Future Intel 'Broadwell' chip could pack up to 18 cores

Shawn Knight

Posts: 15,256   +192
Staff member

intel broadwell cpu chip 18-core

Intel is said to be working on a Broadwell chip that will pack a whopping 18 processing cores. The chip, which would feature the highest core count to date, won’t arrive until sometime in 2015 according to a report from VR-Zone.

Instead of just speeding up the cores, Intel will simply pile more of them on each die. That may not immediately be useful unless you actually have software designed to take advantage of multiple cores, but I digress.

The 18-core Broadwell-EP or EX Xeon chip will be based on the company’s pending 14-nanometer process although there will be multiple variants. For example, one is expected to be an eight to 10 core high performance desktop / workstation part while another with 12 to 16 cores will target enterprise servers.

The first consumer Broadwell chips are expected sometime in the first half of next year as the second processor in the Haswell microarchitecture. In accordance with Intel’s tick-tock production methods, Broadwell will feature the aforementioned reduction in manufacturing process to 14-nanometers.

During a public demonstration at IDF back in September, CEO Brian Krzanich said the chip would allow systems to provide a 30 percent improvement in power use compared to Haswell.

In related news, CPU World is reporting that Intel is also planning to bring Broadwell to mobile devices. With a TDP of as low as 4.5 watts, such a chip could be a perfect fit for tablet use. Naturally, Intel declined to comment on the matter.

Permalink to story.

 
Writing applications that take advantage of multiple cores is amazingly difficult and even if you write the application as multi-threaded, the most cores used is three or four. I don't know of too many applications that would take advantage of anything more than four cores...
 
Writing applications that take advantage of multiple cores is amazingly difficult and even if you write the application as multi-threaded, the most cores used is three or four. I don't know of too many applications that would take advantage of anything more than four cores...
Don't think single application...

Should add, Broadwell can't come quick enough for me!
 
I don't know of too many applications that would take advantage of anything more than four cores...
Which came first, the chicken or the egg?

Which do you think needs to come first, an 18 core CPU or applications that will take advantage of 18 cores? I'm pretty sure applications will never exist until there is physical hardware to support them. Even then most applications won't support 18 cores, until 18 core CPU's become the norm.
 
5 years ago having a Core 2 Duo was enough, soon it wasn't. Today I have a 4th Gen Core i5 and when playing online I can't avoid noticing some lag due to the 100% load to the CPU by having only the game, one site loaded in the browser and Skype opened at once. If I close the browser and Skype I don't have a single noticed frame drop online, it's not a bottleneck with the graphics or RAM (I have a GTX 670 and 16GB of RAM) simply the game is taxing a 100% load to the CPU, if a program uses only 10% of CPU while playing... it gets noticed.

Shortly: bring it on, fast is never enough.
 
Writing applications that take advantage of multiple cores is amazingly difficult and even if you write the application as multi-threaded, the most cores used is three or four. I don't know of too many applications that would take advantage of anything more than four cores...
18 cores are for Xeon (Broadwell-EP/-EX). In an (enterprise) industry where some off the shelf software is licenced on a "per core" basis, and the bulk of the remaining software is custom code, you can guarantee that there will be pretty good core utilization.

As far as I'm aware, Broadwell-D (mainstream desktop) is likely to be quad core.
 
It is about time for a bump to 6 core though, in my opinion.
Intel seem intent on lowering power consumption-and leveraging a higher proportion for graphics use- over increased core count for the mainstream (which also double for mobile and embedded markets). From what I've heard, Broadwell-D will be ~55W TDP. High end desktop/Xeon have less constraints with power (no iGP, users valuing throughput over energy use) so the march in core count continues. Haswell-E looks to retain the 8 core/16 thread count of Ivy-E ( 3GHz from an early 8-core engineering sample looks promising).
 
I always thought a CPU with 10 Cores all at 10Ghz would be truly revolutionary but it seems they stopped trying to find ways to improve Ghz :/ shame...
 
Well AMD's Mantle seems to like as many physical cores as it can get.....so if that takes off then there could could be a big desktop market for MOAR CORES.
 
I always thought a CPU with 10 Cores all at 10Ghz would be truly revolutionary but it seems they stopped trying to find ways to improve Ghz :/ shame...
Outright clock speed generally requires a deeper architectural pipeline which is usually detrimental to overall performance (or results in a larger cache requirement in terms of cohesiveness and die real estate) and increases in power and heat. Shorter and more numerous pipelines mean less misprediction- so while the clockspeed is lower, the actual processor becomes more efficient.

For a quantum leap in performance you also need a major shift in design materials - such as that evidenced by the transition from aluminium to copper interconnects in early processors. Maybe the switch to more exotic metals compounds (I.e. Indium and Gallium based) offers a leap in relative performance, but the focus of modern processors is performance per watt and maximizing the instructions per cycle rather than straight line speed which can be detrimental in any case due to branch prediction misses.
 
I always thought a CPU with 10 Cores all at 10Ghz would be truly revolutionary but it seems they stopped trying to find ways to improve Ghz :/ shame...
Outright clock speed generally requires a deeper architectural pipeline which is usually detrimental to overall performance (or results in a larger cache requirement in terms of cohesiveness and die real estate) and increases in power and heat. Shorter and more numerous pipelines mean less misprediction- so while the clockspeed is lower, the actual processor becomes more efficient.

For a quantum leap in performance you also need a major shift in design materials - such as that evidenced by the transition from aluminium to copper interconnects in early processors. Maybe the switch to more exotic metals compounds (I.e. Indium and Gallium based) offers a leap in relative performance, but the focus of modern processors is performance per watt and maximizing the instructions per cycle rather than straight line speed which can be detrimental in any case due to branch prediction misses.
Case in point was Intel's Pentium 4 processor.
 
When you go online you probably loading malware or botnets more then your stupid game. Hahaha. Welcome to my world zombie!
 
All very nice Intel but what about the CHEAP thermal paste being used on Haswell now? For the prices on these high end CPU chips do you think we could get a drop of solder on them to deal with the furnace temps?
 
Wouldn't more Ghz be more beneficial than cores? I am just thinking wouldnt a 10Ghz 4core cpu beat a 4Ghz 10core cpu in rendering and other real world applications?
 
Wouldn't more Ghz be more beneficial than cores? I am just thinking wouldnt a 10Ghz 4core cpu beat a 4Ghz 10core cpu in rendering and other real world applications?
Depends upon how the application were coded, but assuming that the core usage were fully optimized, then generally no. Basically, the faster you go the more errors creep in, and the more errors (branch prediction misses) the more idle cycles are introduced as the cache hierarchy attempts to resolve the issues.
A simple example would be that performance doesn't scale with overclock. If you overclock a CPU by 50%, you generally don't increase it's performance by the same margin- and as the speed increases the margin becomes smaller because of inefficiencies in the hardware (power requirement, heat build up and dissipation, cache latency) and external restrictions such as memory bandwidth.
 
Back