Is the death of Atom really a Bad Thing? Talking about die sizes

[sm625]

Why is intel actually leaving the phone space?

Because intel is selling chips that are smaller than some phone SoCs for prices that are 5-10 times higher. And people are paying it, for whatever reason. But they wont pay it for phones. So intel stops making them.

Core m isnt that much bigger than atom, maybe 20%, yet they charge 5 times more? Intel is wasting a large amount of die space trying to bring their latest gen graphics to atom, but the cpu cores are so slow it is just a waste. This is probably what made them realize how much of a dead end atom is. It makes much more sense just to make Core m be lower power.

 

[poofyhairguy]

Core m isnt that much bigger than atom, maybe 20%, yet they charge 5 times more?

That is the real problem.

I personally don't see Atom dying as a bad thing- it means that Intel is finally feeling some competition in its segment that AMD could never provide.

I hope Apple and eventually Qualcomm compete with Intel in every important segment and drive down their margins in every segment I care about.

 

[Zodiark1593]

That is the real problem.

I personally don't see Atom dying as a bad thing- it means that Intel is finally feeling some competition in its segment that AMD could never provide.

I hope Apple and eventually Qualcomm compete with Intel in every important segment and drive down their margins in every segment I care about.

Makes me wonder if Intel has a bigger Core waiting in the wings to take over their desktop/server/laptop segments when ARM catches up to Core. I'd assume that Intel probably saw this coming for some time and made some sort of preparations for when they can no longer take such profit margins on the current Core line.

 

[Qwertilot]

Think I've seen some mutterings about it for servers somewhere.

The problem with it for consumer desktops, is well, what ever would they offer beyond a decently clocked quad clocked core that anyone would want? Of course also a lot of sales in low power notebooks etc.

Its basically a non trivial problem for their business model. They'll have some potential solutions lined up. De emphasising consumer chips for one thing.

 

[poofyhairguy]

Makes me wonder if Intel has a bigger Core waiting in the wings to take over their desktop/server/laptop segments when ARM catches up to Core. I'd assume that Intel probably saw this coming for some time and made some sort of preparations for when they can no longer take such profit margins on the current Core line.

I don't think they have an ace up their sleeve they haven't shown us. I think their hope is the HPC market grows, or being willing to bake in special extensions for someone like Facebook helps lock them into Intel as a vendor.

 

[ninaholic37]

Of course you can't ignore "Netbook 2.0", which is what Apollo Lake is for.

Does that mean Apollo Lake won't be in 2-in-1s? Praise the lord! Hopefully they will die out and Acer can resurrect the 10.1" Aspire One from 2012 instead.

 

[IntelUser2000]

I thought about that, but I imagine they would either not do much to the CPU part beyond shrinking it or optimize the core more toward parallel compute rather than something that would be useful as a cheap CPU.

I believe they would continue to do so.

They said future cores would use Atom to reduce single thread performance deficit with Xeon chips. The new Atom cores are 3x+ performance of the P54's in the Knights Corner one.

Considering the area, Silvermont is a huge improvement even over previous Atom, nevermind P54C. They are nearly the same size. That's one part Core cannot think of even doing.

Without single threaded performance it would literally have been impossible to achieve what they have done with Knights Landing. That is, offer competitive performance without corner-case optimizations out of the box practically. SpecFP_Rate performance equivalence to a 2P Xeon setup is simply excellent. You get to run everything you have at a decent rate, and get huge future speedups when you optimize the applications you want. AVX-512 compatibility with SKL Xeon means SKL optimizations benefit Xeon Phi. The same sentiment is mirrored at benchmarks(though few) out there.

They realized that being able to run current code at decent rate with future expansion capabilities are what both Itanium, and Atom on Android lacked for success. They are both gone.

Knights Hill(successor) guesses: Goldmont cores are likely 3-issue cores. I doubt they'll expand core count much, because scaling starts to fall off on many general purpose cores(its also I believe no coincidence their research chip ranged from 60-80 cores). I am guessing the possibility of having 3x 512-bit FP cores, for 50% throughput increase per core. Improved design may also indicate higher clocks. 50% FP increase per core x 20% higher clock x Core increase to 80 from 72 = 2x performance and performance/watt.