Cazalan :
GCN is better but it's going to take more than that to rocket the APU speeds up. AMD already has bandwidth limitations with Piledriver seeing improvements up to DDR3 2400.
GCN on the desktop has the benefit of 150+GB/s of local GDDR5 memory.
AMD is running into scaling issues without adopting faster memory or on-die RAM/cache. DDR4 is still looking at late 2014 for mainstream so that's 3 chips away (SR, EX, ??? ).
imo, all they have to do is find a way to feed the trinity igpu some 6-7+ gb/s of memory bandwidth @ ddr3 1600/1866 ram and scale from there. like intel gets 21 gb/s with ddr3 1600. if you take a look at a10 5800k's gpu-z screenshot, you'll notice that that gpu memory bw is rated at 25~ gb/s. this is a screeny from xbitlabs' trinity review pt2 of overclocked trinity:
http://www.xbitlabs.com/picture/?src=/images/cpu/amd-a10-5800k/trinity-4500-1086.png
if the memory bw reading is correct, the igpu can be fed up to a massive 34 gb/s of data. that's higher than the bw reference radeon hd 6670 ddr3 version(@28gb/s) uses. i fear that with sr, the stronger igpu performance might be diminished due to bottleneck.
here's an excerpt from at's latest memory roundup:
We are told that on the Intel side of things, Haswell is DDR3, as will be Broadwell, the Haswell replacement. Reports expect DDR4 to be less than 10% of the market in late 2014 (early adoption in the high end space), but 50%+ across 2015.
http://www.anandtech.com/show/6372/memory-performance-16gb-ddr31333-to-ddr32400-on-ivy-bridge-igp-with-gskill
afaik, haswell xeons will get ddr4 support first and trickle down from there.
jdwii :
.....
But overclocking performance all depends on how much Amd improved piledriver's power consumption and really i see no improvements beyond resonant clock mesh and possibly a newer stepping at Globalfoundries(or take in the fact that they had 32nm for 1 year and there's improvements i'm sure) and people are "claiming" that at higher clocks the advantage of Clock mesh is smaller but its claimed to be 10% reduction in power consumption OR 10% higher clocks which scares me a bit since the 8350 is clocked 11% higher than a 8150. I know a Phenom II x6 can be clocked at 3.9-4.0Ghz on average on air and 4.2Ghz with water or some advanced heat sink and by looking at the numbers Piledriver needs to be clocked a good 400-500mhz higher to beat a Phenom on average so Piledriver would have to be clocked between 4.4-4.5Ghz on air and 4.7Ghz on water.
......
As for Power consumption i don't see anything beyond a minimal improvement when compared to Bulldozer probably 10% or less but who knows i might be very surprised on the 23rd just like others here.
i think that current impressive showing of trinity's power consumption figures is a two tier effort by amd and their motherboard partners. chances are the motherboards contain aggressive power management logic as trinity does (which is a very good thing imo. but for oc'ers, not so much), combined with bios support, trinity shows idle power use lower than ivy bridge (when no discreet card is being used). all that goes out the window though, when trinity is overclocked. i said a while ago that oc efficiency might not change with pd. i haven't yet seen otherwise.
gamerk316 :
But as techreport showed not too long ago, even top AMD CPU's (BD, and even PII to an extent) runs into latency problems (microstuttering) in SLI/CF configs. Hence why I always felt the hybrid CF approach was a silly one to take, because you run into too many latency problems.
i agree on the latency problem.
hybrid cfx (amd dual gfx) enhances the value attraction the apus have. you have an entry level, dual gfx compatible discreet card, you can extend it's life by putting it in a desktop apu pc. otoh, you have an apu pc, you're looking for ways to get more gfx power but you can't/won't buy a higher end gfx card - in that case, a dual gfx compatible gfx card can get you a modest amount of gfx performance increase. However. like crossfirex, dual gfx depends a Lot on driver support. i've read two reviews with where trinity's dual gfx was tested. the performance was highly inconsistent. where improvement did show, it was big. which means that dual gfx has promise at the entry level. amd would be arrogant to think that they would provide trickled-down driver support for trinity and dual gfx because those are entry level. arrogance has cost them dearly already. and no, having smaller financial resources is not an excuse. it's about efficiently using those resources on products that will earn them money.
p.s. i used 'trickle down' because of blue mountain state...
![Pt1cable :pt1cable: :pt1cable:](/data/assets/smilies/pt1cable.gif)