stupid this is

[Anonymous Coward]

Only to still replace it with air cooling further down the line. What a waste.

I design cooling systems for high-heat semiconductors. Guess what? Liquid cooling SUCKS as you're still limited by how fast you can transfer the heat to the air ultimately.

Direct metal cooling, or phase change. Liquid/water cooling sucks.

You must clearly be lying about your supposed experience. Almost every high power semiconductor is liquid cooled due to the impossibility of removing the imposed losses from the surface area provided. There's a reason why there are no SCR clamps that are air cooled, or why IGBT chillblocks are employed how they are.

Stupid.

Re:here's how stupid this is

[Anonymous Coward]

Yes, it does. Read the article - the heat is dissipated from a radiator on the exhaust fan attached to the back of the computer's chassis, instead of from a radiator on the card itself.

Re:here's how stupid this is

[Anonymous Coward]

I'm sure they could use you now in Fukushima! Air cooled nuclear reactor would be nice.

Re:here's how stupid this is

[Anonymous Coward]

Are you really that dense to keep posting without thought? Yes thermal conductivity is the same in both directions, but the 100+ W generated on the die heats up the heatsink because the conductivity between the chip and heat sink is high, but you can't absorb 100 W of heat with the chip. The source of heat is thermally well connected to the heatsink, but the ultimate destination of the heat (the air) is not.

Re:here's how stupid this is

[DocSavage64109]

Radiators are usually mounted on a case panel with fans blowing outward. How exactly does that heat up the other components?

Re:here's how stupid this is

[K. S. Kyosuke]

I wonder it it weren't simpler to use some gas to cool it. I was thinking of using small amounts of hydrogen at comparatively low pressure. You can hardly beat the thermal capacity. I'm just not sure what this would do to the metal structure.

Re:here's how stupid this is

[Anonymous Coward]

Of course we're not shocked, Khyber didn't design our electrical distribution subsystem.

Re:here's how stupid this is

[K. S. Kyosuke]

Well, in Nordic countries, I'd opt for liquid cooling and simply put the other end into the permafrost. ;-)

Re:here's how stupid this is

[Xicor]

that may be true, but im sure that they could make graphite based water cooling radiators as well.

Re:here's how stupid this is

[K. S. Kyosuke]

Well, it's a completely different issue (I believe the problem with ICs is in the small temperature differential), but there you are. ;-)

Re:here's how stupid this is

[Anonymous Coward]

I design cooling systems for high-heat semiconductors. Guess what? Liquid cooling SUCKS as you're still limited by how fast you can transfer the heat to the air ultimately.

I also design cooling systems for high power semiconductor circuits. You could say liquid cooling sucks, not because of its cooling ability, but because it adds complexity, expense and can require more maintenance depending on its setup. We try to air cool whatever possible, but once the power density gets too high or the temperature requirements too narrow, we switch to water cooling or heat pipes. Even if you are just moving the heat a couple inches, in many setups that gives you loads more room to build a radiator that is more efficient than any heatsink you could have crammed in. Whatever fancy tech you can throw at a heat sink you can throw at the radiator, except now you can optimize the layout of the circuit better, and optimize the airflow through the radiator better, with more surface area.

Re:here's how stupid this is

[drinkypoo]

Now, as for the idiots who introduce the cost and complexity of a liquid loop, all so they can use a radiator no larger than a simple heatpipe-based aircooler... Those guys are just idiots.

If they're cooling the entire system, they're also cutting their fan count down to 1. It's expensive, but it cuts a lot of the noise out of the system. Also, you can get pretty big radiators, for example a motorcycle oil cooler has considerably more area than you can reasonably get onto your processor.

I've never bothered, but I'm starting to think about it again because my computer is noisy and I have a motorcycle oil cooler lying around. Maybe I'll fill the system with some of this magical Ravenol crap they're using in fancy VWs now, since I have it and it looks neato. I milled a deep-passage water block a long while back and never have got around to using it.

Re:here's how stupid this is

[bulled]

If you are dumping air that runs over your radiator back into the case, you are doing it wrong. BRILLIANT!

