Belial's Comprehensive Guide to 7950s!

[Belial88]

Why are we talking about Furmark again?

Because it comes somewhere close to the utilization and stresses of GPGPU work, which is what most 7950s are used for? Maybe some people want to know the limits of their hardware, after all, they paid a lot of money for it and want to make sure they got their money's worth?

Or, maybe, some people simply want to never crash or artifact during a game? My ladder points are important, I'm not going to lose them because my GPU chocked up.

I also play in tournaments (for money), and I stream, so on top of all of this, yea, I cannot have a hardware failure. Not everyone is like you. If you don't care about rock stability, you don't need to run OCCT error testing.

I am simply providing information, on how to correctly use and analyze a GPU, a 7950 specifically here. Now you know how to stress and heat test your 7950. If you dont care to stress test, then you dont need to do it. Simple as that.

Because OP insists that 50% power limit is a must. I contend it only makes a difference for pathological use cases like Furmark, and demonstrated that in Unigine Valley (i.e. gaming) it makes no appreciable difference +50% versus +20%. Whereas it did make an appreciable difference in Furmark (lol).

It does make a difference, it's been proven plenty of times. The difference will be much smaller in games, sure, but it will still be there in intense moments of a game like Crysis3, that .0001% of moments where, if you suddenly have performance reduction, it'd really suck. It also makes quite a sizeable difference in any GPGPU work, which is what most 7950s are used for. Not everyone who buys a 7950 uses it for gaming. I know I didn't, and I know most people don't.

In fact, I really don't know many people who own a 7950 and don't mine with it, simply because it's free money to make if you have it (though this is less true nowadays due to price dips).

Otherwise, I'd say logical, not pathologic, because it makes no sense to put a throttle on your card. If you are worried about heat or power consumption, just reduce your card clocks & voltage so you get a more even and higher performance than a throttled higher clock. You don't put throttles on your CPU. It makes no sense to if it's completely unnecessary. No one goes and installs a governor on their sports car if it didn't come with one. Likewise, no one goes and puts power limits on their CPU or other GPUs.

No one puts in a 200w power limit on their i7-4770k and says 'well it'll likely never hit 200w, even though it could sometimes'. Because that'd be retarded. They set it at 300w+ and never think about it again. They definitely do not say "well i dont stream h264 so i dont need to raise it above 200w"

I mean if you aren't fully utilizing your card to the max, then you wasted your money on something that's more power than you need.

 

[horza]

..

 

[Bubbleawsome]

I see most 7950s being used for gaming/GPGPU/DC computing fairly equally. If one had to have an advantage I would choose DC computing, but I am mainly involved in those communities.

 

[Belial88]

OK, to start with I'd love to see some GPGPU benchmarks with +20 vs +50. I'm just speculating but I'd be surprised if the difference was sizeable.

Addressing the bolded comment, I'm almost certain you're mistaken here but I could be wrong.

I've said +0 vs +30+ is sizeable, not 20 vs 30-50. There is still a difference in 20 vs 30-50, not huge, but it is there.

I've already linked and posted pictures, including in the guide, on the differences of 20 vs 50. So part you being wrong, part you misreading my comment (which, sure, is easily misreadable). There's a large logical aspect of this too - just remove the governor from your card. It makes no logical sense to have a governor, or power limit throttle on your card, even if it's only going to kick in .00001% of the time. You set your CPU power limits to 100000w and 10000a, likewise you make sure your GPU never hits it's power limits. You definitely don't set a 200w i7-4770k at 205w, much less 199w.

I see most 7950s being used for gaming/GPGPU/DC computing fairly equally. If one had to have an advantage I would choose DC computing, but I am mainly involved in those communities.

Most 7950s ≠ Most 7950 Owners.

That said, the balance has tipped since profitability has dropped (plummeted, maybe?) and new releases of cards, but profitability has been dramatically rising again recently too.

 

[Subyman]

I think it is obvious that the OP's claim that a user will see a 40% drop in fps/performance in benchmarks and games while using a 0% power tune compared to a 50% power tune is not backed by real world data.

