Cannot get Dwood Bracket to Work!

[Belial88]

Okay, I'm having a ton of trouble with the Dwood bracket...

Gigabyte 7950 rev1.0 (voltage unlocked version)
H60 2013
H60 dwood bracket w/92mm fan mount

So first off, I see that there's actually some transistors and such that the bracket lays on. I'm guessing this has a lot to do with why some people's cards
'break' when they tighten it too hard, they are actually smashing the transistors. So obviously, use washers.

But that's not even the problem. I just can't get a temp drop using the Dwood bracket. I've got .5mm shims (like what frozencpu sells), I've heard dwood uses 1mm and 1.5mm, but with washers there is a bit more of a gap, although I actually decapped my 7950 to make more room.

At the moment, using AS5 (i know, awful, older than DDR1 paste, just a test run) and both .5mm shims (i know, a single shim would be better), my temps are basically identical to using the stock aftermarket heatsink. That's with the same overclock, and 100% fan speed (3x75mm vs 2x120mm!).

I ordered some 1.5mm shims, hopefully that'll help in creating better contact, and maybe with CLU on top it'll work well....

Could anyone report their temp drops using a dwood bracket or anything like that? I originally used zip ties, but the tube torque was so strong from going from the GPU to a rear mount that it was pulling the block off the die and caused basically 100+ temps (I had my system to shutdown immediately on that so np). Anyone understand what I'm talking about?

You can see the transistors on the lower part above that VRAM and in the bottom left corner.

 

[SickBeast]

I'm not sure that it's worth going from Gigabyte's windforce cooler to an H60. I just don't know if you would see much gain if any. My guess is that it would probably be quieter under full load but the cooling performance would be pretty comparable.

How did you manage to get a dwood bracket? He has not been answering my PMs for the past month.

 

[lavaheadache]

what is wrong with AS5?

 

[T_Yamamoto]

You could always ask dwood himself.

Also, the windforce cooler does quite well on its own...

 

[SickBeast]

dwood is not responding to PMs. He is on some sort of hiatus. He says he's busy with something.

 

[BUnit1701]

what is wrong with AS5?

Nothing at all, but the kids like to act like its as useful as toothpaste for a TIM now. :whiste:

http://forums.anandtech.com/showpost.php?p=34292422&postcount=42 <- IDC's post on AS5 when he was testing a number of TIMs.

http://forums.anandtech.com/showthread.php?t=2285595&highlight=thermal+paste <-- Full TIM testing thread.

http://www.hardwaresecrets.com/article/Thermal-Compound-Roundup-February-2012/1490 <-- TIM roundup

 

[Subyman]

I don't think the bracket is supposed to sit on any transistors at all. Maybe you got the wrong bracket or something. The point of the bracket is to make sure you can't tighten it too much, so if its sitting on transistors that defeats the purpose.

BTW AS5 is good stuff. Just because something is old doesn't mean its bad. Look up Collinite wax, best stuff you've never heard of and old as dirt.

 

[Belial88]

I'm not sure that it's worth going from Gigabyte's windforce cooler to an H60. I just don't know if you would see much gain if any. My guess is that it would probably be quieter under full load but the cooling performance would be pretty comparable.

How did you manage to get a dwood bracket? He has not been answering my PMs for the past month.

The gigabyte windforce heatsink is actually awful. It's significantly worse than any 2x92mm heatsink model, like the Sapphire Dual-X or PCS+, etc etc. It's only better than the 2x75mm models like the XFX DD and MSI TF3, which are really bad in terms of cooling.

It's definitely worth it, most people report 20-40C temp drops. With large air heatsinks, like the Arctic Acceleros, those massive aftermarket air GPU coolers, people usually report 10-30C temp drops.

Definitely the heatsink is significantly quieter with 2x120mm's @ 2K RPM vs 3x75mm @ 4K RPM. The gigabyte windforce is an absolutely awful cooler, because it uses 3x75mm fans, it's just insanely loud and high pitched. In contrast with 2x92mm fan models like the Dual-X, which while still jet-loud at 100%, not as loud as 3x smaller fans (as we all know, smaller fan = higher pitched).

