Originally Posted by maarten12100
Well duh you would need a ~70% increase in perf to match the multicore scaling of a 4670k that isn't going to be solved with just one architecture efficiency improvement.
I would say we'll see it come in a little below in single threaded and 30% bellow in multithreaded.
Kaveri will be a great chip I mean we have already seen it run BF4 720P on ultra with a solid 30/35fps just with some AA settings turned down.
For gamers that are in it for the game rather than the graphics this is great really great and if hybridCF gets the needed improvements also due to Mantle than we can really see why this is what will be in the cheaper steamboxes.
It is a great time to be a gamer (Fallout teaser is also comming
Piledriver isn't that far behind Ivy - 30% would be more of a accurate statement. It's MULTI-THREAD that doesn't scale well.
With SR though? It does. It scales perfectly
Originally Posted by Arni90
AMD APUs have been around for two years now, and we've only seen a handful of programs capable of taking advantage of the integrated GPU for heterogenous computing, it's unlikely that 2014 will up the pace for HSA adoption in any significant manner.
Sandy Bridge performance is very close to Haswell performance by the way, if AMD were to catch up to Sandy Bridge they would also catch up to Haswell.
This is definitely an evolution of the Richland APU, there's no point trying to fool yourself into thinking otherwise.
The fact that AMD is not making an FX-processor using the steamroller architecture points to them not being able to fix the inherent problems with the Bulldozer-architecture, which is kind of sad seeing as the Steamroller architecture should fix a lot of the problems of Bulldozer/Piledriver.
Thing is you haven't been seeing the preparation of HSA apps as Kaveri is the FIRST arch to support HSA
They didn't make a FX processor using SR is simple. FX is long-dead and should have been after 2006
Originally Posted by Blameless
Agreed, though not for precisely the same reasons.
Why wasn't Quicksync or OCL (Intel HD can use OCL) used on the Intel APUs in this comparison?
Even an i5 2500 can beat a 3960k in transcoding software that takes advantage.
Up to a certain point perhaps, but some nominally GPGPU workloads will be CPU limited on Kaveri long before they are on a faster processor.
HSA doesn't magically remove the use for a powerful executive processor that is good at serial tasks.
Any broad comment on CPU vs. GPU budget ratios is an over-generalization, at best.
Intel has shown a great deal of interest in improving their IGPs/APUs, but you are correct about HSA. Intel is not a apart of HSA and hasn't made any serious push for conformance
Because Quicksync = Speed with expense of video quality. BIG expense. My friend had a 3770k and he tried encoding with Quicksync and then i reviewed the footage ... I nearly puked. If this was a 3" screen that would be fine but most of us watch on 4.7" (and up) 720p-1080p displays on the go
OR 23.5" 1080p displays. The result is horrible
And despite being said to be the man of "expansive knowledge" you don't seem to understand HSA. HSA is to make use of the ENTIRE APU to boost applications. If it's CPU "limited" it means it can only use just as much (Lower usage ratio for the APU) or HSA is basically not working
Intel has not REALLY shown absolute deal of interest. Their gains have been at expense of GPU power consumption.
Lower the CPU power consumption and increase the iGPU (for mobile) is what they've been doing.
With AMD? They've been keeping power consumption for both portions the same, only with GCN they surely improved the GPU by a huge amount - Yes. 512 GCN1.1 28nm cores pulls as much power as 384 VLIW4 32nm cores but because per clockspeed Kaveri is going to draw a bit more (But it doesn't need to go so high, considering the MASSIVE IPC improvements they've done with SR)