Skylake was released prematurely. DDR4 offers little more than higher latency and price for CPUs with discreet graphics. There is no EDRAM, which is why Broadwell is faster clock for clock for most games.We have enter the age of no performance improvement unless we get a new node.
I wonder how much more performance can be squeezed out with new uarch. Skylake as far as i can tell is a flop.
AMD has a lot more time to evaluate Intel's design, and may be finally to produce a decent competitive chips at an affordable price. At least on the server side of things.
Sure, but someone here, possibly Shintai suggested cycle time was the problem and he seems to suggest that more capacity was brought on stream.The problem isn't cycle time, which is what you are referring to, but actual yields. They have capacity in place to build X chips at Y yields. If yields are significantly smaller than Y, then you wind up with a shortage.
They already did it seems.
So D1X, D1D, D1C, Fab24 and Fab32 is making 14nm chips now. Fab42 was originally to open and make 14nm chips too.
Sure, but someone here, possibly Shintai suggested cycle time was the problem and he seems to suggest that more capacity was brought on stream.
Cycle time does increase with each successive node as there are more manufacturing "steps" required to actually build a chip. However, this must have been factored into the original projections. The only things that would drive an unanticipated shortage are:
1. Way more demand than anticipated (Intel usually has "whitespace" -- additional capacity in place beyond its anticipated needs in order to respond to potential upside; TSMC does the same thing which is why its "structural gross margin" target is calculated based on 85% capacity utilization, not 100%)
2. Worse yields
Intel management tried to spin this as "super duper demand for Skylake" but the reality is that when your PC chip biz is down double digit y-o-y and you've already guided down your full year sales projections that's not "amazing demand" that's "crappy yields."
WRT additional capacity, Intel didn't come up with any capacity that wasn't originally planned; what they did AFAIK is they began ramping its 2nd major 14nm factory (in Lexlip, Ireland) a little earlier than they had originally anticipated.
Core Duo to Core 2 Duo was mainly single cycle SSE. And even then you dont get 20% average. The actual real life average was more around 5-10% back then.
I meant to say Nehalem to Sandy Bridge. BTW, Westmere doesn't count because aside from cryptography instructions there's zero gain. Also, "Core 2 to Nehalem" for example would mean "Penryn to Nehalem", not "Conroe to Nehalem". That's logical, because that's how everyone would compare it.Core 2 Duo to SB is Conroe->Penryn->Nehalem->Westmere->Sandy bridge.
FIVR should first return with Icelake. So that would be odd and interesting if true.
However do note that Skylake-E/EP and Cannonlake-E/EP will have FIVR.
http://www.notebookcheck.net/HP-Spectre-x360-13-i5-6200U-Convertible-Review.153933.0.html
http://www.notebookcheck.net/HP-Spectre-13-x360-Convertible-Review.144228.0.html
Skylake is just underwhelming, mobile and desktop. With desktop you don't get the peformance, with mobile you don't get the mobility. Despite their claims of SoC power reduction, it doesn't do anything as you can see from the review above. Broadwell achieved better battery life. Specs can't be more similar between the two.
I bet its because taking out FiVR made it harder to achieve the better battery life in practice. It makes sense because they've been touting the advantages of FiVR for mobile so long. FiVR even provides the advantages of thinner laptop and less components.
FIVR should first return with Icelake. So that would be odd and interesting if true.
It makes sense since because of the delays. Kabylake has some features of Cannonlake; Cannonlake probally has some features added to it from Icelake, and so on. Presumably Cannonlake and Icelake are socket compatible with each other while Tigerlake will be compatible with whatever's after it.