Re:here's how stupid this is

[Anonymous Coward]

Just how stupid are you? Have you ever seen a case with liquid cooling? Just shut up and stop embarrasing yourself.

Re:here's how stupid this is

[drinkypoo]

Mostly they use distilled (or if you are feeling fancy, deionized) water with some additive. You can add alcohol, but I would prefer an automotive coolant additive. You could use redline water wetter, for example. I'd just go ahead and install a normal automotive coolant, which comes in a variety of colors. I'd want a low-silicate coolant without any special additives. The old-fashioned green stuff (ethylene glycol) is my favorite. Use about 25 percent to retard corrosion. If you want to get really nerdy, mount a voltmeter in the system. Ground to the metal of the water block or the radiator and suspend the positive electrode in the coolant solution. If you're making more than about 100mV then change the coolant. The meter is under ten bucks, digital or analog. Use a stainless or brass electrode to reduce corrosion of the electrode itself. A screw of appropriate length pushed through the top of the reservoir and sealed with epoxy or goop (tm) will serve.

Re:here's how stupid this is

[drinkypoo]

Thermal conductivity is exactly the same in both directions.

Well no, no it isn't. Because the thermal conductivity on the inward direction is based on the interface between two metals, and the outward direction is based on the interface between metal and air. See, one of the laws of thermodynamics is that heat energy always moves from hot to cold. So in fact, we do have an actual directionality to the heat flow; into the heat sink from the hot processor, and into the air from the heat sink. You fail both physics and English, and I suck at physics.

The rate of thermal transfer is based on both conductivity and delta T. That's why you put your cream in your coffee right away if you want it to stay hot longer.

Re:here's how stupid this is

[drinkypoo]

If you had a filter (say, RO) running all the time you could feed it through your water block :)

Re:here's how stupid this is

[L4t3r4lu5]

The radiator for the liquid loop is not on the graphics card, only the waterblock and pump are. The radiator is separate and designed to be mounted on the 120mm case-fan mounting at the rear of your case (or wherever else it will fit) and exhaust out of the case. You lose can only gain in cooling efficiency as you are increasing the air-cooled surface area (120mm rad compared to standard GPU heatsink), you're exhausting directly out of the case thereby reducing ambient temperature compared to an enclosed air-cooling system, and you're getting a lower RPM, therefore lower noise, fan to boot.

Re:here's how stupid this is

[serviscope_minor]

The main point of enthusiasm is that liquid cooling is easier to do than phase change cooling with hobbyist

There are several advantages. One key one is that coolant pipes are longer and much, much, much more flexible than heatpipe pipes. This means you can easily place the large radiator on the edge of the case, with the fan blowing outwards. This has substantial advantages than the heatpipe coolers which recircuate air withing the case and rely on other fans to cycle fresh air.

The cost and complexity is low, these days now that you can buy sealed, closed loop liquid coolers.

Even so, it's not just a question of radiator size, there's the question of placement too. I've used one big non-closed loop liquid cooled machine which had a top end CPU and a bunch of high end graphics cards (2 or 3?). The total heatsink area was probably less than having decent heatpipe coolers on all the components. It is however much easier to build. The heat pipes must be close to the thing being cooled, and the thing being cooled gets in the way of the air flow.

With the water cooler, you have a nice, simple mesh of fine fins with straight linear airflow all the way through it.

The machine dissipated nearly a kilowatt at full load and wasn't noisy.

From my point of view the complexity was low since the machine was bought like that from a specialist vendor. The cost was a little higher, but that was better than having an aggressively noisy monster with overheating trouble.

Re:here's how stupid this is

[Anonymous Coward]

Depends on what type of airflow situation you have inside your case. Theres no reason to blow 45C air over your radiator, when you can draw 20-25C air from the outside in, and let your exhaust fans do their job.

Re:here's how stupid this is

[Anonymous Coward]

The actual cooling capacity of the system depends primarily on the radiator (exposed surface aria and airflow over the surface area, as well as temperature difference between the radiator and ambient).

