(no title)
Nelson69 | 5 years ago
AMD doesn't look nearly as one dimensional as they did then, power consumption wasn't as interesting and Intel came out with a low power play; this time AMD seems to have offerings in every category that are compelling. It's really hard to bet against Intel with their long history though. I wouldn't be surprised if they come out strong when they get their process stuff sorted.
paulmd|5 years ago
Yeah, it's primarily a problem of node here. AMD shrunk and Intel has been struggling to get their node up and going. If Intel had a working 10nm-class node the picture would be very different. They have a whole bunch of new architectures in the pipe that get back to making substantial IPC improvements, they simply can't manufacture them yet. Even if they could simply port Skylake to 10nm it would do OK.
TSMC are kind of the real star behind AMD's success. AMD is benefitting hugely from Apple and Qualcomm and others who sink a lot of money into TSMC, while Intel has to get it running all by themselves. TSMC has substantially outrun every other foundry on the planet, the situation would be equally bad if Intel were stuck with GloFo or Samsung or IBM, right now you're either on TSMC or you're not competitive.
The one part that AMD got right is the chiplet design. Being able to manufacture server processors out of chiplets that are a fraction the size of a monolithic laptop processor and have them lose effectively no performance from scaling like this lets them use TSMC even if yields might not be fantastic on an equivalent monolithic chip.
Part of the reason they have laptop processors running a year behind the desktop/server chips is, those are monolithic processors, not chiplet, and they'e bigger and yield worse than chiplets. In this segment, Intel beat AMD to market substantially - Ice Lake has been in the market since like September, the first Renoir laptops are just shipping like sometime this month. I was looking at laptops at Costco before Thanksgiving and just under half the laptops there had Ice Lake, so it's been available in substantial numbers for a while. Renoir is still better, but it is a leapfrogging dynamic unlike, say, the server market where AMD is just better. Ice Lake actually still outperforms Renoir in per-thread performance, just not iGPU performance and has fewer cores overall, so Intel's uarch isn't terribly uncompetitive when they can actually manufacture it. Zen3 will probably match Intel and then Tiger Lake will leapfrog AMD again a bit.
I have my doubts that giant monolithic Ice Lake-SP will ever be manufacturable at any competitive cost. The lack of consumer laptop/desktop 8C Ice Lake speaks against this as well, if you can't yield an 8C at competitive prices how are you supposed to yield a 38 core processor? But, Intel seems to be plowing forward with the launch anyway this year, so maybe it is, who knows.
To make a short story long, Intel really needs to get its node situation straightened out, and probably needs to transition to a chiplet style layout to make that happen, especially for the server stuff. Obviously it is not trivial to get chiplets to scale well in terms of performance. But Intel is not behind AMD so much as they're behind TSMC, and once they can actually manufacture their products on a competitive node then they'll be back in the game.
DeathArrow|5 years ago
The best bet for Intel would be to sell the fabs like AMD did, or keep them but start attracting other customers to spread costs when optimizing for a particular node.