The statement should be rephrased to say "To see absolutely no throttling in any application a higher power tune is mandatory."

 

[BallaTheFeared]

I never had an issue mining my cards overclocked while using less than 20% on powertune.

That said who wants to waste their time modding a card or regedit for the power tune slider?

My 7950 has a 900w power tune TDP.

 

[Belial88]

I think it is obvious that the OP's claim that a user will see a 40% drop in fps/performance in benchmarks and games while using a 0% power tune compared to a 50% power tune is not backed by real world data.

The statement should be rephrased to say "To see absolutely no throttling in any application a higher power tune is mandatory."

Wut? Did you not read all the links, pictures, and evidence provided? Are you seriously that dense?

A lot of things change power draw, such as temps, but you should see around a 40% difference on air cooling with 0% vs 40% on a 1ghz@1.25v clock.

I never had an issue mining my cards overclocked while using less than 20% on powertune.

That said who wants to waste their time modding a card or regedit for the power tune slider?

My 7950 has a 900w power tune TDP.

It takes 5 seconds to do... And saying you've 'never had issue' simply means you have no clue what it's supposed to be like. As for your gpu flasher image, the card definitely isn't 900w TDP.

Warning issued for personal attack.
-- stahlhart

 

[Haider]

Belial, do you spend alot of your time playing Furmark? I know I dont...

Don't be absurd! Lots of people do not play Prime95 or Linpack or Memtest but they are used for stress testing.

 

[Haider]

My cooling appears to be limited by how quickly my radiator can dissipate heat. The Fujipoly at 0.5mm thick is plenty to have proper die contact with my CLC water block. For the sake of comparison, I did try using TIM+shim+TIM, but it made no appreciable difference for me in idle or load temps. For reference, the shim was an EK 7970 copper shim and the TIM was Arctic MX-2, properly applied.

Cooling aside, I will concede that for pathological loads (read: Furmark) a higher power limit can be of benefit. I experimentally determined that a power limit of 50% (or actually, even 40%, YMMV) will prevent throttling versus a power limit of 20%. However, I would caution the average user who doesn't have proper cooling of GPU and VRMs against doing this, as even with my overkill VRM cooling (heatsinks + 120mm Scythe fan @ 90%) I was hitting 79C on the VRMs under Furmark as opposed to staying under 60C.

50% and 40% were within margin of error of each other, with no observable throttling in Furmark or in GPU-Z.

50%:


40%:


30% and 20% were hitting power limits and thus being throttled back.

30%:


20%:


Given my previous Heaven benchmark results showing negligible differences between 50% and 20% power limits and displaying no evidence of throttling in GPU-Z, I still wouldn't use 50% power limits for gaming without extreme cooling. For pathological use cases like Furmark or extreme benchmarking, it seems to eliminate another variable and can be used with proper cooling.

Interesting so if the game pushes the GPU enough then it will throttle due to available power. So one eye to the future you could increase to +50 as long as you upgrade the cooling system appropriately. For me I concentrate on the delta between peak frame-rate to minimum frame rate with a minimum bar set at 30fps. Even playing 1080p Crysis 1 with 90% gaphical options set to max I got some areas where it could not keep a 30fps frame-rate. I have a save game around that area. HD 7950 Vapor-X@850MHz, i5-760@3.6GHz, Asus P7H55/UBS3, 8GB DDR1600 Crucial Ballistix Elite, 128GB Sandisk SSD, 2TB Seagate Barracuda. So it's not what you would call puny. I'm trying to find how much I can OC Vapor-X whilst maintaining temps, stability and now throttling. I'll try Furmark and the other stress test tool coupled with monitoring in GPU-Z. Will have to wait a week or two for me to comeback with results.

 

[Haider]

Because OP insists that 50% power limit is a must. I contend it only makes a difference for pathological use cases like Furmark, and demonstrated that in Unigine Valley (i.e. gaming) it makes no appreciable difference +50% versus +20%. Whereas it did make an appreciable difference in Furmark (lol).

Why do people talk about Prime95, Linpack & Intel TAT when stress testing their CPU overclock? All of those stress your CPU way more than any game.