I've used almost every 7950 model, so I do know what I'm talking about here. I wrote the 7950 guide on OCN:http://www.overclock.net/t/1394623/belials-comprehensive-guide-to-7950s/160

For reference, at 1068mhz@1.15v at 100% fan speed, the Gigabyte WF3 hits ~85-90C. The Sapphire Dual-X hits about 80-85C (and not nearly as loud). It might be comparable if you are really bad review site and you compare 'stock' clocks like 925mhz@1.05v, but on any significant overclock, it makes a difference. 7950s are hot GPUs, basically any 7950 is going to overheat on any mild overclock, especially with overvoltage, and usually even on 'stock' settings (especially 'boost' 925@1.25v).

what is wrong with AS5?

What's wrong with DDR1? What's wrong with the Pentium 4? What's wrong with a FX5200?

They are over a decade old. Arctic Silver 5 is literally as old as DDR1 RAM. Just as RAM, CPUs, GPUs, cases, etc has advanced by leaps and bounds over the last 10 years, so has thermal paste.

It is true that the difference between high end, modern thermal paste is minimal, but AS5 is neither high end nor modern by any stretch for over half a decade. It's awful paste, it's worse than the stock paste on most heatsinks.

I've done a ton of thermal paste testing and review, but to put it in perspective, the difference I got between Coolaboratory Liquid Ultra and PK-3 (the best ceramique thermal paste, ie 'standard' thermal paste), was ~8C. PK-3 was about 5C better than PK-1, which for a long time was reigning champion as the best thermal paste around (and considered significantly better than AS5, ie Skinneelabs).

Most review sites do an awful job of thermal paste review because they'll do something awful like test a non-overclock like 4ghz@1.2v - that way they can fit a high end paste or cooler on the same graph as a low end one, but in reality a low performing paste isn't going to be much different until higher temps/wattages.

Like a Hyper 212+ will actually outperform a custom loop, high end water system with a 480mm radiator, on a low overclock or stock settings, and likely at idle too (which is why idle means absolutely nothing). You put a moderate overclock, like say 4.4ghz@1.3v, and you might see a 20C difference. Sure, large, but put it up to 5.2ghz@1.6v, and it'll be a difference larger than 40C (ie the hyper 212+ will overheat and crash while the water loop will be at 70C).

So yea. Arctic silver is awful, awful, awful paste. There is no reason to be using it when you can buy some of the highest end thermal paste ceramiques, like PK-3 or Masscool, for $3 for a 1g tube (enough to last about 10-20 CPU applications, ie most people's lifetimes...). On top of that, getting some extreme TIM like Coolaboratory Liquid Ultra for $16 or Hegrease for $9 is definitely one of the best values you can find. For example, a $20 Hyper 212+ vs a $40 H60, will be about a 4-8C temp drop, but getting CLU will easily yield double that temp drop on a high overclock.

Like I said, the difference between modern, high end pastes is negligible (pk-3, gelid, masscool, shin etsu, mx-4, tq-4, nh-t1). But AS5 is not modern at all. Why in the world you wouldn't spend less than $5 to get a performance boost larger than a 2nd fan or a step up in cooling is a bit silly. Of course, on a low, near-stock overclock the deltas are going to be a lot smaller.

/rant

 

[Belial88]

You could always ask dwood himself.

Also, the windforce cooler does quite well on its own...

Actually, you can't ask dwood. He only responds once a month, his website has been sporadically been shutting down recently, and he no longer is selling anything.

And the windforce cooler is absolutely awful. Maybe if you go by bad cooler reviews that test 'stock' settings, even though 'stock' settings on every 7950 is different and use different fan speeds, than sure, it's A+++! just like every other 7950 (gee, let's rank every model as awesome so as to please the manufacturers so they send us more stuff, rather than make honest reviews! Yea, of course, every single 7950 model is A+, none of them is better or worse than the other even though they use wildly different components!).

I won't mention names, but it's an absolute joke when sites compare the exact same Tahiti chip, on different model cards, with different fan settings and clock settings. A windforce on 925mhz@1.05v is surely going to seem cool and quiet compared to a 950mhz@1.25v card, it's just stupid. You need apples to apples settings, at 100% fan speed (or, if you got a dBa tool, at the same sound levels, though that doesn't factor in pitch) and then post the sound levels. The Gigabyte Windforce 7950 cools worse than every other 7950 model, except the XFX DD and MSI TF3 which use just 2x75mm fans.

