Intel Shows Latest Xeon Roadmap at Data Centric Innovation Summit

Subject: Processors | August 8, 2018 - 07:39 PM |
Tagged: xeon, intel DL boost, Intel, ice lake, dcg, data centric, cooper lake, cascade lake

Today at Intel's Data Center Group's Data-Centric Innovation Summit, they provided a peek into the future of Xeon processors.

View Full Size

Coming later this year are the oft-rumored Cascade Lake Xeons. In addition to supporting Optane DC Persistent memory, Cascade Lake will offer hardware-based mitigations for Spectre/Meltdown vulnerabilities.  

View Full Size

Intel Deep Learning Boost will also make its first appearance in the Cascade Lake products. In its first iteration, DL Boost will provide a vector neural network instruction set (VNNI) based on AVX-512 for faster inference acceleration. Intel is working to add VNNI to industry standard deep learning frameworks like TensorFlow and Caffe.

View Full Size

Next, in late 2019, we have the Cooper Lake architecture. Still based on 14nm technology, Cooper Lake will expand upon Intel DL Boost and add support for the BFloat16 data type, which provides the same level of precision as double precision (32-bit) floating points, but in a smaller (16-bit) data size. 

In 2020, after Cooper Lake, comes Ice Lake – the first 10nm-based Xeon. While details are sparse about what improvements Ice Lake will bring architecturally, Intel has said that it will be compatible with Cooper Lake platforms, giving users an upgrade path.

Source: Intel

Video News

August 8, 2018 | 11:56 PM - Posted by MustSeeCrapertainmentFromFatlizzardBloatBelly (not verified)

Dogs and Ponys mixed with nauseating amounts of Snake Oil marketing speak. This is theater of the paper kind at its best and any Talk of 10nm is taboo. That S/A article says it all and Epyc will clean up the overpriced mess that is Xeon with Xeon's secondary and tertiary layers of product segementation and obsfucated extra costs.

With one fell swoop AMD's Epyc/SP3 MB platform offers 128 PCIe 3.0 lanes and 8 memory channels per socket, on 1P and 2P Epyc/SP3 server platforms. Intel does not offer that strght up because of all that segementation for Milking the suckers again and again and no attempt at all at providing more feature value via less unsegemented offerings. Epyc/Rome will continue the process of adding value and features and more compute per socket while Intel will continue with its milking schemes.

Watch those fat Intel markups and Quarterly Gross margins as they begin their fall back to earth. And plenty of Server Customers will not be wanting the take it or leave it from Intel with no alternative options possible routine! Now with an AMD Epyc Option around, that leave it option for server customers with regards to Xeon now has an Epyc option for those server customers to do just that and not miss a beat.

Just look at that Show that Intel is putting on to distract on paper from what does not exist in a functioning 10nm hardware form. Those 10nm cutout bin chinese market SOCs with fused off GPUs and lower performance than 14nm do not count as they are 100% lower binned residue thrown out there Potemkin Village style to obsfucate the truth.

Just Look at Epyc/Platform's Price/Performance metrics and Price/Feature metrics and Intel is still charging a bit too much to be taken serious. It looks like Intel's 10nm process node is a process node of the future-- and always will be!

August 9, 2018 | 06:41 AM - Posted by Alb (not verified)

Bet Rome will be an huge disappontment ??

You are too much confident and fanboy about AMD and TSMC.

Remember IBM refused 7nm from GF and shifted on a relaxed custom GF 10nm for POWER 10 in 2020.

This tell us actual foundry 7nm is not done for server cpus. Too much leakage,too much power density, too low clock speed.

August 9, 2018 | 11:27 AM - Posted by MarginDeclinesForShortyViaIntelsExcessiveHubris (not verified)

Bet Rome will not even be a disappontment if it's only 48 cores per MCM/Socket instead of 64 cores per MCM/socket! That's still a bit better than what Intel will be offering and Zen2 will come with the expected IPC improvments among others. But it's obvious that you are an Intel shareholder so it's best to begin to hedge your bets. And that is before the Wallstreet Quants more fully plot the, necessary for competition with Epyc's Price/Performance metrics, Intel Gross Margin declines that Intel will suffer in order to stave off Epyc's market share gains into only the upper teens to lower 20s% market share range! And that may not even be possible for Intel this time around with Epyc like it was against Opteron with that being kept in the low 20s% market share range.

I'm not an AMD fanboy as much as I am an Intel Gross Margin decline fanboy in the face of Epyc's, Power9's/10's, and ThunderX2's/Others competition that will help make Intel's high margin/markups take the Icarus route after flying too high and too close to that big burning ball that's fusing H into He.

Intel's has more than just AMD in direct competition and that's includes both AMD's and Nvidia's GPU based compute/AI accelerator products that Intel can not and will not be able to even begin to compete with in the 2 to 3 years time frame. Those things take time and AMD has been firing on all cylinders while Intel has been in meltdown mode in more ways than one lately.

AMD has been switch hitting on the CPU and GPU compute/AI front lately something that Nvidia can not do in the server/HPC markets for x86 based CPUs at least. But with all the new competition and all the Contra Revenue wasted by Intel in that mobile market phone/mainstream tablet market grab fiasco that Intel could have been spending on preparing its server business for the future is flushed down the toilet which is, incidentally, just where Intel's gross margins will be heading shortly over the coming business quarters.

AMD's got that heat resistant glue while Intel is just trying to wax over the truth and see where that takes Intel's Gross Margins over the next few years of nosedive!

August 10, 2018 | 11:31 AM - Posted by msroadkill612

What puzzles me is why business isn't as mad as hell with Intel for all the deliberate time wasting fragmentation practices listed here.

I would seethe with impotent rage at the cost and extra risks of staff being jerked around (on top of gouged prices), yet being compelled to buy from them.

I could live with a lot of warts rather than deal with such weasels.

Post new comment

The content of this field is kept private and will not be shown publicly.
  • Lines and paragraphs break automatically.
  • Allowed HTML tags: <a> <em> <strong> <cite> <code> <ul> <ol> <li> <dl> <dt> <dd> <blockquote><p><br>
  • Web page addresses and e-mail addresses turn into links automatically.

More information about formatting options

This question is for testing whether you are a human visitor and to prevent automated spam submissions.