Well now boys, I'm going to sand the heatsink. Specifically I am going to sand the chip area to a lower level than the HBM2 area, which I plan to keep 40μmhigher than the rest. Not only will leveling out the heatsink give a more even contact, it should also get rid of imprefections. While getting the HBM part of the heatsink higher, should offer better contact to it, as the default Vega 56 package I got is not modled and the HBM dies are lower.
I don't have pressure paper or precise enough tools to measure before after and during the process so I will have to eye ball it.
Just so you guys understand 1–10 μm – length of a typical bacterium 10 μm – Size of fungal hyphae 5 μm – length of a typical human spermatozoon's head[3] 3–8 μm – width of strand of spider web silk[4] about 10 μm – size of a fog, mist, or cloud water droplet
For more examples of things measuring 1 to 10 micrometres, see Orders of magnitude (length) § 1 micrometre.
Between 10 μm and 100 μm
about 10–12 μm – thickness of plastic wrap (cling wrap) 10 to 55 μm – width of wool fibre[5] 17 to 181 μm – diameter of human hair[6] 70 to 180 μm – thickness of paper
And I have to remove 40~ on one side of the heatsink.
Will post progress pictures of the process and so on, as well as a before and after tests in terms of thermals.
washers seem to fit nicely in terms of size, they should be around 0.5mm thick which should compensate for any heatsink contact I sand off and in fact increase the pressure some a bit.
Here is the old paste application and you can even see the pressure distribution of the heatsink and package. As you can see the chip itself had good contact in the center, but near the edges it seems to be weaker and as you expect questionable contact on the HBM. Note also the thermal pads and how they are not making the fullest contact they could be. I plan to adjust and move them as well.
Applying more pressure is fine and nice but don't overdo it and irreparably damage your hbms
Landon Wood
Started off with 2000grit, just so see how things go. Look at what pattern the sanding gives, how my hand movements work and figuring out what would be the best way to go about this sanding business. You can see it get shinier and then duller, because I went to 500, then 1000, then 2000, so the polish I worked up first goes away, but in thery 1200grit already provides all the "polish" you need, its mostly about the thing being even and giving a nice contact, not making a mirror polish and removing all imperfections, as with thermal paste usage, you get dminishing returns past 1200~ grit or so.
I know, in ideal conditions I would have pressure paper, precise tools to measure the height diffferences and so on. But it's hard to get them and expensive in other case. So I use just a ruler and eyeball it. Kinda daring, considering I will have to make a flat surface by hand and I will have to make one area of the heatsink higher by less than a wool fiber.
you are a moron. thermal paste is supposed to level the heatsink
Henry Russell
flat surfaces tend to go convex when sanded down
Brayden Russell
Thermal paste evens out small fractures and uneven surfaces, it's intended for scratch tier problems. However 40μm is not a scratch tier problem. That's enough of a difference to feel an edge, if you were to run a piece of paper from HBM to the chip die, it would get stuck, if you would apply an even object on the chip that would go over the HBM as well, plenty of light would shine through as there is that much space there.
you cant see or feel 40 microns. maybe with your fingernail if there was a hard edge, but otherwise no.
Levi Murphy
>you cant see or feel 40 microns. It's a fiber size. More importantly, I can definitely see and feel that when comparing the HBM and chip die height, based on the description here. Will I be able to adjust the height using my human hands and eyes, is another question though.
you're eyeballing sub-millimetre work... without any equipment other than a sanding block? inb4 finished surface isn't level
Gavin Hughes
You also could have bought a 1070 or something and had a gpu that just werks.
Jaxon White
Keep going, OP. So long as you have a near 3 digit IQ you can do this easy. I'd love to see the results. OC is the lifeblood of this shithole and you're doing God's work now.
Shut the fuck up, fag.
Austin Phillips
ahhh close enough
>you're eyeballing sub-millimetre work... without any equipment other than a sanding block? inb4 finished surface isn't level I might not have any equipment, but I got balls to go where no retard has gone before!