Of course, on the 'stock' settings that bad review sites test, you are only going to see 1-4C temp differences between the different models, especially with the auto fan profiles. But put the same clock settings on every card, and set it to 100% fan speed, you'll see very large differences between the different coolers, about 5-15C difference.

Nothing at all, but the kids like to act like its as useful as toothpaste for a TIM now.

Because it is... why would you use something as old as DDR1 RAM? You don't use a TRUE 120, an IDE HDD, 80mm fans, or a LGA775 cpu, so why would you use just a terrible paste, when you can get some of the best paste around for under $4 shipped? You can even get free samples from most companies, which is enough for at least ~5 applications.

If you were building some bottom barrel, $50-100 system, sure, use the free AS5 you got at microcenter. But if you spent $300 on your CPU, and $50+ on your cooling, why would you not spend $5 on your paste? It's about diminishing returns, and thermal paste is some of the best values you can get in cooling.

A 2nd fan, ~1-3C temp drop (2-5C on water rads). Going from low end cooler with 4 heatpipes like hyper 212, to mid-range cooler like 6 heatpipe large towers or 120mm closed loops, will cost about $20+, and only give you about a 5-10C temp drop, which is about the same a $5 tube of paste will get you.

If you are doing some mild, basic overclock, sure it's not a big deal (in which case, buying anything more than a $19 hyper 212+ is a waste of money), but for any moderate or high overclock, it's just a huge value.

BTW AS5 is good stuff. Just because something is old doesn't mean its bad. Look up Collinite wax, best stuff you've never heard of and old as dirt.

Collinite wax has been tested for almost a century. We are definitely seeing smaller and smaller gains in thermal paste performance (likewise in CPU, RAM, GPU, etc), but I'm sure in ~80 years, there will be some paste that stands out as a reliable, time-tested favorite.

Also, lubrication is completely different than thermal conductivity + ease of use.

AS5 is awful stuff, perpetuated by people who don't know what they are talking about. There isn't even any review that says AS5 is good, and definitely not any modern review that says it's good. You have tons of reviews saying it's bad.

What you have though, is a lot of misinformed people regurgitating the same bad information from bad sites, saying thermal paste doesn't matter. In which case, sure, as long as you use some modern, high end paste, it really doesn't matter what you use. PK-3, Shin Etsu, Masscool, Gelid, NH-T1, Tuniq, MX-4, they all perform similiarly. But older pastes, like PK-1/PK-2, MX-2, TQ-2, perform noticeably worse. Even older pastes, like AS5, Ceramique, no-name pastes... horrible stuff responsible for 10C+ temp increases on higher overclocks.

It's like if you read a review of heatsinks. Say, the H100, most review sites will say it's maybe just 1-3C better than, say, the NH-D14. Put it on an actual overclock, instead of a terribly misleading 4.2ghz@1.2v, and like 4.6ghz@1.4v, and you'll see a difference more like 5-10C. Same with thermal pastes.

Trust me, I used to think buying stuff like Coollaboratory Liquid Ultra was totally stupid, that any standard paste would perform similarly, that paying almost $20 for a thermal paste that might give me just 1-2C more, is a total waste of money. Turns out the average temp drop on a delidded chip using Liquid Ultra, instead of a high end ceramique (never mind something awful like AS5 or MX-4), is about 20-30C, compared to just 5-15C. Huge difference.

 

[Belial88]

I don't think the bracket is supposed to sit on any transistors at all. Maybe you got the wrong bracket or something. The point of the bracket is to make sure you can't tighten it too much, so if its sitting on transistors that defeats the purpose.

It definitely sits on transisitors... If you look at pictures of any 7950 PCb, or even 670 pcb, or any other card, you'll see the dwood bracket sits on transistors.

There's no way the bracket wouldn't sit on them. The transistors are INSIDE the mounting holes, as well as straight between them. There is no way, no matter how the bracket is designed, that it wouldn't sit on them (the h60 bracket, the 620 one won't since it doesnt sit on the pcb).