 

[Belial88]

Interesting so if the game pushes the GPU enough then it will throttle due to available power. So one eye to the future you could increase to +50 as long as you upgrade the cooling system appropriately. For me I concentrate on the delta between peak frame-rate to minimum frame rate with a minimum bar set at 30fps. Even playing 1080p Crysis 1 with 90% gaphical options set to max I got some areas where it could not keep a 30fps frame-rate. I have a save game around that area. HD 7950 Vapor-X@850MHz, i5-760@3.6GHz, Asus P7H55/UBS3, 8GB DDR1600 Crucial Ballistix Elite, 128GB Sandisk SSD, 2TB Seagate Barracuda. So it's not what you would call puny. I'm trying to find how much I can OC Vapor-X whilst maintaining temps, stability and now throttling. I'll try Furmark and the other stress test tool coupled with monitoring in GPU-Z. Will have to wait a week or two for me to comeback with results.

Vapor-x is still a 7950, as bad a model it is. Your overclock results are purely limited by your cooling, and all aftermarket air coolers really are going to hit around 1.15v as the max they'll do under 90C.

Hwinfo is a lot better than using GPU-Z for monitoring. OCCT Error is also better than Furmark (basically identical test but OCCT Error gives better options to stress GPU and test for artifacts).

Crysis3 is the only game right now that'll ever really make a 7950 hurt, your results sound right.

Why do people talk about Prime95, Linpack & Intel TAT when stress testing their CPU overclock? All of those stress your CPU way more than any game.

Because these programs are useful for testing CPU stability? Not everyone games, many people do a lot more than game, like streaming x264 while gaming, or they need 100% stability and can't afford a .00001% chance of crash or slowdown like if they play tournaments or ladder-ranked games.

And there's also CPU intensive games. I mean no one is really talking about this stuff in this thread except as analogies, but having a stable system is a combination of factors, making sure your CPU, System/RAM, and GPU are all stable together.

 

[monkeydelmagico]

My die isn't recessed on the Powercolor 7950 Boost. Using a copper shim would not help.


Works fine and I overclocked to 1150/1700 with a +20% power limit and have run verified Heaven loops.

Same here. No shim, Cooler master seidon zip tied to a powercolor 7950 with some mx-4 paste and bingo 1170's with a +20% power limit. Also no problems running furmark.

Alot of good work here Belial but your clearly not interested in listening to what others are trying to tell you so oh well.

 

[BUnit1701]

Don't be absurd! Lots of people do not play Prime95 or Linpack or Memtest but they are used for stress testing.

Yea, but do you argue for altering the system to improve performance in those stability tests?

 

[Haider]

Yea, but do you argue for altering the system to improve performance in those stability tests?

Of course, if you fail a prime95 15min run you make changes to make your overclock stable i.e. voltage or lower the OC. Once you hit the speed you want to achieve and pass the 15 minute run, you leave it on an overnight run. If you pass that you should be good to go. If it's like a extreme overclock that you want to use for say a couple hours then put it through a two hour run. See if it's stable for those hours and doesn't shoot the temps too high.

 

[Haider]

Vapor-x is still a 7950, as bad a model it is. Your overclock results are purely limited by your cooling, and all aftermarket air coolers really are going to hit around 1.15v as the max they'll do under 90C.

Hwinfo is a lot better than using GPU-Z for monitoring. OCCT Error is also better than Furmark (basically identical test but OCCT Error gives better options to stress GPU and test for artifacts).

Crysis3 is the only game right now that'll ever really make a 7950 hurt, your results sound right.



Because these programs are useful for testing CPU stability? Not everyone games, many people do a lot more than game, like streaming x264 while gaming, or they need 100% stability and can't afford a .00001% chance of crash or slowdown like if they play tournaments or ladder-ranked games.

And there's also CPU intensive games. I mean no one is really talking about this stuff in this thread except as analogies, but having a stable system is a combination of factors, making sure your CPU, System/RAM, and GPU are all stable together.

