My XFX 290X teardown [56k]

[Solaris17]

Well my thermals on the ref design were like everyone elses and I kinda got tired of hitting 95C every time I wanted to watch a GIF So I decided to buy some MX-4 (after grilling prolimatech and noctua pastes) suck it up and get to work.

First the beautiful card shes hot in more ways then one ya dig?

Her top is cute she looks good in black.

XFX as well as others play hard to get and feel the need to include warranty stickers on the GPU clamp screws. These aren't normal stickers. They have a plastic kind of build and they crack and shatter when folded. To remedy this I used a hair dryer to warm the adhesive before removal.

I then took a razor blade to her under it. Tetanus included.

After I removed the GPU clamp it was time to get to work with the other screws holding the PCB to the cooler.

After taking the back screws off it was time to take the shroud off in traditional solaris style all of this was done in pajamas. philips size t00

she has her top off.....scandalous

the back of her shroud

Aluminum fins cover a copper base plate at this point unplug the fan.

Moving on we need to remove the last 2 screws on the front I/O panel that hold on to a brace beam connected to the cooler sub assembly.

hello my darlings

Mr. Bracebar now with 6 cups of coffee

The cooler is stuck on and requires some wiggling to get off. I slowly pry from the fan toward the center the PCB bends easily so be very careful prying or you could damage the card. After I got about toward the center I switched to the other side of the cooler and started to pry the whole process took around 5min and it popped off. Oh my 290X stop not here my GF is in the house. She could come in at any second.....Ok but quick.

The TIM used is dry to the touch and has the consistency of bubble gum. While their is good contact between the die and the cooler too much TIM was used and I will call the quality of the TIM into question as well.

The bottom of the cooler has the typical mem VRM and GPU VRM TIM pads as well as the memory pads. My particular card uses Elpida chips.

The close up of the block shows the spread the TIM used is hard to get off and I reccomend 90% alcohol and I had to borrow some acetone from my GF (polish remover) to get it clean. After most of the removal the alcohol (70%) just smeared it around.

The block surface is scarred up and while I expected better I have seen worse. I dont have the $$ atm for proper sand paper or I would have lapped it. I imagine that will help ALOT when I do.

Why buy expensive telescopes when the surface of mars is on your family room table?

After cleaning the core I grabbed my new tube of MX-4 and used the pea method.

My fan ramp is set to custom in after burner and its baseline is 50% with the fan speed ramping 1% for every 1ºC the GPU climbs.

My tests were done driving 2 monitors at stock clock speeds for my GPU

Idle:
Before- 50ºC
After- 48ºC

GPU-Z Render test 100% Load
Before- 88ºC
After- 71ºC

Unigine Heaven Benchmark (Max Temp)

DX 11
Ultra
Extreme
x8 AA
Full Screen
1600x1200

Before- 95ºC
After 77ºC

OCCT Stress Test
Before- 95ºC
After- 88ºC

I'm very pleased with the results. Throttling only happened 5 times in very short ~2 second bursts while running OCCT but did not throttle for the other tests. Instead of buying an aftermarket cooler I may tear it apart again to lap the cooler to a mirror finish and re measure my results from that. I'm more interested with the results from doing that then a new cooler.

Well it was fun

 

[vega22]

keep an eye on your temps over the coming weeks as mx4 will dry out and lose some of its ability to move heat.

 

[newtekie1]

I'm impressed that simply replacing the paste could give such good results. If this is all that is needed to keep the card from throttling and drop temps 5-10°C, it makes me wonder why AMD cheaped out on the paste in the first place. Seems like a crippling move for such hot cards(though nVidia did the same thing with Fermi cards...).

 

[HammerON]

I have been curious on how temps may look on these reference cards after replacing the stock TIM. Thanks for sharing

 

[kn00tcn]

it will be worse if the shroud is removed, right? since the air wont tunnel through out the back?

about your ramp, are you saying before it ran 88% in gpuz, 95% in heaven? after 69% & 75%?

