Do we really need another thread?
Yea, we already have 1500 posts in a thread about the mythical zen apu with 16 cores, Greenland APU, and HBM. Maybe we should cool the speculation until like you know 2017 when we might actually see a zen APU.
So we make one thread for general speculation, one thread for small/no iGPU speculation, one for big mobile iGPU speculation, one for desktop iGPU, one for CPU performance, one for mobile Zen power draw, one for a theoretical SKU with 6 cores and even larger iGPU, one for rumors about the HBM Zen APU, one for Zen SMT implementation, one poll for Zen including FIVR and subsequent discussion about power delivery on the new APU, and the list can probably go on.This is about the large iGPU, not the small one.
Totally different use case scenarios.
So we make one thread for general speculation, one thread for small/no iGPU speculation, one for big mobile iGPU speculation, one for desktop iGPU, one for CPU performance, one for mobile Zen power draw, one for a theoretical SKU with 6 cores and even larger iGPU, one for rumors about the HBM Zen APU, one for Zen SMT implementation, one poll for Zen including FIVR and subsequent discussion about power delivery on the new APU, and the list can probably go on.
This is about the large iGPU, not the small one.
Totally different use case scenarios.
We don't even know if there will be "small" or "large" iGPUs! Seriously, why do we need another thread?
That server APU with HBM is a totally different animal from a laptop APU.
I think there will be small and large iGPU Zen APUs because AMD needs to drive up laptop volume.
One recurring complaint I see about AMD APU laptops is that they have a higher than expected price (poor construction is another complaint). Low volume of large iGPU mobile APUs (eg, Kaveri) hurts the ability of OEMs to make large runs of a particular laptop. This, in turn, increases price.
However, one thing I already see AMD doing with Carrizo (Excavator CPU) and Carrizo-L (Puma CPU) is have the two APUs share the same BGA socket (FP4). Having a common socket should help reduce design costs to some degree as the large iGPU APU (Carrizo) can benefit from some of the small iGPU APU's (Carrizo-L) higher volume. Unfortunately though the socket is the same, the BIOS would be different.
Therefore, in order to share both socket and BIOS, I believe AMD will have both a large iGPU Zen APU (the topic of this thread) and a small iGPU Zen APU (which was the topic of this thread).
Mobile Tonga is 2048 SP @ 125W. Assuming they optimized a bit, removed the GDDR5 tax, and clocked it down to 700 MHz, it could probably be made to run at 60W, leaving 40W for the CPU cores. That would be a 100W package, which is not at all unprecedented for a gaming notebook. With 8GB of shared HBM it would surely be a total gaming beast. But since the form factor would be so small, they would surely want to target the 35W segment as well. For that, they wouldnt want to go past 1024 cores.
What if AMD managed to build a socket with the capacity to power HBM equipped APUs?
But since the form factor would be so small, they would surely want to target the 35W segment as well. For that, they wouldnt want to go past 1024 cores.
We could use, let us say 2 x 1/2 GB stacks as VRAM and L4 cache, with the rest of the memory being relatively cheap dual channel DDR3. BIOS could be configurable in how the HBM is treated, either as purely video memory or a configurable unified pool with higher capacity stacks.
We don't even know if there will be "small" or "large" iGPUs! Seriously, why do we need another thread?
Could HBM/HBM2 work as system memory in this case?
