A lot of it has to do with the layout of the card, you'll notice the VRMs on the GTX280 are very sparsely spaced, which makes the cards absolutely enormous, but allows for much better cooling of the VRMs as a large amount of surface area can be used:

The HD5850 VRMs are linearly organised, so the card is more compact [and arguably, compatible] but it means more difficulty with custom cooling.
Thermalright have 'solved' this problem by using larger VRM sinks that overspill:

Thanks for clarifications.

If I am not mistaken, another thing that may have to see is that Radeons HD have a Digital PWM while GeForces GTX have an analogic PWM.

Anyway, have HD 6870 VRM the same fury than HD 5870/50 VRM has?

Sparkle announces their Calibre X580 graphics card, using this Arctic cooling solution:
http://www.techpowerup.com/134204/Spark ... -Card.html

Thanks for a nice review!

Anybody know why its not compatible with the radeon 5850?

edit: I did some research and it seems it is compatible with the ATI Radeon HD 5850, but only for single DVI port layout version.
So I guess the top DVI connection gets in the way somehow. I wonder if you can remove the metal covers over the DVI port and get around this problem?

I've been considering the Accelero Xtreme for my Powercolor 6950 and so was reading the review and this thread again and a couple of things came to mind.

Firstly, I don't understand why the VRM temp wasn't measured with the stock cooler, as that's a function of the board's controller and has nothing to do with whatever cooler is fitted.

As for supporting the end of the card to prevent it sagging, perhaps they could have provided some legs that would rest on the bottom of the case to do that, perhaps adjustable in length to fit different cases.

Another problem with a long and deep cooler like this is it will block some of my SATA ports. The Powercolor already blocks two and with the Accelero it will block another 1-2, so I might have to buy a PCI SATA card. If you've got a motherboard with right-angled SATA ports this won't be a problem, but mine doesn't (I'm sure there were good reasons why I bought this motherboard and had no idea at the time that I'd be needing such a big GPU cooler).

My Powercolor 6950 isn't exactly reference though as you can see from this picture, so I'll have to check the Xtreme will even fit it first. The VRM is at the opposite end of the card to reference and has a one piece heatsink, which hopefully I won't have to replace. Unfortunately, the CHiL controller doesn't seem to support VRM Temperature monitoring (or at least none of the current apps, such as GPU-Z, can read it), so I've got no way of telling if it's adequately cooled!

I think my ideal setup would be a motherboard with a suitable PCI-E slot near the bottom of the motherboard (but still with enough room for a 3-4 slot cooler) and the SATA ports higher than that slot, along with 2-3 PCI slots in case I need them. Then I'd want a case with fans or passive intake holes in the bottom so that the GPU cooler could get cool air directly from outside the case. Having the graphics card nearer the bottom of the case would also mean I'd only need fairly short legs at the end to support the cooler.

I seem to be able to buy the Accelero Xtreme 5870 about £5 cheaper than the Accelero Xtreme Plus, even though they're the same except that the former supposedly comes with the VR001 heatsink set, which is all I need for my 6950. Even if it turns out not to have the right sinks and I have to buy the VR001 set, it will still work out cheaper as I'd have to buy that if I got the Plus anyway.

I have a 5850 with an Xtreme Plus, which has a very similar layout, see (http://www.overclockersclub.com/reviews/sapphire_hd5830_hd5850/images/28.htm). I had to change the VRM heatsinks, too big to fit under the Xtreme Plus. You will have to change it (easy) or cut it.


For a 5850, with no overvolting, using these heatsinks (http://www.dealextreme.com/p/pure-copper-pc-memory-heat-spreader-8-pieces-24500) with Accelero Xtreme Plus fan at 30% in Furmark is more than enough (I keep checking VRM temperature via monitoring device). You might be able to make a good guess about your card comparing 6950 vs 5850 power consumption.


I have two 5850 with Accelero Xtreme plus in crossfire: both bend a bit, but you won't need a "leg".



I'm not completely sure whether the Accelero Xtreme wil fit the graphics card, the rectangular box enclosing the DVI connectors is a big too big: in fact, that's the only difference between the Accelero Xtreme 5870 and the "normal" one, a little indentation to make sure the heatsink does not collide with the DVI connectors box. Check twice the Accelero Xtreme 5870 too, though I think it will fit perfectly.

You mean the original one is too tall? It doesn't look that tall but obviously you know. I might have needed to replace it anyway to cope with overclocking the card.



