Intel is reportedly prepping the 18-core / 36-thread Core i9-7980XE (and other high-end...

Shawn Knight

Posts: 15,240   +192
Staff member

AMD’s resurgence in the high-end desktop scene has apparently sparked a fire under Intel. The chipmaker, which has stood unchallenged atop the performance desktop mountain for years, is reportedly working on a potent lineup for its X299 platform that includes an 18-core monster.

Update: And they are official! Check out Intel's confirmed Core-X and Core i9 lineup here.

According to VideoCardz, which says its intelligence was obtained from multiple sources, the Intel Core i9-7980XE will be the new performance king with 18 cores and 36 threads (other details like L3 cache size and clock speeds wasn't shared). There’s also said to be a Core i9-7960X with 16 cores / 32 threads and a Core i9-7940X with 14 cores and 28 threads in addition to the stable of chips we’ve previously heard about.

Intel even reportedly has new boxes in the works as spied above.

As we learned earlier this month, AMD is also gearing up for an epic battle this summer with its Ryzen 9 “Threadripper” chips. That lineup, if you recall, is said to be capped by the Ryzen 9 1998X, a 16-core / 32-thread beast with a base clock of 3.5GHz (Boost to 3.9GHz), quad-channel DDR4 support and access to 44 PCIe lanes. That chip, along with the 1998, will carry a 155W TDP, rumors indicate.

Some expect Intel to announce its new X299 platform as early as next month although as always, rumors of this nature should be taken with a grain of salt until we hear official word from the horse’s mouth.

Permalink to story.

 
These multi-core processors are SMPs (Symmetrical Multi Processors) and do little to nothing for the desktop user. They can be useful for server systems ... but only marginally.

Studies on increase processing throughput compared to the number of SMPs shows two conclusions:
  1. the improvement by adding another SMP is not linear.
  2. after 4 processors, the returns really diminish and after 8, improvements are in the 10% range.
Why? Because the system needs to protect access to the shared resources, (memory, common devices) using locks and soon the system is spending more time in lock management than doing useful work.

Hi multi-core systems quickly become cost ineffective.
 
Sadly, the only reason this is happening is to make sure AMD doesn't get to leap ahead of Intel. Without AMD, Intel would continue pitching the current products for another decade.

Fair competition makes all the difference in the world.

I used to be a consumer of the top CPU products from Intel. But these days, for the lack of very specific tasks, a regular 4-core is sufficient for 99% of tasks I'm doing. Now I just enjoy reading about the progress :)
 
So all it took for Intel to use the kaioken was for AMD to show up with a little bit of aura. Interesting.
 
These multi-core processors are SMPs (Symmetrical Multi Processors) and do little to nothing for the desktop user. They can be useful for server systems ... but only marginally.

Studies on increase processing throughput compared to the number of SMPs shows two conclusions:
  1. the improvement by adding another SMP is not linear.
  2. after 4 processors, the returns really diminish and after 8, improvements are in the 10% range.
Why? Because the system needs to protect access to the shared resources, (memory, common devices) using locks and soon the system is spending more time in lock management than doing useful work.

Hi multi-core systems quickly become cost ineffective.

For short tasks assigned on the fly by an OS, there's some truth to this.

For highly-parallel operations optimized for available threads, though, more cores and threads can be quite useful.

So it sounds like ordinary desktop users shouldn't be interested in more than 4 cores...

Until you think about it a little deeper. Y'see, that's exactly what we used to hear from proponents of single-core CPUs when confronted by the arrival of dual- and four-core CPUs. Nobody needs that! Operating Systems don't do a good job of using the extra cores!

Thing is, that's entirely a software issue. And software evolves.
 
These multi-core processors are SMPs (Symmetrical Multi Processors) and do little to nothing for the desktop user. They can be useful for server systems ... but only marginally.

Studies on increase processing throughput compared to the number of SMPs shows two conclusions:
  1. the improvement by adding another SMP is not linear.
  2. after 4 processors, the returns really diminish and after 8, improvements are in the 10% range.
Why? Because the system needs to protect access to the shared resources, (memory, common devices) using locks and soon the system is spending more time in lock management than doing useful work.

Hi multi-core systems quickly become cost ineffective.

