There were some leaks about a APQ8084. Pretty much the same as APQ8074 except with Adreno 420 instead of 330 (and also featuring h.265 decode). No idea what the name would be. There's indeed not much roadmaps to be seen from them, even the "leaked" stuff only covers like half a year in advance . But surely they'd be working on something new for 20nm for quite a while already.
So presumably, APQ8084 is still on 28nm, right? I googled it a bit, but there doesn't seem to be any information about Adreno 420, beyond h.265 support.
You may be right that doubling 330 performance on 28nm could be too much to ask for. So probably 420 will be slightly faster GPU than 330(maybe higher fillrate) but with full directx 11 support and higher compute performance. Who knows, maybe they'll be able to reduce power consumption due to advancements made to the micro-architectureYeah APQ8084 would still be on 28nm (probably the last high-end chip before 20nm from Qualcomm would be my guess, there's going to be a couple more budget oriented chips coming out on 28nm). As for graphics I don't think it's just going to be doubled performance, the number implies the chip is significantly different (Adreno 3xx is quite different to 2xx too), and doubled performance might be a stretch imho (330 is quite a monster and it's still on 28nm after all).
That's same story as with desktop/notebooks really, gflops increasing much faster than bandwidth. So you need to make sure bandwidth efficiency increases (aside from the fact that more modern apps will use shaders requiring high flops/bandwidth ratio).Aren't these GPUs starting to suffer from bandwidth starvation?
Is there a point in continuing to increase the GFLOPS like hell without substantial upgrades in the memory bandwidth?
Aren't these GPUs starting to suffer from bandwidth starvation?
Is there a point in continuing to increase the GFLOPS like hell without substantial upgrades in the memory bandwidth?
Aren't these GPUs starting to suffer from bandwidth starvation?
Is there a point in continuing to increase the GFLOPS like hell without substantial upgrades in the memory bandwidth?
Ah yes. I forgot that there won't be large density improvements with 16nm FinFET. I believe its the backend that's the same as 20nm though.TSMC's 16nm process uses the same frontend as the 20nm one and will barely offer density improvements. TSMC says something like 10%. The big improvement is the move to FinFETs.
Ah yes. I forgot that there won't be large density improvements with 16nm FinFET. I believe its the backend that's the same as 20nm though.
The wikipedia is semi-right on this one.
Both the HTC One and the Galaxy S4 have a SoC labeled APQxxxxx, which means it's a chip without baseband processor. AFAIK, both are paired with the MDM9215 (or whatever is its CDMA counterpart).
I don't know of any S600 implementation with integrated baseband, at the moment.
Isn't it a bit weird how it says 4G LTE and no 3G UMTS support? Is it omitted?