Discussion in 'Frontpage news' started by Hilbert Hagedoorn, May 18, 2015.
As long as there is no air gap at all, heat will spread with no issues.
Thanks for that info, as I suspected it could be then, I've seen some other links since you posted that confirm this too.
Well, different materials have different thermal conductivity, which is why heatsinks and thermal pastes are made of the materials they're made of - high thermal conductivity where possible. (Well, thermal pastes aren't highly thermal conductive, but at least the better ones have higher thermal conductivity figures). But I agree, an air gap is horrendous for heat transfer, air is an insulator effectively, but it is important to use the right material for thermal transfer - see the heat issues caused by moving from solder for the IHS with Intel.
That heat would just sit there in that air pocket, and another issue that could happen while it's very low chance of happening but moisture could get in that air pocket.
So my concerns about the skyscraper effect are unwarranted as H personally asked AMD about my questioning of HSF removal and the worry about damaging gfx cards.
Still they may only be answering the question i posed about 1st Gen HBM because it can only achieve 4 stacks but Gen 2 will be 8, 16, 32 stacks(?) and then they might have an issue with the skyscraper effect. I'm not sure if it means 4x4 Gen 1 =4 gb but future Gens maybe 8x8 or 4x8 or 8x4?
Ide imagine the new gpu's are going to look like the old ps3's gpu when delided. http://community.us.playstation.com...U-look-different-to-repair-guide/td-p/6883680
Talking four years now where a crossfire 7950 or 7970 setup is much more powerful than most (any?) single gpu. This is getting real boring real quick. With dx12 they get pooled memory. So basically, who f'ing cares it's just another little boost.
Thanks Hilbert, I think understand how HBM works now. I know I didn't before reading this article.
What are the demerits of HBM compared to DDR5?
Wrote it before, 4GB of vram is enough for me, just bring in The GPU AMD.
Once APU gets HBM as L3 cache or any other cache, we are on next level.
I can see mobile Zen APU equipped with HBM2 only, to save power and so much space.
We need unified RAM tech...(similar to consoles), imagine 16+ of these..
So, I am getting the impression AMD is holding the reigns on HBM and will be able to charge licence fees to NVidia / Intel etc for it's use for years to come?
If so this could help give AMD a serious financial boost for years to come helping them become more competative in the R&D field, even though they seem to be doing amazingly well on such a low current budget! ^^
I do not know what who exactly is holder of IP, since it is AMD+Hynix who made this.
But would be nice if AMD got few $ per HBM sold by Hynix.