The advantage of water cooling is that you can attach your heat generating parts to an arbitrarily large radiator (or array of radiators) which can be placed externally for betetr airflow if your case doesn't allow a good mounting (While technicly you could also do this with metal conductors, it would be more expensive, and harder to fit pieces). You can also use the same radiator to cool multiple components. You could even do something exotic like submerge the radiator in an ice-bath, place it in front of your air conditioner, etc.

It's pretty pointless on a closed-loop single component setup like this (copper pipes filled with state change material or "heat pipes" would be better here), but if you're looking for a way to cool an entire system than there are a lot of advantages to a well thought out water cooling system.

Re:here's how stupid this is

[DarthVain]

Its size isn't so much the matter (though it helps) it is about moving the heat to be radiated someplace else, preferably no where near the thing you want to actually cool. This is where water or liquid cooling come into play, as a heat transportation medium.

Heatpipes is another halfassed way to do it, though not as efficient and limited in range.

The only issue with liquid cooling is a leak obviously, and if the pump ever fails, your small copper and water heatsink isn't going to perform very well.

Re:here's how stupid this is

[Khyber]

Not really, due to how carbon fibers tend to conduct and radiate heat - only at their ends.

Re:here's how stupid this is

[Khyber]

I have one case with liquid cooling, and old Smilodon.

I get better cooling with my graphite-core aluminum fin spreader, outperforming a solid copper Itanium II heat sink.

Come back when you have to deal with 1,000w in a 50mmx50mm package and need to keep junction temps BELOW 50C.

Re:here's how stupid this is

[Anonymous Coward]

You might want to consider the possibility that you don't know what you're talking about instead of just being an arrogant asshole right off the bat.

Re:here's how stupid this is

[wagnerrp]

In this case, they're sacrificing the blower fan, rear exhaust, and all that surface area directly on the card, for a single 120mmx20mm radiator that gets mounted to the back of your case. Using water cooling to feed a giant external heat exchanger is beneficial compared to traditional air exchangers directly in the case. "closed loop" water cooling systems are nothing more than the Monster cables of the heat sink market.

Re:here's how stupid this is

[wagnerrp]

A standard GPU heatsink is really god damned big. It's about an inch thick, and extends the entire length of the card, with a big centrifugal blower fan at one end, and exhaust vents out the back of the case on the other. The 120mmx20mm radiator they replaced it with is actually smaller.

Re:here's how stupid this is

[wagnerrp]

The main point of enthusiasm is that liquid cooling is easier to do than phase change cooling with hobbyist

There are several advantages. One key one is that coolant pipes are longer and much, much, much more flexible than heatpipe pipes. This means you can easily place the large radiator on the edge of the case, with the fan blowing outwards. This has substantial advantages than the heatpipe coolers which recircuate air withing the case and rely on other fans to cycle fresh air.

That really sounds like a flaw in your case than the radiators contained within it. Back when we had blow-down heatsinks instead of tower ones, cases would come with side vents and shrouds to provide cool air directly to the fans. If you open up a server case, chances are it will have a shroud that covers both CPUs and all your memory, ducting air directly from several of the mid-plane fans. You can get cases that move the power supply to the bottom, so the CPU is directly in the top rear corner of the case, next to exhaust fans. You can get cases that rotate the board sideways so that everything vents upwards, taking advantage of convection to prevent recirculation outside the case.

Re:here's how stupid this is

[wagnerrp]

Except in this case, it's carrying the heat to a single 120mm exhaust port, instead of the entire surface area of that huge card it would have otherwise normally had.

Re:here's how stupid this is

[fuzzyfuzzyfungus]

It doesn't have the passive convenience of heatpipes, so you don't see it in computers (much, there is or was some outfit selling them at one point); but systems that use phase-change refrigerants can have hoses of basically any length your tube-fu lets you get away with. Extremely common in buildings with AC and external heat exchangers and the like. Heatpipes are great because they get you better results than solid copper, with less copper than solid copper; but are only the most basic application of phase change. Once you add pumps and compressors back into the system, though, the laptop is so difficult to get screwed shut again...