I use HWInfo for all the other temps. Real temp for CPU. I like the GPU-Z VRM temp, GPU temp, utilisation and speed. It's how I found out without the +20 the GPU speed was throttling at 950MHz. In the end I had to change PC case to keep the temps in check at +20%. I'm looking at getting a better front case fan with 50% better CFM AND static pressure to blow across the vrms and GPU whilst only being a couple of dBAs than the stock one in the Define R4.

 

[Belial88]

Same here. No shim, Cooler master seidon zip tied to a powercolor 7950 with some mx-4 paste and bingo 1170's with a +20% power limit. Also no problems running furmark.

Alot of good work here Belial but your clearly not interested in listening to what others are trying to tell you so oh well.

You mean, others are not interested in listening to what everyone has said? There's a reason every single 7950 stock cooler, aftermarket cooler, water block, and custom mod uses a shim. Because there's a gap. Gunking up that gap with thermal paste is fine, the gap is smaller than 0.5mm most of the time, but it still exists. You realize a ~0.3mm gap is too small for you to notice with your eye right? Saying 'well i dont need a shim cuz it works without one' is not evidence at all. All it means is that a CLC is way powerful for cooling a GPU.

Sure, technically you dont 'need' a shim. But, there IS a gap, and while you can still have super cool 7950, 40C overclocked without one, it's pretty stupid not to use a shim when it's less than a dollar. I'll friggin send you one for free if you ask.

AMD literally designed it to be a gap, so people wouldn't crush the die with blocks or aftermarket heatsinks and RMA their card every day like they used to. There is a gap, and I'll email AMD and ask them for technical specs to prove it. Jesus christ. I also got an extra shim I can give you, if you'll promise to post before/afters.

Also, mx-4 is awful, get a better paste. You can get a 1.5g tube of PK-3 or Masscool for ~$3, that's enough for about 20-30 applications (ie a lifetime if you use it right, less is more), or just get CLU for $12 at performancepcs right now for a dramatic difference.

As for you 'not having problems running furmark', pull up graphs on HWInfo, and then see furmark. You'll see minor throttling with +20% power limit compared to +50%. This has been well documented, I've already provided tons of pictures and evidence yet no one wants to address that or provide counter-evidence.

Literally, IEC just proved power limit throttling occured at 20% and still people won't listen to evidence or what has been proven long ago on other forums. Anandtech and Tomshardware forums are always pretty far behind...

Yea, but do you argue for altering the system to improve performance in those stability tests?

...wut? Everything we do is altering the system. Overclocking, installing an OS, changing the settings like Aero theme, fonts, borders, what browser you use. It's your system. Of course you alter the system, you want to alter it, for max performance... Are you serious? Go buy a mac, seriously (mac has their place, dont get me wrong, but the hardware and firmware is pretty locked is my point).

I use HWInfo for all the other temps. Real temp for CPU. I like the GPU-Z VRM temp, GPU temp, utilisation and speed. It's how I found out without the +20 the GPU speed was throttling at 950MHz. In the end I had to change PC case to keep the temps in check at +20%. I'm looking at getting a better front case fan with 50% better CFM AND static pressure to blow across the vrms and GPU whilst only being a couple of dBAs than the stock one in the Define R4.

You're better off using +50 power limit and just reducing your core clocks and voltage to where they need to be than using +20. +20 vs +30+ isn't huge so you shouldn't have to downclock that much.

The stock Fractal Design fans are actually really good, but just low speed. The vast majority of fans are utter crap (especially most stock and popular fans, ie noctua, nzxt, coolermaster, etc), but those are really good fans. Of course, they are low speed and low speed can only move so much air, but you should understand that you aren't going to get any more cooling unless you ramp up the fan speed, which means louder. Which is fine, but just realize your fans are so good already that you wouldn't see much improvement going to anything else.

Adding a medium speed fan usually accounts for about 1-2C, if that, and only about the first 3-4 fans do that. So upgrading a fan isn't really going to give you much improvement man. Sorry to say, but the only way you're going to get more cooling is to replace your cooler or up your GPU fan speed if you haven't already.

