AMD Reveals the Monsterous ‘Exascale Heterogeneous Processor’

[Johan45]

Talk about a monster !

There had been rumors about AMD working on a huge APU with Zen cores and Greenland HBM graphics, something that AMD had hinted upon in its official roadmap. However, it has (finally) officially revealed details about the upcoming APU in a paper submitted to IEEE (Institute of Electrical and Electronics Engineer). The APU, dubbed an “Exascale Heterogeneous Processor” or EHP for short is the mother of all APUs with 32 Zen Cores, an absolutely huge Greenland graphics die and upto 32 GB of HBM2 memory – all on a 2.5D interposer.

Read more: http://wccftech.com/amd-exascale-heterogeneous-processor-ehp-apu-32-zen-cores-hbm2/#ixzz3i9clLtAS

 

[Scout]

I hope it turns into real product not just vaporware, wonder what process it would be on if it ever got made.

 

[Johan45]

Last I heard they were jumping right to 14nM

 

[Mandrake4565]

Seems like by the time AMD comes out with a new chip worth buying I'll be considering if I'll be replacing a Skylake chip with whatever Intel is coming out next.

 

[Silver_Pharaoh]

Seems like by the time AMD comes out with a new chip worth buying I'll be considering if I'll be replacing a Skylake chip with whatever Intel is coming out next.

*slaps across face*


Sounds like a beast, let's see what really happens here though

 

[ShrimpBrime]

But can it overclock?

 

[Tech Tweaker]

This sounds vaguely like the "Sky Bridge" architecture I heard about earlier this year (or was it late last year? I don't remember). Except it sounds much bigger and more complex than what I remember hearing about initially.

But can it overclock?

It darn well better!

I'm sure someone will make it overclock, even if the designers never intended it to. Probably Dolk, if he's not already busy.

 

[Dolk]

What do you guys want to know about this type of architecture and what Intel is doing that counters this?

HBM type memory and the open memory architecture is going to be pushed to every type of chip pretty soon. Stacked memory is here and now; TSMC, GF, Intel/Micron, and IBM are all heading that direction. What does this mean server side? Faster memory bandwidth and higher density of secondary memory (now RAM). Will this be brought to consumer side? Maybe.

All in all, I'm excited for this turn in architecture design.

 

[ShrimpBrime]

Ya.... Again Ill ask....

Can it freeking overclock. If not.... Im not interested hahahaa.

 

[trents]

As fab size gets smaller and smaller, overlcocking becomes harder and less productive. 32nm chips seems to have been the golden age of overclocking. Best combination of low voltage and heat dissipation in my opinion, for both Intel and AMD.

 

[EarthDog]

As fab size gets smaller and smaller, overlcocking becomes harder and less productive.

How so? Particularly with the 'less productive'... what do you mean by that?

 

[trents]

How so? Particularly with the 'less productive'... what do you mean by that?

Check the post again. I added to it. By "less productive" I mean the amount of overclock possible over stock. The power brick densities have become so high that even small amounts of voltage increases causes temps to zoom and the surface area of contact between the die and the IHS gets less and less, thus not as efficient at removing the heat.

 

[Woomack]

I can't really agree. You can OC new chips by 50%+ by raising voltage about 10-15%. In the past you could make exactly the same.
The main difference is that new chips have often IGP, IMC, PCIE lanes, much more cache and some other stuff inside the CPU so it simply has to take more power but TDP in most cases isn't much higher.

Socket 462/756/939/478/775/1156 could OC ~50% stable in most cases at voltage bump ~10-20%. Since 1366 it was harder but still ~30% was possible at ~10% voltage bump. Right now you can OC IB/HW/HW-E/SL about ~30% at ~10% higher voltage, in some cases on stock voltage.

CPUs with the highest OC potential are ... locked. Mobile chips could OC 100% at stock voltage if only multi was unlocked.

I'm mentioning mainly Intel here as AMD stuck 5-7 years ago and can't seem to move from that point. FX8000/9000 series are scalling like 8 core P4. AMD has nothing new to offer and it's hard to compare their products with older generations as they're actually old.


Back to article ... this new chip is planned for ~2025. We will be lucky if AMD will be still on the market till then I hope they won't let us wait 10 years for next CPU. FX will be slightly outdated till then.