Re:here's how stupid this is

[wagnerrp]

In fact, heatpipes being a form of phase change cooling, are much more efficient than pumped water at short and medium ranges. The only issue with heatpipes is that the average hobbyiest can't make them themselves, and they require static routing.

There's no problem with convecting your heat near the thing you actually want to cool, so long as you can get enough airflow there.

Re:here's how stupid this is

[wagnerrp]

Except we don't use copper any longer, we use ammonia and a passive refrigeration cycle.

Re:here's how stupid this is

[wagnerrp]

Water cooling is far superior because water has a very high thermal capacity, so it stays cool over a much longer period than a metal heat sink.

Are you suggesting that heatsinks work by absorbing and storing the waste heat generated by the CPU?

Re:here's how stupid this is

[xfade551]

The other thing that is being missed in all this discussion about liquid vs. air cooling is the fact that the resistivity of the metal and silicon in the electronics rises as the temperature of the material rises. Remember this equation: P=i^2*R? Well that R, resistance, is also a function of Temperature. If you can transfer a higher amount of heat away from the processors, etc. with liquid cooling, the temperature inside the case will stay lower and less heat will be generated in the first place. Therefore less heat gets dumped into the air in your mom's basement (so maybe you don't want that if you live in Minnesota in the winter).

Re:here's how stupid this is

[serviscope_minor]

That really sounds like a flaw in your case than the radiators contained within it.

That's how most non-server cases are made. It's much easier to move a radiator to the edge than to faff with air channels and stuff.

Re:here's how stupid this is

[wonderingponderer]

They can't fit a giant heatsink on the card because there are two dies in close proximity. What they can do is fit a small copper block and run fluid through it since fluid can carry more heat than air.

Honestly, while I somewhat agree with what you're saying it really doesn't apply here. There is simply not enough room for two giant HSFs on the card.

Re:here's how stupid this is

[Khashishi]

Could you use gallium instead? Granted, gallium is a lot more expensive than water, but if you are the type to spend $1500 on video cards, you clearly have money to burn.

Re:here's how stupid this is

[Anonymous Coward]

Hahaha what a dumbass

Re:here's how stupid this is

[Anonymous Coward]

As an auto mechanic I really hate it when owners literally give a shit. It makes my job very messy.

Re:here's how stupid this is

[Anonymous Coward]

If it can attach to a cpu die or heatspreader plate, then it can attach to a water block and be a radiator. Except now you can use a bigger one, or two if you need to, or have more choice in where to direct the airflow and heat.

Re:here's how stupid this is

[wagnerrp]

The entire side of the card is one giant heatsink, shared between the two dies. The heatsink is separate from the dies by a heatspreader, basically a giant flat heatpipe, to ensure roughly equal cooling between the two dies.

Re:here's how stupid this is

[ByteSlicer]

He was told a Steam box is great for playing games...

Re:here's how stupid this is

[gregmac]

My last CPU upgrade in my desktop came with a fan that about doubled the noise from my system, and more when the CPU got cranked up. At suggestion of a friend, I installed a closed-loop water cooler on it (which cost me ~$70 IIRC). Night and day.

It actually runs cooler all the time (I was monitoring the temp before/after because I was curious) and the whole computer now makes less noise than before. When the furnace comes on, the sound of the air coming from the vent is actually louder, to put it in perspective.

Also, my computer has no 'window' in the case, extra lights, or any of that 'enthusiast' crap that you'd probably associate with liquid cooling and people that brag about it. In fact, aside from recommending it to another friend once, this is the most I've ever said/typed on the subject.

TL;DR: liquid cooling makes your computer nearly silent, is very effective at cooling, and you don't have to be a fanatic to use it.

Re:here's how stupid this is

[bulled]

Or you can isolate the rest of the case from the airflove over your radiator.

Re:here's how stupid this is

[bughunter]

Well if you're gonna go there, then ultimately, pretty much everything is radiatively coupled to interstellar space.