What clock and voltage are you at? A sapphire-x should be doing a decent ~1.15v without going over 90C, which should give you decent room to play around with and overclock, even with +50 power limit.

The best fans of all are the Gentle Typhoons AP14/AP15, as well as the other Scythe fans (ultra/kama kaze), but the yate loons really aren't much worse and the best value fans of all by far. There aren't many other fans that come close to those 2 in value, most fans are pretty crappy unfortunately (a few exist - vortex, sp120s, thermalright TYs). If you upgrade your fractals I'd recommend you just get some yates.

 

[Haider]

What I found in benchmarking was that running at 1050MHz at +20 makes my GPU temps hit 84C and the GPU start to go into thermal protection mode i.e. throttle down. Whilst playing Crysis 1 I was hit a peak of 70C and median was about 67C. So first things first is that thermals need improvement if I want to run harder and/or faster.

The two games I am playing at the moment are Crysis 1 and Supreme Commander 1. By approximately what factor are the two benchmarks stressing my graphics card more than the two games i.e. 25%? Then I can calculate how much 'fat' I want to build into the system.

On the fans from I have one Silent Series R2 140mm for exhaust and one for front intake. It (front intake) is covered by a filter and therefore does not take in as much as the rear un-obstructed exhaust fan puts out. The Prolimatech Vortex Aluminium can provide upto 50-60% more CFM and static pressure. I will have this mounted in the top front fan slot controlled via the case fan controller and move the existing Fractal Design fan to the lower front intake position - this is currently vacant. Hopefully this will increase the air pressure in the case and I will running a positive pressure system. Though I think I will have to add another fan TBH to the bottom intake next to the PSU. Obviously I do not want to increase the pressure inside the case to the point where it explodes but enough to get air moving and escaping out of any gaps etc.

The only thing I changed on the card is Powerli +20, nothing else but from my A Level Physics: -

p=vi
v=ir
p=i squared r where p=power, v=voltage, r=resistance and i=current

therefore if power is being increased voltage and/or current must also be changing; I believe increasing. Generally for most materials as temps increase resistance increases too. I reckon allowing the power to increase by 20% must allow the current to increase by a certain factor too. So although I have not altered my voltages and current in Trixx but changing the powerli slider to +20 in effect has done that.

 

[BUnit1701]

Of course, if you fail a prime95 15min run you make changes to make your overclock stable i.e. voltage or lower the OC. Once you hit the speed you want to achieve and pass the 15 minute run, you leave it on an overnight run. If you pass that you should be good to go. If it's like a extreme overclock that you want to use for say a couple hours then put it through a two hour run. See if it's stable for those hours and doesn't shoot the temps too high.

Not improving stability, improving performance. There is a difference. And what Belial is arguing for is the second.

 

[Stuka87]

What I found in benchmarking was that running at 1050MHz at +20 makes my GPU temps hit 84C and the GPU start to go into thermal protection mode i.e. throttle down. Whilst playing Crysis 1 I was hit a peak of 70C and median was about 67C. So first things first is that thermals need improvement if I want to run harder and/or faster.

The two games I am playing at the moment are Crysis 1 and Supreme Commander 1. By approximately what factor are the two benchmarks stressing my graphics card more than the two games i.e. 25%? Then I can calculate how much 'fat' I want to build into the system.

On the fans from I have one Silent Series R2 140mm for exhaust and one for front intake. It (front intake) is covered by a filter and therefore does not take in as much as the rear un-obstructed exhaust fan puts out. The Prolimatech Vortex Aluminium can provide upto 50-60% more CFM and static pressure. I will have this mounted in the top front fan slot controlled via the case fan controller and move the existing Fractal Design fan to the lower front intake position - this is currently vacant. Hopefully this will increase the air pressure in the case and I will running a positive pressure system. Though I think I will have to add another fan TBH to the bottom intake next to the PSU. Obviously I do not want to increase the pressure inside the case to the point where it explodes but enough to get air moving and escaping out of any gaps etc.