 

[Solaris17]

it will be worse if the shroud is removed, right? since the air wont tunnel through out the back?

about your ramp, are you saying before it ran 88% in gpuz, 95% in heaven? after 69% & 75%?

Yes they coincide. I didn't change my fan ramp at all between original and new TIM. so before temps indicate the fan speed percentage as well. they are a 1:1 ratio.

 

[kn00tcn]

that sounds unacceptable for most people i'm interested in what % fan you need to have heaven running at about 90 degrees target with the new paste

also, that was your 16,666th post

 

[Solaris17]

i figured someone would say something about fan speed when they read my ratio but all I can really say to them is that they are a pussy. It does sound ridiculous 90+% and is noticeable around 85% but if you have a headset their is no reason for it to bother you unless you live inside your case and if you use speakers then buy another card.


is what I would say to them.

Everyone seems to forget fermi and 290X bashing is easy given its price but all 2xx and 4xx cards did it. and trying to vs titan or the 78x cards is laughable because their coolers span the entire length of the card.

 

[DF is BUSY]

great teardown, the contact point of the copper cooler and the chip looks terrible, what is up with the concave bump? not to mention the scratches. i imagine the stock TIM is probably the same cheap pos that intel used for 3rd and 4th gen processors

 

[Solaris17]

great teardown, the contact point of the copper cooler and the chip looks terrible, what is up with the concave bump? not to mention the scratches. i imagine the stock TIM is probably the same cheap pos that intel used for 3rd and 4th gen processors

honestly it feels very similar.

 

[kn00tcn]

well i come from using a 4870x2 that i made a custom fan ramp for, 38% idle target 90 degrees, so i'm not saying let's make the fan capped at 50%, but going past 80% is a concern for both loudness & hitting 100% if it gets dusty (not being able to cool further)

90 degrees has been a common reference temp for years, even cpus get close to that nowadays (my i7-2670QM reaches that when rendering, even on max fan), so i'm not looking for arbitrary sub 80 degree fancy cooling from massive fans or custom coolers

the loudness concern is the fact that my case is within arm's reach to my right on the table, meaning, the fan is directly going to my ears even with headphones

on top of that, i replaced an antec900 with a corsair 500r, everything is LOUDER now! probably due to the fact that the case window opens up closer to the front, revealing the position of a reference gfx fan within line of sight (but also hard drive activity is louder)

 

[the54thvoid]

This all goes to show that AMD really need to try harder when they release cards. If all it takes is a little care when assembling the cooler and perhaps having better contact (with reference to the copper heatsink) they could have had much more favorable reviews regarding heat and noise at launch.

Well done Solaris on the cooler re-assembly. You tempted by water at all? Seems that is the best way forward for lots of todays high end cards.

 

[Solaris17]

This all goes to show that AMD really need to try harder when they release cards. If all it takes is a little care when assembling the cooler and perhaps having better contact (with reference to the copper heatsink) they could have had much more favorable reviews regarding heat and noise at launch.

Well done Solaris on the cooler re-assembly. You tempted by water at all? Seems that is the best way forward for lots of todays high end cards.

I did water with my DD loop but I did hardware reviews at the time and got tired of breaking everything down when I got a cooler. So I sold everything and walked away. I might do it again but if I do im probably going to go EK everything and do all the components. That way I just drop the cash and do it right the first time. Instead of adding to the loop over time.

 

[kn00tcn]

This all goes to show that AMD really need to try harder when they release cards. If all it takes is a little care when assembling the cooler and perhaps having better contact (with reference to the copper heatsink) they could have had much more favorable reviews regarding heat and noise at launch.

Well done Solaris on the cooler re-assembly. You tempted by water at all? Seems that is the best way forward for lots of todays high end cards.

seems like every single company's product can be improved by redoing the paste, intel/nvidia/amd/sony/microsoft/via/apple

i would think in a mass production situation, every little cost counts & every second it takes to apply the paste counts

i wouldnt mind having a DIY gfx cooler product like we have for cpu coolers, saves them assembly time

 

[rougal]

Good tear down... For me I really like the way the cooler looks, but that's one hell of a bad copper base... no wonder all the stock cards are way hotter than expected. Very informative indeed...