Re:here's how stupid this is

[DarthVain]

While generally true that really depends on how hot things are going to get relative to the thing you wish to cool. Too hot is too hot. Also though a bit off topic is that the hotter it gets the more and more airflow that is required. At a certain point not only are mechanical fans loud, but so is the moving air, not to mention the literal vacuum cleaner/leaf blower you just created.

I this case (sorry no pun intended) making a 1500$ ridiculous video card that takes your two biggest hottest chips that use the most energy (tho this version apparently tweaked to take less somehow), puts them on the same board, would require something a bit more than your standard heat pipe. Hence why it is probably liquid cooled in the first place.

I recently experimented with one of the newer AIO liquid cooled solutions (H90 I believe it was). Not sure how much better it would be from a high end conventional heat sink heat pipe fan combo, however it does have some advantages. One of which that isn't always apparent is displaced space. Putting a huge 1kg chunk of copper, pipes, and 140mm fan takes of a lot of real estate literally in the middle of all the business. At least with the liquid cooling solution you can relocate that someplace else, though you still have to route the tubes. In an ITX case for example it can be a big deal, particularly if you actually want to access anything on the MB without removing your cooling solution every time.

Re:here's how stupid this is

[Anonymous Coward]

No, it's not.
The actual fin area even on a 6990 is rather small, even compared to the "small" radiator they're using. Measure it yourself.
You're also ignoring the issue of actually getting airflow through a long dense fin array, those didn't have 12V 2.4A centrifugals for fun.

Re:here's how stupid this is

[Anonymous Coward]

Might as well be a fool and use a Peltier inside the computer at that pint.

I'll drink to that!

Re:here's how stupid this is

[slashmydots]

Here's my magic plan to solve that, considering I've built 300 computers. Look up the noise specs prior to buying an air-cooled copper solution. Or just get a properly sized Silenx fan and put it on the heatsink in place of whatever atrocity of a fan came with it.

Re:here's how stupid this is

[Khyber]

You'd thaw the entire Nordic region with like two of these and an nVidia 9800.

Re:here's how stupid this is

[L4t3r4lu5]

That is only one consideration, and a questionable conclusion at best. There is also airflow; Very good through the 25mm thick radiator an unrestricted 120mm^2 case fan mounting, poor through 100mm of densly packed fins and the height of a PCI slot and out through the single slot exhaust. If you're running an aftermarket cooler (Windforce 3, as I have) then that's different, but that exhausts into the case, increasing ambient temperature. There's also acoustic preferences; The stock AMD coolers *howl* under load, as the tornado-style fans are small and have to spin pretty damn fast to get the air pressure required to push air through those small fin gaps at a reasonable pressure. There's no such issue with the water loop; 120mm fan can be silent and still easily cool a GPU without the into-case exhaust of the aftermarket coolers. My CPU is cooled by a Corsair Hydro closed loop system; Not ideal when coupled with the Windforce, but it gets the job done.

If I was in the market for a $1000+ graphics card and didn't already have my own water system, I'd snap up one of these AMD cards in a second. It's almost guaranteed to be next to silent as long as the tornado fan can keep the RAM and MOSFETs cool at a reasonable RPM.

Conductivity is close to irrelevant.

[Anonymous Coward]

Water is just used as a transport medium to move heat from one place (the GPU) to another (the radiator). While metal works decently it's, well, solid. Water on the other hand can and does move inside pipes and it's easier to distribute heat energy over bigger areas.

Water does beat pure-air when the absolute energy quantites get big enough.

Just imagine

[Dishwasha]

A Raspberry-Pi Beowulf cluster of those R9 295X2's.

this is the future

[Anonymous Coward]

less pwer using and same graphics ...ontario canada - 42% hike in energy costs....

thats 12 million people of the nation that has oil

Re:this is the future

[mellon]

What the hell does this have to do with high performance 3D graphics cards? And which would you rather eat: oil, or food?

Re:this is the future

[Anonymous Coward]

It's obviously a huge conspiracy by big oil, make sure America doesn't drill, use up all the middle east reserves now, sell America's reserves back to them at hugely inflated costs later down the line.

18 pages, really??