>sand off 10μm >replace it with 2mm of thermal paste hope you took before temps to compare to
Juan Rogers
>replace it with 2mm of thermal paste I will get squished out, I rather have a tad more than less in GPU in general, but even more so on vega with it's uneven die, hbm. And now I did sanding, so there could be unevenness caused by me, trust me, bit more rather than bit less, will be better here.
That and I plan to warm up the card without fans on. pull it out and tighten the screws a bit more while the paste and heatsink is nice and hot/warm.
Jaxson Harris
>sanding down a cnc machined surface by hand to make it flatter
I should add to this, that the only one thing I changed that wasn't shown on the images so far, is that I removed the rubber between the noctua fans and the heatsink, previously I was testing more "airflow" oriented setup. This time I though the fans being closer, would yield a higher pressure and thus possibly better results.
There you have it boys. The temps went down >-6 on GPU > -5 on HBM > -4 on hotspot But that could be because of the slightly different fan placement. The fan speed and literally everything else remained the same between the tests.
I was hoping the HBM and GPU thermals would equalize more to be honest, yet there is still a difference in fact it increased from 3 deg difference to 4 deg difference. Did I fail in my quest to get better cooling for HBM? Did I not sand enough? There is many possibly reasons, and feel free to tell me what you think the reason is. At least the over all temps are down, so I guess it's not a complete failure. Still I was aiming for equal HBM and GPU temps.
All of this for -5C? How little do you value your time?
Ryder Gray
>All of this for -5C? How little do you value your time? Hey, if you can't play video games on this card, then might as well thinker with it and derive my "fun" that way.
But I'm more interest interested why the HBM2 temp isn't equal to the core temp. Both temps went down, but the uneven temp is still there for some reason, in fact one might say it increased by 1C. But is that because I fucked up, or because the core cooling just improved much more than HBM.
Jason James
This is some next level shit. Seems like too much effort for the marginal drop in temps.
Matthew Morgan
nvidiots confirmed for retards that can't tinker with their card for max performance, how gay tracing and blurry DLSS treating you
Ethan Roberts
You're a seeker of efficiency, nice. I like that.
Jack Thompson
OP actually risked his hardware. He had money on the line. He took that risk for better temps, better performance before thermal throttling. This is a real enthusiast. OP today was not a faggot.
Jordan Moore
Elaborate
There is that small part of you that thinks >but what if I fail, what if I make things worse, what if I void my warranty But then I realize that this cuck mentality disgusts me, to be afraid of touching the thing I bought. Think about it a lot of GPUs can get better thermals with just thermal paste replacement, yet people use the stock shit for 2-3years until warranty expires and in all that time its running hotter/louder/less performance. I reject this shit! It's like that girl you like but you never talk to. I rather go and talk to her and get rejected, than live the rest of my life feeling, knowing, that I was too scared, too much of a cuck to go and talk to her.
That said, I did think think this thing through from the very start and considered worst possible outcomes and the risk I took was calculate and deemed safe enough. I imagine there is people who would have fucked something up if they tried to do this though.
Look into getting higher quality thermal tape from Amazon. That stock thermal tape doesn't look like it's providing full coverage and will also interfere with how well the heatsink applies pressure on the gpu die/hbm area.
mod/smart Fujipoly Ultra Extreme XR-m Thermal Pad - 60 x 50 x 1.0 - Thermal Conductivity 17.0 W/mK
Austin Cox
While you could have just molded it yourself and used liquid metal... Gives actually better results than factory molded, since you have direct die contact.
Xavier Wood
My V56 used to beat a V64 with it's OC/UV while NEVER going over 55C after I put some adhesive paste between the gaps of the dies (wasn't factory molded) and used liquid metal for the HBM and core. It already had a beefy cooler since it was a Nitro+.
Carson Edwards
Just polish to mirror and use liquid metal OP, sanding for height is a negligible benefit when you have insanely high thermal contact.
Samuel Wright
>and used liquid metal With the exposed interposer, I have high HIGH doubts about that. Additionally copper does react with liquid metal. If the HBM and interpose were covered with epox and nice and leveled, if I had a nickle platted heatsink, then I would have used liquid metal for sure.
David Harris
should be mirror finished kid. comb your hair in that shit. cloth ;)
Aiden Wright
Ah shit, I thought they capped off the interposer after the Fury.