Intel's 28-Core 5GHz Processor And Test System Breaks Cover

[PaulAlcorn]

We caught up with Gigabyte at the show to get a closer look at its new motherboard that supports Intel's 28-core processor. We also grabbed pics of the internals of Intel's test system.

Intel's 28-Core 5GHz Processor And Test System Breaks Cover : Read more

 

[Hoetan]

How many cores is too many cores? *thonk*

 

[mihen]

There is no such thing as too many cores as long as it does not hinder single threaded performance.

 

[giovanni86]

Intel needs to price it right. It can't be more then 2k. I feel 2k is too steep. With AMD having a big piece of the pie now, intel shouldnt hold out. We all know threadripper 2 will be priced equal to its predecessor thats just how AMD rolls and they've won strides doing so. Now we just need AMD to announce a killer AMD GPU to dethrone Nvidias jenson so he can finally wake up and see that delaying a release will cost you greatly!

 

[singemagique]

Gah these would be awesome for virtualization. Need a couple for my homelab!!!

 

[Valantar]

Intel needs to price it right. It can't be more then 2k. I feel 2k is too steep. With AMD having a big piece of the pie now, intel shouldnt hold out. We all know threadripper 2 will be priced equal to its predecessor thats just how AMD rolls and they've won strides doing so. Now we just need AMD to announce a killer AMD GPU to dethrone Nvidias jenson so he can finally wake up and see that delaying a release will cost you greatly!

That, my friend, is some serious wishful thinking. The current 18-core model is already at $2000, and the enterprise variant of this is ~$10000. Selling the same product (even with ECC or some PCIe lanes removed) at 1/5th of the price? Yeah, that's not going to happen.

My guess: around ~$3500. Remember, chips like these are mainly for the "never even looks at the price tag" crowd.


Also: has nobody noticed that utterly insane VRM setup? I think I counted 28 phases (or 14 doubled, I suppose, which is mostly the same thing). Holy utter ****. No wonder, really, as this thing was likely pulling 800W through the CPU cores alone (CFL at 5-5.1GHz is around 160-200W - subtract ~10W for the uncore and divide by 6, and you have 25-30W per core). But still: wow. Wonder how many current cases the motherboards for these things will actually fit into.

 

[BulkZerker]

"The 5.0 GHz overclock makes Intel's new processor all the more impressive compared to AMD's relatively tame clock speeds. "

Not at all impressive when..

"there wasn't enough power handy to run both the chiller and the system simultaneously."

This setup needs more Than 20 amps to run? Ha, hah hah, *amused chortle*.

Intel is in full panic right now to do this as their response to a processor that will cost less (I'm guessing 1/4 the price of this chip if not less), will use less power (maybe 400 watts full load), and "somehow" manages to provide comparable performance.

The only reason to buy Intel hedt right now is you flat do not have space and must have as much performance as possible in that limited space.

 

[Valantar]

"The 5.0 GHz overclock makes Intel's new processor all the more impressive compared to AMD's relatively tame clock speeds. "

Not at all impressive when..

"there wasn't enough power handy to run both the chiller and the system simultaneously."

This setup needs more Than 20 amps to run? Ha, hah hah, *amused chortle*.

Intel is in full panic right now to do this as their response to a processor that will cost less (I'm guessing 1/4 the price of this chip if not less), will use less power (maybe 400 watts full load), and "somehow" manages to provide comparable performance.

The only reason to buy Intel hedt right now is you flat do not have space and must have as much performance as possible in that limited space.

AnandTech got some hands-on time with the PC and the chiller. Apparently the chiller requires 9.6A alone, with a cooling capability of ~1.7KW. Alongside that, we can expect a single ~5GHz CFL core to consume ~25-30W. Times 28, that's 700-840 for the CPU cores alone. Then there are VRM losses, which even given the silly 28-phase VRM on the test board would be significant at that kind of power (even at 95% efficiency that's another 35-42W). Then there's the CPU Uncore (negligible at this wattage, but still). and 6-channel memory (looks like the build was fully populated with 12 DIMMs) which should account for a decent amount of power again. And whatever GPU and so on they had in there, motherboard power, and PSU losses. And probably a hefty water pump, if the chiller wasn't running while the demo was done.

 

[Non-Euclidean]

What a joke. This monstrosity by Intel is as feasible as driving a rocket powered car down the road next year.