Definitely not the wrong bracket, there's only 2 brackets, the 620 bracket and h60 bracket, there's no way I got the 'wrong bracket'.

As for making sure you can't tighten too much, I assume you mean it's because you can't reach the screws once the block is attached to the bracket. I suppose with a large enough shim, you wouldn't be able to reach the transistors... but even with 1mm in shim, I can reach the transistors, and it isn't enough to make full contact with the die.

We'll see what happens when the 1.5mm shims come in.

edit: before people ask why I got the gigabyte 7950, i got it because it's the best 7950 model... the first revision, that is. There are 2 different Gigabyte 7950s, the older one is 6+6 power using a 7950 reference PCB, no VRM heatsinks (easy/cheap fix, chipset sinks are like $5 for 40 pc), and the newer one (like if you bought in last ~5 months) is 8+6, using a custom 7950PCB (NOT 7970, not all 8+6 are 7970), VRM heatsink (dedicated, vrm sink rather than plate), and is voltage locked by using a different PWM chip.

The older gigabyte 7950, and the Asus DC2 7950s, are the only 2 versions have Memory Voltage unlocked, VRM Temp sensors (VRM tends to overheat before core, yet has a similar temp limit), and Hynix VRAM (~1900-2000mhz ram speed vs ~1400-1500 of Elpida 50f and 1600-1700 Elpida 60f - not to mention tighter timings so it's faster clock for clock). I do scrypt litecoin mining, which actually cares more about memory clock than core clock, so memory overclocking to me is literally more important than core overclock in my usage. The Asus is also more expensive, and has a weird, uneditable BIOS that is suboptimal (so like if you want increased stability, not have to use a resource-intensive software program to have to be on for your overclock to be applied, want to overclock on Linux, or if you want to have your overclock applied while not plugged into any monitor since mining while plugged into an active monitor will make your computer unusable). It also uses higher quality mosfets and chokes compared to most other 7950 models.

The downside is that the Windforce is an awful heatsink, but when you get down to it, every 7950 cooler is inadequate for overclocking, especially overvolting, even at 100% fan speed, which is insanely loud on the 75mm,92mm, or even 100mm asus fans, and since I'm replacing the heatsink anyways, I didn't care and can just sell it on ebay for $30-40. The asus dc2 is the only cooler that comes close to being adequate for cooling 7950s, but like I said - still louder than 120/140mm fans, still not nearly as good as an closed loop mod or the larger aftermarket air heatsinks (ie acceleros), more expensive, and uneditable BIOS.

You can read all about the differences in the 7950s here in my guide: http://www.overclock.net/t/1394623/belials-comprehensive-guide-to-7950s/160

You'd be surprised, but there's actually major differences between all the 7950 models, something that all of the review sites fail to cover. Now I get that they are all 7950s, and what difference does 400mhz VRAM speed make (about $10/mo, actually, quite a dramatic difference frankly), but even for just gaming, why would you pick a sub-optimal 7950? The TF3 is best for aesthetics and build quality but is insanely loud/poor cooling, the Dual-X is a better cooler but no VRAM voltage control (so better for gaming), the HIS IceQx2 is actually an extremely good card for overclockers/gamers, probably the best choice but is slightly more expensive and sometimes is hynix (unfortunately it's 7970 pcb, which is great for most users but awful for scrypt mining like litecoin, as it behaves like a 7970).

Also be aware a lot of 7950s use a custom, cheapo 7950 PCB - a specifically, tailor made 7950PCB that actually is made to cut costs and uses significantly lower quality components (TF3, Dual-X).

tldr: Core Voltage, Memory Voltage, VRM Temp Sensor, Hynix vs Elpida RAM, VRM Heatsinks, and PCB are what make one 7950 model better than another. While when you get down to it, what difference does it make if it's all the same 7950 core, in a game the best card and the worst card still is only a 5-10FPS difference, but hey, if they are all the same price, +/- $20-40, why would you not pick the better card, OR conversely, pick a cheaper card that has the features you care more about. However since I'm a scrypt miner, these actually make a huge difference to me, about $10/mo increase in money made, so it's worth the increased cost.