For short tasks assigned on the fly by an OS, there's some truth to this.

For highly-parallel operations optimized for available threads, though, more cores and threads can be quite useful.

So it sounds like ordinary desktop users shouldn't be interested in more than 4 cores...

Until you think about it a little deeper. Y'see, that's exactly what we used to hear from proponents of single-core CPUs when confronted by the arrival of dual- and four-core CPUs. Nobody needs that! Operating Systems don't do a good job of using the extra cores!

Thing is, that's entirely a software issue. And software evolves.

Unfortunately software typically only evolves the *right* way after all WRONG ways are tried.
 
We shouldn't be shortsighted. A 8+ core CPU may not mean much for desktop users right now but these designs will eventually push software devs to create their software in ways to exploit multi-core hardware's potentials, maybe not in in short-term but soon enough (and I know not all kinds of software can be multi threaded)
 
Unfortunately software typically only evolves the *right* way after all WRONG ways are tried.
Heh, so true!
But software *will* get there. Not immediately, of course. In the short term, 4 cores is fine for most desktop users. But as more cores reach the mainstream at reasonable prices, there will be a rising motive among coders to take advantage.
 
I guess we're on the slippery slope of increased software licensing costs with OS and applications being core based.
 
I guess we're on the slippery slope of increased software licensing costs with OS and applications being core based.
In the short term, sure.
Longer term, I expect AI to generate efficiencies in coding, same as it will do in most other fields.
 
Realistically, 99% of consumers don't need anything that has more than 4 cores. The other 1% incorrectly think they do.
Anyone who does any typical (read typical) heavy workloads that arent gaming usually benefit greatly from more cores.

In the next 10 years, IPC improvements will mean much less than proper multithreaded software design. With consoles and their 8+ cores helping move gaming into a more heavily multithreaded timeline, 4 cores (without 8 threads anyway) will become a bottleneck soon.

Currently, no gamer needs more than 4 cores, but in 2-3 years? buying a new 4 core CPU now is not a great buy unless you upgrade every 2 years.
 
Currently, no gamer needs more than 4 cores, but in 2-3 years? buying a new 4 core CPU now is not a great buy unless you upgrade every 2 years.

Ryzen owners will strongly disagree. If you stream your games, you want more than 4 cores because 4 are dedicated to the game and the others can be used for discord, skype, twitch, recording gameplay, etc.
 
In the short term, sure.
Longer term, I expect AI to generate efficiencies in coding, same as it will do in most other fields.
Speaking of AI, why can't an AI determine which cores are to be used? Is that too shortsighted for AI usage?
 
Hmm; The alternative to SMP is the MPP (Massive Parallel Processing):
  • Massively Parallel Processing (MPP) is the coordinated processing of a single task by multiple processors, each processor using its own OS and memory and communicating with each other using some form of messaging interface. MPP can be setup with a shared nothing or shared disk architecture
Systems of this architecture interconnect via a High Speed Bus and process portions of a problem on one processor and other portions on other processors, coordinating, controlling and finalizing the results, very similar to using Threaded Architecture. As locking is no longer a problem as in the SMP case, adding another MPP is not perfectly linear, but almost. The gains are above 90% improvement per.

The Tandem T16 system was MPP and could be expanded to 255 processors. Parallel sorts of massive data was amazingly fast as it performed nearly at P/N-1; Processing for 1cpu divided by N-1 MPPS

While AI has been hyping for decades now, the MPP will be necessary to sift the massive amounts of data need to make AI work in the real world.
 
Speaking of AI, why can't an AI determine which cores are to be used? Is that too shortsighted for AI usage?
That's a misuse of cores, just like on windows, picking processor affinity typically hamstrings the OS. When there's a DMBS and a Network server on the same machine, the OS does a better job of picking processors than user assignments.
 
In the short term, sure.
Longer term, I expect AI to generate efficiencies in coding, same as it will do in most other fields.
Have you done any AI coding? Prolog, Lisp, ... which did you use?
You should see the code needed to make IBM's Watson just play Chess . . . you wouldn't be so glib
 
It's not good enough to match a competitors equivalent product you got to make sure it's priced very similarly as well and Intel rarely does that .
 
Back