 

[John Nemesh]

This is a desperation move from Intel. It's nowhere NEAR release, as was shown when Intel (wisely) decided to pull these from the show floor (including from other OEMs).

AMD, on the other hand, showed a consumer-ready product that is functional with air cooling, and will be released this fall.

Some people don't want to admit that the company with an 89% marketshare is in trouble. This show proves them wrong. Intel is going to have a rough 2018, and an even rougher 2019.

 

[bugnguts]

1.7kW of cooling is more than the largest window mounted AC units, 750W to 1400W is typical for window AC. On top of that, the test was only 7s long. Intel has hit a thermal wall. I've said it before and I will repeat Intel is stuck at 14nm because 10nm is too small. Yes, less power but also much less cross sectional area to transfer heat through chip and then chip to the spreader. This is why Intel only releases low power low core count at 10nm. Who wants to upgrade to a slower processor? This hits Intel's server-class hard due to high core counts. This is a big problem because the problem is inherent in the architecture. They would have had a few more years to work on it but AMD shows up with Ryzen which has higher IPC, core counts and clock speeds with the 2nd iteration which suggest this architecture has room to grow. AMD then shows an 7nm EPYC and says the chip will ship by end of year. If the 7nmEPYC has higher clocks and better efficiency Intel server chips will have little appeal.

 

[LordVile]

Intel needs that 10nm node fast or they're going to lose more and more market share to AMD. Their only advantage now is in adobe premier due to Ryzen not having an iGPU and high FPS gaming which Zen 2 might rectify if they can bump the speed with a die shrink.

 

[shpankey]

Intel using an unrealistic phase change cooling system that no consumer will ever use. hahahaha

 

[mischon123]

The age of silicon has now officially ended in 2018 and no new tech in sight. The stuff is so dismal we throw 1.5kw at a chip and cannot even simulate a fly brain. We are now stuck. No way to get to the singularity. Sad news.

 

[Giroro]

All those cores, but still no hardware fix for Spectre.

 

[mapesdhs]

GN summed this up nicely:

https://www.youtube.com/watch?v=tRH0-QwhvVQ

Intel's PR is complete nonsense. Basic math on how a 7980XE behaves proves there's no way such a thing could function with normal ambient-based cooling, so for consumers it's completely irrelevant, yet tech headlines all over the place are reporting it as if it's a genuine 28c 5GHz part. This article is guilty of the same misguided implication in its wording. This is an oc'd sub-ambient cooled part and should be reported as such.

 

[mapesdhs]

Intel needs to price it right. ...

Price what right? There's no product to price. This was a sub-ambiently cooled oc'd tech demo, nothing more. Big whoop, yet people are falling for it left, right and centre.

We had months of hue & cry over previous Intel PR nonsense with security flaws, spin over KL-X, chipset change lockouts, etc., have people learned nothing? If it sounds too good to be true (well duh) then it probably is. A few seconds of math on the power draw of a 7980XE would show there's no way Intel can ship a 5GHz 28c SL-X part, it's just deceptive PR designed to take the spotlight away from AMD.

 

[Karadjgne]

Told you there was a fish tank and chiller... Hah.

Prolly see a resurgence of dumb postings like 'how can I make this work with my refrigerator loop?'

Kudos to Intel for finally getting off its collective duff and beating Amd in every department tjis time, but I think they got a few wins backwards, like price for one. The better cpu should be cheaper than the competition, not priced out of anybody but NASA's pocketbook range.

 

[saintrasputintm]

Karadjgne...Kudos to Intel for finally getting off its collective duff and beating Amd in every department tjis time...damn son, you went full retard !

 

[PaulAlcorn]

GN summed this up nicely:

https://www.youtube.com/watch?v=tRH0-QwhvVQ

Intel's PR is complete nonsense. Basic math on how a 7980XE behaves proves there's no way such a thing could function with normal ambient-based cooling, so for consumers it's completely irrelevant, yet tech headlines all over the place are reporting it as if it's a genuine 28c 5GHz part. This article is guilty of the same misguided implication in its wording. This is an oc'd sub-ambient cooled part and should be reported as such.

You should read the article before commenting. We were the first to report that Intel used some type of sub-ambient cooling --

https://www.tomshardware.com/news/intel-cpu-28-core-5-ghz,37201.html

Tom's and AT were also the first to track down the demo system and the water chiller, which is outlined in this article.