The only thing I changed on the card is Powerli +20, nothing else but from my A Level Physics: -

p=vi
v=ir
p=i squared r where p=power, v=voltage, r=resistance and i=current

therefore if power is being increased voltage and/or current must also be changing; I believe increasing. Generally for most materials as temps increase resistance increases too. I reckon allowing the power to increase by 20% must allow the current to increase by a certain factor too. So although I have not altered my voltages and current in Trixx but changing the powerli slider to +20 in effect has done that.

i didn't see you mention this, but which 7950 do you have? My card has never gotten anywhere close to that temp, even when I tried clocking it higher than it could handle.

 

[Haider]

i didn't see you mention this, but which 7950 do you have? My card has never gotten anywhere close to that temp, even when I tried clocking it higher than it could handle.

I was running Furmark at 1080P and it hit 84C on Sapphire Radeon HD 7950 Vapor-X@1050MHz, the card fans were doing 100% and it downclocked to cool off...No heat issues in Crysis 1 runs at about 66C at 1080P with everything turned to highest setting...

 

[Bubbleawsome]

I was running Furmark at 1080P and it hit 84C on Sapphire Radeon HD 7950 Vapor-X@1050MHz, the card fans were doing 100% and it downclocked to cool off...No heat issues in Crysis 1 runs at about 66C at 1080P with everything turned to highest setting...

Vapor x is a good card. Maybe faulty?

 

[Belial88]

What I found in benchmarking was that running at 1050MHz at +20 makes my GPU temps hit 84C and the GPU start to go into thermal protection mode i.e. throttle down. Whilst playing Crysis 1 I was hit a peak of 70C and median was about 67C. So first things first is that thermals need improvement if I want to run harder and/or faster.

Yes, as I've said, even the aftermarket heatsinks@100% fan speed really aren't adequate. You simply need to tune down your overclock/voltages. I recommend you run the +50 power limit with your card slightly downclocked. Max temps are about 80-85C so hitting 84C in stress test isn't too bad, though it'll likely be slightly higher with +50 since your card will be utilized more.

The two games I am playing at the moment are Crysis 1 and Supreme Commander 1. By approximately what factor are the two benchmarks stressing my graphics card more than the two games i.e. 25%? Then I can calculate how much 'fat' I want to build into the system.

Crysis1 is pretty dependent, SC isn't. Not much, even Crysis3, which is very GPU heavy, isn't stressing 100% all day long.

If you are asking as a benchmark, games are pretty poor for benchmarking due to not really stressing the card hard and consistently. I'd really recommend using OCCT or Furmark as a benchmark. You can use 3dmark to compare to other results, but it has a lot of problems, like the fact it's as CPU heavy as GPU heavy so CPU ends up influencing the score a lot.

On the fans from I have one Silent Series R2 140mm for exhaust and one for front intake. It (front intake) is covered by a filter and therefore does not take in as much as the rear un-obstructed exhaust fan puts out. The Prolimatech Vortex Aluminium can provide upto 50-60% more CFM and static pressure. I will have this mounted in the top front fan slot controlled via the case fan controller and move the existing Fractal Design fan to the lower front intake position - this is currently vacant. Hopefully this will increase the air pressure in the case and I will running a positive pressure system. Though I think I will have to add another fan TBH to the bottom intake next to the PSU. Obviously I do not want to increase the pressure inside the case to the point where it explodes but enough to get air moving and escaping out of any gaps etc.

Fan specs are never honest (i believe there's only been one 'honest' fan and the official specs are abysmal but it's okay as a fan). It does not provide 'up to 50-60% more CFM and static pressure', I can promise you that (and compared to what? See: Marketing). The Vortex fans aren't particularly noteworthy compared to their peers. In general 140mm fans aren't very good though, besides from pure airflow (ie poor cfm/dba).

Don't worry too much about positive, negative pressures, just put fans where it's blowing on what needs to be blown, and as long as your pressure isn't neutral you're good. Pressure is pretty moot point otherwise, you really should not worry on it. Each additional fan to about 4-5 adds 1-2C, from then on it's really dependent on where it is (ie heatsink vs bottom intake or top exhaust).

