Trinity vs Ivy Bridge

Yeah, london is the family consisting of Thames, Chelsea, Heathrow, Wimbledon.. and supposedly the gpus on khrisna and trinity.
28nmmdgpumobile_1a_Dh_fx57.jpg.jpeg


Erm.. I can't see "London" in that graph, nor I remember seeing it in any other graphs related to SI recently.
 
....or everything, including Trinity and Krishna/Wichita are using already GCN.

Not quite probable that one. Also, GCN strikes me as something that will forfeit their density advantage and as such I'd expect it to trickle down far more slowly into the Lolcat/Atom world, but I may be wrong.
 
Weeeeell...
Wimbledon, Heathrow, Chelsea, Thames - just enter those terms in the online map of your choice. :)

Oh okay.

Even then, looking at the lowest performing Thames, it's supposed to carry 2x Seymour's performance. That's 2x the performance of a 160shader, 8TMUs, 4ROPs GPU.

Unless there's another, lower-performance London GPU, that should be the GPU going into the ~9W Krishna.

We may be looking at something like >240 VLIW4 shaders for the successors of C-50 and E-350. That's mighty powerful for a ~11" low cost subnotebook.
 
Not quite probable that one. Also, GCN strikes me as something that will forfeit their density advantage and as such I'd expect it to trickle down far more slowly into the Lolcat/Atom world, but I may be wrong.

GCN will of course be less dense (flops/mm wise), but as long as perf/mm is more (which should be the case), then I don't see why GCN's introduction has to be delayed beyond the usual integration/validation issues.
 
I'm not enamored with the arbitrarily rotated 3d graphs. It only serves to obfuscate differences. If it weren't for the numerical tables, half the time we'd probably guess some of the larger values are smaller based on how needlessly skewed the perspective is.
Were these graphed like this as a joke?
 
It seems IvyBridge's new iGPU will indeed have lower clocks.

All things considered, Intel's new iGPU can only be considered "functional" and not "performant", imo.
 
Because it scores similarly to the HD3000 in synthetic benchmarks, while supposedly having more EUs.


This is an early ES sample with low clock frequencies both CPU and GPU. I don't think we can make a statement for Retail CPUs based on this early ES. 39% better in Vantage is pretty good considering 2600k GPU clocked 50% higher.
 
I must admit I thought AMD was getting ripped off by switching to 'good wafer' pricing since it seemed the process should be mature by now, and it looked like nothing more than management panicking at low early yields which were unlikely to be representative of production yields (see: TSMC 40nm, which had an awful start but excellent yields ever since they fixed the via issue, anyone switching to 'good wafer' there too late would have wasted a lot of money).

So I'm both pleasantly and unpleasantly surprised by the fact 32nm yields are still subpar - AMD's management made a very good call, but GF underdelivering is still very bad news for both GF and AMD. It will be interesting to see how TSMC handles the 28nm and High-K transitions - if they do much better, that's going to hurt GF's prospects with 3rd parties a fair bit.
 
Back
Top