[rodrigoandrade]

Does anyone actually reads those ridiculousy long tech reviews? Or just skip to the verdict/conclusion page?

Re:18 pages, really??

[Amtrak]

To be fair most of the review is graphs and to answer your question sometimes. I'm just hoping some 3rd party vendor will release this card with standard attachments for hose barbs so I can just drop this card into my already watercooled computer. I don't want another fan/pump/rad assembly when I can just add this card into my current loop.

Re:18 pages, really??

[TubeSteak]

These look like standard hose barbs to me:
http://media.bestofmicro.com/4/H/430433/original/radeon-r9-295x2-blown-up.png
http://media.bestofmicro.com/4/J/430435/original/radeon-r9-295x2-pieces.jpg

You can also see that the GPUs are cooled serially and not in parallel.
Tomshardware doesn't break out the GPU temps individually,
so we don't know if the second GPU is running hotter.

Re:18 pages, really??

[Anonymous Coward]

If we knew the flow rate, it'd be easy to say.
Let's say we have 5l/min.
Water has a specific thermal capacity of about 4.18 J/(g*K).
5l/min are about 83.33 ml/s, and for our purposes one ml of water is close enough to 1 gram.
So to heat those 83.33g/s of water up by 1 K, we need... 4.18*83.33 J/s. Well, J/s are W... about 348W.

So if this loop is anywhere close to that flow rate, chip to chip variation and slight variations in mounting pressure and TIM will completely swamp that temp delta.

Comment removed

[account_deleted]

Comment removed based on user account deletion

Do not want

[Anonymous Coward]

Didn't find any need for the $1000 Titan card, doubtful I will find a need for a $1500 flavor either.

Patience works well. Wait a year or two and you can pick up this awesome horsepower at a fraction of the price. Pick up any games that require this much horsepower at the same time and you're golden. It's similar to how I buy games today. I'll be damned if I'm paying full price for what is effectively Beta III. I'll let them sit for a while, let the world test it and complain, watch all the patches get applied and ultimately pick it up when it goes on sale for $20 or so.

I learned long ago to quit buying bleeding edge gear.

Re:Do not want

[slashmydots]

LITECOIN MINING! lol. Make dem dolla dolla bills y'all.

Re:Do not want

[Entropius]

The Titan was a clever thing from Nvidia: a product marketed to gamers that the budget supercomputing crowd is buying. This means that they don't have to provide professional-level support for the things, but can sell them at semi-professional-level prices. They give top-end performance and are (as far as we can tell -- we have a few dozen of the things) as stable as anything else.

An Nvidia K20X costs many thousands of dollars and is actually slower than a $1K Titan (by about 10%, according to lattice QCD benchmarks). My research group recently bought 32 Titans (for $32K) and used them to upgrade an older cluster; we get better performance per GPU than the top-end supercomputer across campus that uses K20's. Both computers have two GPU's per node, but the top-end supercomputer cost somewhere around $20K/node.

13 watts at idle is better than I expected

[Dr. Spork]

So finally, AMD came out with a power algorithm that reduces power consumption when the resources of the GPU aren't needed. I'm not sure where in their product line they introduced it, but it's about damn time. There are all kinds of good reasons to leave our computers on all the time, but I haven't been doing it because the idle power consumption has been needlessly high - in my case, over 50 watts. This adds up over time. Just how much power does a high-end computer need to idle, serve files, run non-demanding background processes, etc.? Millions of computers do just that, for many hours every day. A focus on reducing the power draw of these basically idling computers could make a huge difference to the world.

Re:13 watts at idle is better than I expected

[drinkypoo]

Just how much power does a high-end computer need to idle, serve files, run non-demanding background processes, etc.? Millions of computers do just that, for many hours every day.

Serving files, best done by something dedicated. In my case, I am using an ultra-low-power solution, a Pogoplug connected to the disk via USB3. Performance will not exactly set the world alight; via my crappy dlink GigE switch I'm getting about 10-15MB/sec real-world as reported by various file managers. Performance is better via samba than nfs, but I haven't looked into why at all. The disk spins down when not in use and the pogoplug uses jack diddly for power.