I know it can hit about 250W once unlocked and overclocked to 6970 speeds with powertune at +20%. Apparently a 6970 runs about 187W Peak, 287W Maximum but I don't know if that has a beefier VRM heatsink or just cooler running VRMs than the 6950. A stock 6950 is about 128W Peak, 182W Maximum and a 5850 108W Peak, 150W Maximum.



Well sure, you don't have to use anything to support the end of the card, but I doubt it's good for it to be bent like that and I don't really want to risk damaging a £200 card. I'll probably just ziptie the end of the heatsink to something to hold it up though, as making legs is probably beyond me.



Thanks for pointing that out, I would never have noticed. It seems rather bizarre that the Accelero Xtreme Plus, which is marketed as compatible with the 6950/6970 won't fit (don't all this range of cards have the DVI connectors box?), whilst the cooler marketed as only compatible with the 5870, will fit the 6950!

Assuming the 20W idle rating is accurate, the maximum loading I can create on my HD6970s suggests their TDP maxes out in the real world at 238-242W. Using stuff like furmark, OCCT or the bitcoin miner the greatest difference I can create between idle and load power is 500-510W AC, between two cards. Applying an estimated 87% efficiency, that gives a total of 20+0.87(250)=237.5W or so.

I don't know for sure because the photo doesn't have the adequate perspective, but it is only logical that this heatsink will have the same height, as there is space enough. The one I had was 2 or 2,5 cm tall at the very minimum. I doubt you'll find a better one that fits the reduced space there is available, as VRMs are sandwiched by capacitors.


If you are going to overvolt in order to overclock, then consumption will grow real fast and the VRM will have to handle that. I want to insist, because power consumption does increase non linearly with voltage increase. As an example: by moving from 1.088 to 1.193 V one of my cards incurred in a 30% (!) increase in power consumption: that was furmark stable, though.

With respect to the VRM quality, this depends a lot on the graphics card you get for non reference designs. Of course, the 6970 is a top dog carrying a considerably higher price: it is very logical to think they will provide better components.

6970 consumption figures and using a worse VRM cooler than the one that comes with the card makes me feel a bit uncomfortable about VRM temperatures, though it will very likely work if you use good VRM heatsinks (beware of small ones). Can't provide guarantees, though.



These guys have sold many cards, and I have not heard of a single one failing becasue of this: I think this is the kind of problem that would have caused lots of noise and fury on the Internet if it ever happened. But zipties will not do any harm if you can attach them



This kind of thing always happens unless you get cards following the reference design, that is what aftermarket coolers are designed for. They just can't accomodate all custom designs, you have to check your card when going the aftermarket cooling route.

Regards

I'm not planning to overvolt really. I seem to be stable at 900Mhz Core, 1350Mhz Memory (using MSI Afterburner) at stock 1.1v, which I'm happy with and hasn't seemed to have made much, if any, difference to power consumption.

I'm having problems with PowerTune though, in that increasing it (I've tried +5, +10, +15, +20%) doesn't make any difference to the power consumption under load at the above overclock, so something doesn't seem right.



Is GPU-Z useful for determining power consumption? Idling it shows 4A @ 0.9v and under load testing at 900/1350Mhz it shows 75A @ 1.094v VDDC, which works out to about 82W, which is clearly not the full picture. My power meter shows the idle AC draw around 100-115W, increasing to around 260W at load, so around 150W AC increase.



I'm still confused though, as it looks like even the reference cards have the same double PCI bracket and DVI connector box, so surely the Accelero Xtreme Plus wouldn't fit on them either?

Thanks for that. This article puts the stock 6870 in Furmark at 228W, a +20% PT'd 6870 at 277W (looking at the other article again, the 287W Maximum given for the 6870 was obtained at +20%) and a +20% PT'd 6950-modded-to-6970 at 252W.

I've found it difficult to generate much of an increase with powertune being changed from 0% to +20% really, I'll have to do some more tests. As far as I can tell though, it's always in the 'highest power' setting, not stuck in the limited setting, as performance tests from those programs seem to be about right.

I think we must both be doing something wrong

That article shows power consumption increasing by 50W and Furmark's FPS increasing from 88 to 115 on a 6950-modded-to-6970 at PT+20% . Even with an unmodded 6950 it shows an increase of 20W and 11 FPS.

I have been testing with GPU Tool rather than Furmark, but it still shows GPU activity at 98% and I would still expect to see a difference.

Nothing wrong, overclocking does not increase power consumption per se. But, to get even higher overvclocks, you will eventually need to overvolt, and that's what increases power consumption. It is likely you can overclock quite a bit without overvolting, I do: it seems you have a very good overclocker.

I decided to not overvolt due to the higher temperature increase with two cards, which forced me to go from "silent pc" to "quiet pc": I just prefer silent and slower.