 

[Solaris17]

Good tear down... For me I really like the way the cooler looks, but that's one hell of a bad copper base... no wonder all the stock cards are way hotter than expected. Very informative indeed...

Thanks yeah I honestly dont mind the stock coolers I like the simplicity but I think its missing something not sure what though.

 

[OneCool]

Thanks yeah I honestly dont mind the stock coolers I like the simplicity but I think its missing something not sure what though.

Your not sure what?

I could find a few things...

 

[Solaris17]

lol oh you.

 

[manofthem]

Thanks for the thread and the pics, Solaris; I enjoyed looking through it. It still surprises me at how poor the TIM job always is on these AMD cards... Mine looked just like yours, it's a shame

 

[Solaris17]

Thanks for the thread and the pics, Solaris; I enjoyed looking through it. It still surprises me at how poor the TIM job always is on these AMD cards... Mine looked just like yours, it's a shame

It really is I was more disappointed by the copper job. I understand companies can save money on paste w/e every CPU I buy comes with a travel sized colgate. but to not mirror finish a flag ship? even if not mirror I shouldnt be able to feel burrs even if the surface looks scratched. I guess they just cant spare the time on the mill. Did you go water? or after market air mano?

My personal biggest challenge is going to be lapping that space. the copper block is actually epoxied to the black parts you saw where the memory pads were. In order to lap it correctly I will need to bake the block off. buy thermal epoxy then buy emory cloth and a small block of wood to re polish that surface. then I will need to thermal epoxy the block back on (might be an improvement its normal epoxy stock afaik) I might need clamps but ill hit that road when i get to it.

 

[manofthem]

It really is I was more disappointed by the copper job. I understand companies can save money on paste w/e every CPU I buy comes with a travel sized colgate. but to not mirror finish a flag ship? even if not mirror I shouldnt be able to feel burrs even if the surface looks scratched. I guess they just cant spare the time on the mill. Did you go water? or after market air mano?

I put an EK block on my card and it's been pretty good so far. Come to think of it, I don't remember how the copper looked on my stock cooler, and now I'm wondering... tomorrow I'll pull it out of the box and take a gander, compare to yours. But you're right, it's pretty bad when little things could have made a nice difference on these high end cards. (i also don't recall which ram chips my card has, mad I'm getting forgetful )

 

[Solaris17]

I put an EK block on my card and it's been pretty good so far. Come to think of it, I don't remember how the copper looked on my stock cooler, and now I'm wondering... tomorrow I'll pull it out of the box and take a gander, compare to yours. But you're right, it's pretty bad when little things could have made a nice difference on these high end cards.

Im actually thinking of getting an EK backplate for my card even with stock cooler. They say its NOT a stand alone unit. but looking at the schematics Im pretty sure if anything will cause an issue its the screws they give you but the head diameter looks the same so I might get one and use the original screws that came with the stock cooler and just use those to put the plate on.

but thats a $40 adventure I cant afford atm.

EDIT:: You can check your chips with this. Ill upload it

http://www.filedropper.com/memoryinfo

 

[Mussels]

its atrocious how such an expensive device can be crippled buy one cost cutting measure, especially one with known thermal issues.

 

[Solaris17]

You know the more I think about it the more I want to try does anyone know much about arctic alumina adhesive? or a better one?

http://www.frozencpu.com/products/3...ramic_Thermal_Epoxy_-_5_Gram_set_AATA-5G.html

I kinda want to test the absolute limits of the stock cooler with proper materials.

Mirror lap

thermal adhesive between the block and subframe

3m pads and VRM thermal strips

and an EK backplate (because im sure I could get it to work)

its kind of exciting.

ill probably switch to the liquid TIM at that point just for kicks.

 

[kn00tcn]

oh that reminds me... amd says their gpus are fine at 90 for years, but what about the vram? my 4870x2's vram went bad in june while playing limbo, but i have the feeling i stressed it physically when handling the card & holding it by the backplate (for sure i killed an x800 while handling it that same week, bsod when the driver loads)