Most of the rest of the idle consumption could be eliminated by simply letting the machine sleep... If it's not serving as a filer, it doesn't need to be on. Especially if your filer is capable of making downloads.

Re:13 watts at idle is better than I expected

[Anonymous Coward]

Great news Dr. Spork!

Reducing idle power consumption has been a major focus of Intel, AMD, and Nvidia.

Not just in the CPUs, but in other components like PSUs and even fan controls.

See the idle modes for Haswell, for example.

Re:13 watts at idle is better than I expected

[AmiMoJo]

It's been there since the 5000 series. I had one of the high idle power 4000s, but they fixed it with the 5000s. Shame the promised driver update to reduce power on the 4000s never materialized.

Re:13 watts at idle is better than I expected

[Anonymous Coward]

Well, kinda, but not really.

4xxx had neither clock nor power gating for the shaders, they did have clock/power gating for the UVD block and display encoders/drivers. So beyond those all they could do is reduce core voltage/clock at idle. Technically they did have hardware for shader clock gating, but it's so buggy it simply can't be used.

5xxx introduced working shader clock gating, so now idle shader blocks got their clocks turned off.

7xxx introduced core power gating, so when *no* shaders are being used the whole core powers down. Issue here is... the only time the whole core is idle when there's no outputs active. Which means secondary card in CF when nothing using CF is running, or a primary/single card when displays are off/DPMS standby.

Hawaii introduced more granular shader power gating, so it can power off idle shaders in banks of about 128. As you can imagine, you don't need very many active shader blocks just to display a composited static desktop or some video.

Punny, that

[Anonymous Coward]

Shure to make waves, that liquid cooled GPU

Yep

[goldcd]

I was running a pair of 6990s (previous gen dual-GPU AMD cards).
Bought the first one when I realized that Bitcoin app I'd accidentally installed a few months earlier on my server had produced something I could flog for 10 bucks and I'd earnt myself a free card. Then used that one to cover the cost of the second one.
Quit mining when difficulty meant I was pulling in less than a BTC a day.. Looking back..

By that logic...

[Junta]

Might as well remove the radiator from your car, after all, it only gets cooled by the air, so you might as well just let air flow over the engine and it will be just as good.

Here, it looks like they are looking for additional heatsink and exhaust volume than they can fit in a dual-high form factor, meaning liquid transfer to the additional exhaust sink/fan. I personally think it a bit much in terms of GPU capabilities, but it doesn't mean it's totally silly.

Easy solution!

[Anonymous Coward]

> [...] serve files [...]

simply don't use your GPU to serve files.

(SCNR)

Drivers

[Anonymous Coward]

Great now if I could just get Linux Drivers that don't hose my system.

Re:Drivers

[amirishere]

Second that, my pc hangs once every two times that I boot into linux because of a race condition in the drivers. I don't have this problem when using the open source drivers, but then gnome shell has problems.

Also their OpenCL implementation has problems which prevents Blenders Cycles engine from being usable through their GPU.

I'd rather they spend money on their drivers than on some Frankenstein GPU manifestation.

Timing is everything

[Anonymous Coward]

This news is carefully timed to proceed AMD's Q1 2014 earnings report due out next week. Nuff said.

It's for gamers

[Powercntrl]

I really hope for AMD's sake that gamers are looking to snatch these things up, because GPU-based cryptocurrency mining is looking pretty abysmal at the moment. Assuming this card can do about 2MH/s, that's 0.0108 of a Bitcoin each day (by CleverMining's pump-and-dump pool). Of course, you aren't mining Bitcoins, you're mining altcoins that are being dumped onto exchanges where fools buy them with Bitcoins, in the (most likely false) hope that one day they'll be worth more than they paid. As the supply of fools dries up, profitability goes down.

Altcoins only have what little value they currently carry because not everyone who mines them, immediately sells. High end Scrypt ASICs will concentrate the distribution of coins to people who are only interested in cashing out. If a larger proportion of coins are dumped onto exchanges rather than held, well... Supply and demand - you do the math.