AMD Sampling DIFFERENT Vega 64 GPUs?

FrgMstr

FrgMstr

Just Plain Mean
Staff member
Joined
May 18, 1997
Messages
55,602
OMG! There are packaging differences in AMD Vega 64 GPUs! What the hell is AMD trying to pull? The short answer is, "nothing." Some have epoxy around the silicon, others don't. Here is the official answer from AMD. - "We have multiple manufacturing partners for Vega, one of them uses an epoxy fill to level the SOC surface, otherwise they’re functionally identical."

Check out the pic.

Crisis averted. No AMD conspiracy today. Trombone sound...
 
Last edited:
The bigger issue is likely going to be the height difference from the core and the HBM2:

eA01q34.jpg
 
Last edited:
I will post this here, as I did in the review thread when it comes to temps on these GPUs.

"I am not sure you can contribute that to the substrate packaging. To actually test that you would need to do some seriously controlled heat testing with multiple GPUs from each group then compare, THEN you would need to pull the heatspreaders and test all that again with bare GPUs to figure out if it was packaging or IHS tolerances."
 
The assemblies WITHOUT epoxy should be cheaper, underfill costs quite a lot in application and inspection. Not to mention rework, you know how hard it is to remove that with an epoxy underfill? We recently purchased a machine for BGA packages, all I can tell you is it's in the 6 digits. This is primarily for high vibration use cases, i.e., planes trains n automobiles. Haven't seen this on a GPU before...

Unless I'm completely wrong and it's just around the perimeter. The machine is still expensive, but not much in the way of inspection. Sort of like a conformal coating application.
 
Mugato said:
The assemblies WITHOUT epoxy should be cheaper, underfill costs quite a lot in application and inspection. Not to mention rework, you know how hard it is to remove that with an epoxy underfill? We recently purchased a machine for BGA packages, all I can tell you is it's in the 6 digits. This is primarily for high vibration use cases, i.e., planes trains n automobiles. Haven't seen this on a GPU before...

Unless I'm completely wrong and it's just around the perimeter. The machine is still expensive, but not much in the way of inspection. Sort of like a conformal coating application.
Click to expand...

Any reason to think its NOT just around the edges? That's what I'd assume
 
ecmaster76 said:
Any reason to think its NOT just around the edges? That's what I'd assume
Click to expand...

Most likely, but why either way is the question. I've seen some odd things from other fabs I guess, just seems like a waste of money.
 
Ocellaris said:
The bigger issue is likely going to be the height difference from the core and the HBM2:

View attachment 33545
Click to expand...
Which is why this line in the article annoys me:

The difference is so incredibly small that a bit of thermal compound likely would make this observation irrelevant.

Oh, how wrong they are. Remember that the gap between the IHS and die on Haswell was only about the width of a piece of paper, but it caused all kinds of thermal issues which led to delidding becoming a common practice for overclockers. I'd be very concerned about the HBM2 stacks not getting sufficient cooling in this situation.
 
Armenius said:
Which is why this line in the article annoys me:

The difference is so incredibly small that a bit of thermal compound likely would make this observation irrelevant.

Oh, how wrong they are. Remember that the gap between the IHS and die on Haswell was only about the width of a piece of paper, but it caused all kinds of thermal issues which led to delidding becoming a common practice for overclockers. I'd be very concerned about the HBM2 stacks not getting sufficient cooling in this situation.
Click to expand...
You would be making the assumption that all GPUs are built with the same IHS. There is no reason to NOT believe that each manufacturer's GPU is packaged with its "own" IHS. I would highly suggest that is the case if there are Z height differences.
 
Armenius said:
The Haswell thing was just an example, but I think there is no doubt that cooling would be suboptimal if the heatsink isn't making full contact with the dies or at least with a shim to make up the difference.
Click to expand...
Uh...yeah.
 
Any makers of watercool blocks for GPUs will have some kind of issue with this circumstance. They either offer 2 or 3 different blocks for the different chips used ( means you would have to take it apart 1st to order the correct block ) or they will use an offset that is big enough to allow a thermal pad for for anyone of the chips, which would be suboptimal.

This is another "not so nice" with Vega.

I am tied to NV anyway through my Gsync screen, I just wished AMD would do better in the GPU sector. This all doesnt look like it from my perspective
 
BitMaster said:
Any makers of watercool blocks for GPUs will have some kind of issue with this circumstance. They either offer 2 or 3 different blocks for the different chips used ( means you would have to take it apart 1st to order the correct block ) or they will use an offset that is big enough to allow a thermal pad for for anyone of the chips, which would be suboptimal.

This is another "not so nice" with Vega.

I am tied to NV anyway through my Gsync screen, I just wished AMD would do better in the GPU sector. This all doesnt look like it from my perspective
Click to expand...

Or the first post in this thread is correct and people and just reaching.
 
I don't think there is a height difference. The chips look like they are offset by a little bit so the angle of that picture just makes it look like there is a height difference.

When looking at one of the pictures where it has the black fill, to me, it looks like the fill around the big chip is a little thicker.


eA01q34.jpg
 
Last edited:
Until someone measures with a set of calipers we'll never know the actual difference, if it even exists.
 
Dudes, the difference is called "Underfill", and if you want a mojo cooler on there with a ton of contact pressure, you want this Epoxy to be there.

With underfill is more robust; the HS pressure is spread thru the epoxy, not only the solderballs under the big BGA chip, which then don't break due to pressure and heat.

:)
 
HBM is like 20w of power, probably not as critical in cooling the gap distance. Does not HBM have temp indication built in?
 
You must log in or register to reply here.
Back
Top