Intel Launches New Core i9-9980XE 18-Core CPU With 4.5GHz Boost Clock (hothardware.com)

MojoKid writes: When Intel officially announced its 9th Generation Core processors, it used the opportunity to also unveil a refreshed line-up of 9th Gen-branded Core-X series processors. Unlike other 9th Gen Core i products, however, which leverage an updated Coffee Lake microarchitecture, new processors in Intel's Core-X series remain based on Skylake-X architecture but employ notable tweaks in manufacturing and packaging of the chips, specifically with a solder TIM (Thermal Interface Material) under their heat spreaders for better cooling and more overclocking headroom. The Core i9-9980XE is the new top-end CPU that supplants the Core i9-7980XE at the top of Intel's stack. The chip features 18 Skylake-X cores (36 threads) with a base clock of 3.0GHz that's 400MHz higher than the previous gen. The Core i9-9980XE has max Turbo Boost 2.0 and Turbo Boost Max 3.0 frequencies of 4.4GHz and 4.5GHz, which are 200MHz and 100MHz higher than Intel's previous gen Core i9-7980XE, respectively.

In the benchmarks, the new Core i9-9980XE is easily the fastest many-core desktop processor Intel has released to date, out-pacing all previous-gen Intel processors and AMD Threadripper X series processors in heavily threaded applications. However, the 18-core Core i9-9980XE typically trailed AMD's 24 and 32-core Threadripper WX series processors. Intel's Core i9-9980XE also offered relatively strong single-threaded performance, with an IPC advantage that's superior to any AMD Ryzen processor currently.

Comments

Pricing

[TFlan91]

The pricing though... AMD still edges out in my book.

Re:Pricing

[Joce640k]

If that's the price of bragging rights then I'll skip this one.

Re:Pricing

[omnichad]

SMP should be perma-off I'd say

So you want a single core computer? It has nothing to do with hyperthreading (which itself is related to SMT). Speculative execution is only one of the optimizations involved, and the one that has all the security issues.

Re:Pricing

[alvinrod]

It's not quite that simple. Both companies do all of their packaging (final assembly) in Asia, so no one gets bragging rights on that part and everyone's CPUs read "Made in China" or "Made in Malaysia".

You would need to look at where the chip is diffused. I believe that the current AMD Zen chips are being produced at Global Foundaries Fab 8 which is located in New York. The next batch are being done at TSMC, so you would have a point there.

I believe that the design teams for both CPUs are largely in the U.S. as well, but both AMD and Intel have some smaller teams throughout the world.

Re:Pricing

[Anonymous Coward]

Actually you are both wrong. Intel's major fabs are in the US and AMD uses Global Foundries whose main CPU fabs are in Germany and the US. They both have plants elsewhere but they handle other products. Most other products including Apple's SoCs are made in Taiwan though under contract with TSMC who are aside from Global Foundries probably the largest gun for hire fab around. Samsung has their own fabs as well. There are a lot of fabs in the PRC and many chinese chips are made there but Intel and AMD chips aren't

Fewer Cores and Hypertrheading is likely better

[Crashmarik]

For almost all desktop use.

Unless your desktop is doing something that parallelizes really well you probably will never notice the benefits of this.
Even things that benefit from parallel processing are far better served by running them on truly parallel architectures. You have an application that can support fine grained parallelism, why run it on 18 cores of X86 when you run it on 1500 cores off a graphics card ?
 

Re:Fewer Cores and Hypertrheading is likely better

[GregMmm]

What someone could use this for is virtualization on your desktop. But at that price you might as well get a Xeon proc and call it a day.

Re:Fewer Cores and Hypertrheading is likely better

[macklin01]

Shared memory parallel codes (OpenMP) could benefit, though. Many (originally single-threaded) or homemade scientific applications run in this space: get some parallelism for relatively little work (insert pragmas, be careful to be thread safe, and test test test), without all the extra work of redesigning those simulations for efficient message passing.

You certainly find problems where it is much better bang for the buck to throw an expensive processor and OpenMP ( O($10^2 to $10^3) ) at a problem than to throw specialized MPI development effort ( O($10^4 to $10^6) ) at that problem. Especially when the first step to any hybrid OpenMP-MPI code is to work on single compute node performance with OpenMP, before connecting nodes with MPI.

Re:Fewer Cores and Hypertrheading is likely better

[thegarbz]

You have an application that can support fine grained parallelism, why run it on 18 cores of X86 when you run it on 1500 cores off a graphics card ?

Because a graphics card is not just 375 traditional CPUs jammed into a single package and just because something can scale to 18 cores doesn't mean it would run better on 1500 GPU cores.

Re:Fewer Cores and Hypertrheading is likely better

[Crashmarik]

Browsing separate tabs really shouldn't be taking cpu at all. Haven't looked at Firefox's source but I will typically have a hundred plus tabs open and notice very little draw on my cpu resource. Right now it's pulling 5.1% on a quad core cpu with a guess at around a 100 tabs open and and at least 5 that are interactive. Streaming once again suspect the optimum use of the dollars is buying a higher end graphics card.

Re:Fewer Cores and Hypertrheading is likely better

[Crashmarik]

Thanks for showing me wrong.

I should have gone with my first impression that you were an idiotic troll when you said this

Play videos in multiple tabs. Access multiple crappy javascript sites. There are any number of ways to consume cpu in multiple tabs
https://slashdot.org/comments....

But I gave you the benefit of the doubt. My bad.

Now you are coming up with this

Why is it necessary to explain it to you in words of one syllable? Lose frame rate.

And you have removed all doubt.

You turn on streaming you are going to lose frame rate no matter what you do. The questions are, how are you are going to lose more and what is the best use of resources to build the system.

Vulnerabilites

[cyberchondriac]

I didn't see any mention of addressing Meltdown, Spectre, L1TF.. so I assume those general architecture issues are not yet addressed, this is still Skylake.

Re:Just in time for Winter..

[DamnOregonian]

warm my home and play Crysis.

On medium settings

Hello again Mr.7980XE

[Z80a]

I see you got a fancy new power curve, soldered TIM and nothing else!

Re:Or maybe not

[barc0001]

Yep, and you demonstrate that wonderfully. The AMD Threadripper is better for some things, the Intel chip is better for some things, and *depending on your needs and budget* each could be "better".

Intel wins the IPCs, but is crushingly expensive. Many people would take the half price of the AMD part and be quite happy with it. Others for whom money isn't that much of an object will go with the 9980XE. Still others who need Blender, Cinebench or POVRay workloads done would be fools to buy anything but the Threadripper.

All in all, everyone has a different need and will cherrypick based on that need.

It's not the fastest desktop processor

[aliquis]

It's not the fastest desktop processor when it trails 24 and 32 core ThreadRippers.

That's not how it works. Fastest doesn't mean slower.

Re:Or maybe not

[Tough Love]

Intel wins the IPCs

Not for very much longer. And for me, like most of us, value is the decider. I'm also finding AMD's thermal performance excellent these days, and I just love how long the sockets last. AM4 really delivered on its future-proof promise.

Re:Or maybe not

[Tough Love]

I like the TR parts but they really need to cut down the idle power, 100W+ at idle (!)

Tom's Hardware says 35 watts for the 2990WX at idle.

For my trusty Ryzen 1700 box, the entire system power measured at the wall is 38 watts at idle.