That said a fan on the bottom intake is good for GPU cooling. Adds about 1-3C.

herefore if power is being increased voltage and/or current must also be changing; I believe increasing. Generally for most materials as temps increase resistance increases too. I reckon allowing the power to increase by 20% must allow the current to increase by a certain factor too. So although I have not altered my voltages and current in Trixx but changing the powerli slider to +20 in effect has done that.

Yes, you are right. Currently 7950s at stock are sold as 250w cards and will throttle above 250w. However, a simply 'stock' 950mhz@1.15v card will easily consume over 300w on full load, so they heavily throttle. 250 x 1.2, ie +20% power tune, means 300w limit, but then something as simple as a change of 5C ambient temps will cause your power draw to increase roughly 10-20w, enough to put your not-throttling 300w card to throttling.

Have an actually overclocked card, or one of those 'Boost' BIOS where the stock voltage is 1.25v (or set it to 1.25v via overclock), and you'll consume more like 340w. You'll find heavy throttling at stock 0 power limit, and find considerable throttling at +20%. It's not common to throttle after +35 power limit, but it's definitely possible if you do a heavy overclock on water, for example.

You can actually reduce power draw by increasing fan speed/cooling though (a 120mm fan consumes ~2-3w at 100% 2k RPM, so half of that by going from 50 to 100% is about 1-2w, much less than the difference of 20-40w from having 10C higher temps).

i didn't see you mention this, but which 7950 do you have? My card has never gotten anywhere close to that temp, even when I tried clocking it higher than it could handle.

You will definitely see 80C+ on a Sapphire 7950 on 100% fan speed on a proper stressor like OCCT, Furmark, or GPGPU work like mining.

Even the Sapphire Vapor-X will hit that. It's cooler really isn't much better than the normal Sapphire Dual-X, it's only slightly larger but you're talking an couple extra fins. Even the Vapor-X heatsink is still about half the size of a Hyper 212+, it's like a stock CPU heatsink in terms of cooling capability.

I've used the Vapor-X, multiple ones. And a card running hot isn't a 'faulty' card, a faulty card doesn't magically run hotter. A heatsink can't be faulty. It can be mounted improperly perhaps, but that's easily fixable and not really heard of outside of the Asus V1 7950.

 

[Belial88]

Bear in mind that even aftermarket heatsinks like the Vapor-X are roughly the same surface area and FPI as the stock CPU coolers. You would never recommend someone to run a 200w overclock on a Haswell or FX CPU, yet you constantly see people recommend 300w+ overclocks on graphics cards.

Just like the stock cooler, you are going to run hot when pumping 200-400w through the heatsink, especially if its the size of the stock CPU cooler. Putting 3 loud fans on it helps, but not by much.

 

[Haider]

Bear in mind that even aftermarket heatsinks like the Vapor-X are roughly the same surface area and FPI as the stock CPU coolers. You would never recommend someone to run a 200w overclock on a Haswell or FX CPU, yet you constantly see people recommend 300w+ overclocks on graphics cards.

Just like the stock cooler, you are going to run hot when pumping 200-400w through the heatsink, especially if its the size of the stock CPU cooler. Putting 3 loud fans on it helps, but not by much.

Now we're on the same hymn sheet. My Enermax ETS-T40-TB HSF is a lot bigger. Vapor-X cooling system hasn't really delivered in terms of performance.

My backup for this was situation was Alpenfoehn Peter 79XX Edition with 2 x 120mm Noiseblocker BlackSilent Pro Fan PL2 or 2 x 140mm Prolimatech Vortex fan. The Sapphire Vapor-X isn't reference design, I would welcome input on this. As I haven't really taken a graphics card apart a bit scared...

 

[Belial88]

You can look up the PCB on EK's coolingconfigurator.com. THe vapor-x shows to be a custom pcb.

Air cooling doesnt mate like full block water, so I can't imagine a problem. It says it's compatible with the 79xx. Be aware you will need a shim, but from the pictures it almost looks like it already comes with a shim... can't tell though.

Those massive aftermarket GPU heatsinks actually do a pretty decent job, a good 20-30C temp drop.