Replies to post #142752 on Intel Corp (INTC)

[This Causes an Error]

Once you start giving up on x86 in mobility what will be its long term future when you see how mobile volume has transformed ARM and Apple ? x86 will then become the niche Server Risc of the future. This is where your short-term strategy will lead Intel.

I don't agree here. Intel sells billions of dollars worth of X86 processors into the PC market as well as to a very broad range of server/datacenter customers. Barring a major screw-up on Intel's part, I don't see x86 becoming a niche SPARC-like ISA even if Intel doesn't win with x86 in mobility.

I know you like to rag on Atom but it is still only a dual-issue cpu competing against triple-issue A57/A72 and hex-issue A9, there really is so much future potential scope to develop both Atom and Core with the latter still only a quad-issue cpu and as Intel can change its ISA at the same time it can maximize the performance increase rather than waiting in line like everyone else for the next release of ARMv8.

FYI, Skylake is 5-wide and if there is a hit in the uOp cache it is effectively 6-wide.

http://pc.watch.impress.co.jp/video/pcw/docs/724/408/p01.pdf

But yes, I agree, if Intel put the work into it, Atom could be a very nice, high performance mobile architecture. Intel needs to execute though. I am very interested to see what uArch changes Intel has made with Goldmont, which we should see in the market in the first half of 2016.

I really could see Core-M in a phone eventually if Intel choose to put it there maybe at 10nm-7nm. The way Core has developed lately is that it has lost power more than it has gained performance which is fine for the moment and good for its eventual use in phones but I really could see the eventual need of a third line of x86 cpus, say a 6-issue desktop/server cpu specialist of say Itanium ipc levels as the performance distinction lines between x86 and ARM are definitely getting blurred with each new ARM generation.

100% agree. I think we are beginning to see this though with Skylake. The Skylake core that is in client PCs is not the same one that will be in the Xeon E5/E7 chips come next year. It is said to come with a number of performance enhancements/improvements.

The obvious known improvement is the inclusion of AVX-512 which should be awesome for HPC, but I suspect that they will need to do a lot of additional tweaks in the architecture in order to be fully able to effectively use AVX-512.

I would like to see even larger distinction between the server cores and the client Core processors, though, as you suggested.

[chipguy]

I know you like to rag on Atom but it is still only a dual-issue cpu competing against triple-issue A57/A72 and hex-issue A9, there really is so much future potential scope to develop both Atom and Core

I think Intel should make Atom even leaner and lower power and redirect
it to replace the Quark (486? P5?) core for embedded control, IoT, and
also target cheap ass smart phones.

Intel should probably split Core into two similar uarchs, Core heavy and
Core light. Core heavy pushes single thread and FP performance and is
circuit optimized around higher frequency and power targets. This would
be used in high end laptop, desktop, workstation, and moderate core count
processors for low socket count servers for HPC and technical computing.
The Core light would have lighter FP resources and be circuit optimized
around moderate frequency and lower power targets. It would be used in
products for high end phones/phablets, tablets, low to medium end laptops,
and high core count chips for scale-out servers and high core count chips
for scale-up, high socket count, high RAS, commercial workload servers.