Yeah, GPUz does not give full picture, current sensor seems to be for the GPU core only, excluding memory and maybe all other components. If you have a good power meter that seems the best way to measure



There are other factors besides the "dvi box" getting in the way: you can't know for sure unless you measure distance from the mounting holes to the nearest "obstacle", in your case I think this is the "DVI box". Depending on the layout you can find that the holes are located further away to the right: I think that for my 5850 they were located exactly where your holes are, but I don't have the "dvi box". That's why the 5870 cooler just makes me feel more confident. But it is possible both coolers fit.

To be completely sure, you need to compare areference card that you know for sure is ok for that cooler with your graphics card. You might be able to find an standard layout photo, and carefully compare that layout with your own layout: again, pay attention to holes and obstacles. A lot of work, but that's how I decided that the Sapphire Xtreme 5850 would be able to accomodate the Xtreme Plus.

BTW, why not contacting your graphics card vendor and Arctic Cooling? Maybe they will go the extra mile and check it for you.

But in that article, when they refer to "6950 modded" they mean a shader unlocked 6950, overclocked to 6970 clock speeds. According to this article the 6970 is clocked at 880/1375, so I'm already 20Mhz past the core and almost matching the memory clocks, therefore I would expect my consumption to increase in line with their findings for 6950 modded and 6950 modded+20%. Even before modding the card they found that PT at +20% increased consumption by 20W and FPS by 11.




To put the question another way, as the reference card seems to have the DVI box just like my card, how does the Extreme Plus fit over it? If it stops short of the box, then unless the holes on my card are further left it should fit that too. If anything I'd expect the holes on my card to be further to the right, to allow for the fact that the VRMs have been relocated from the right end to the left end of the card. If true, that would mean that the Extreme Plus would extend past the right end of my card, but stop even further short of the box than on the reference card. I'm guessing that the holes/core are in the same place though, mainly because it makes it simpler and cheaper for Powercolor if they only have to design one cooler which fits several of their cards.



Unfortunately I tried that and Arctic Cooling said they couldn't say as my card isn't reference and Powercolor said they couldn't say as they didn't have an Accelero Xtreme Plus and besides, it would invalidate my warranty if I changed the cooler. I have replied, linking them to the height restriction PDF and asked them if they could check using that (obviously their designs show where and how high the components are, so it's a lot easier for them to check than for me to try and work it out).

I'm not familiar with powertune, because that functionality is not present in my 5850.

That said, what powertune *seems* to be doing is limiting the maximum power consumption by lowering frequencies (and I'm sure voltage) a bit. *If* that's true, then if your settings are such that you are not beyond the default powertune consumption limits, nothing will happen: you'll get always the same performance no matter the powertune limits, and of course you will get the same power consumption. If those guys got better performance by raising powertune limits that's because the card was restricting itself by automatically lowering frequency a bit at default settings: by telling the card to raise the power limit they allowed it not to lower frequency, and therefore got better performance and of course higher power consumption. So, be happy if you are getting maximum performance without increasing PowerTune imposed limits

Now, be careful with Powertune: you are telling the card "increase your maximum power consumption limit by x%". That' especially dangerous if you end up having weaker VRM cooling, because then the reasoning should be more on the lines of "decrease your maximum power consumption limit by...". Besides, the card now seems to "believe" it is a 6970, which we should suppose is more resilient that a 6950, and is probably considering a higher power consumption to be safe.

Yes, we all know that default settings are very conservative so that they work even with a lemon, but be careful.



Very reasonable, but *only* way to know for sure is to find the right photos for a reference card implementation and compare.

Regards

That is more or less how I understand PT to work (IE it throttles back the card if the thermal load reaches a certain point). Although the card is rated to 225W, PT seems to throttle well before that, as the unmodded 6950 went from 169W / 88 FPS to 189W / 99 FPS with PT +20%. Quite why I can't get my card to use more than 150W (AC) I don't know, but I guess the first thing I should try is running Furmark, to see what FPS I get.



As I say, the 6950 is rated to 225W, so it'll be fine with the stock VRM up to that and as it says in the review, even 252W is only "12% beyond the specified power limit will not have any ill effects on your hardware and is well covered by manufacturing tolerances and overspeccing". I take your point about being careful if I replace the VRM heatsink though.

I haven't flashed my 6950 with a 6970 BIOS but just unlocked the shaders with RBE, so it doesn't "believe" it's a 6970 (which it would if I'd flashed it with a 6970 BIOS) and still has the 6950's consumption limits in the BIOS.



I'll certainly try that if the companies don't provide an answer.