11:44AM EDT - We're live here at Intel HQ in Santa Clara for the company's 2018 Datacenter Summit. The most lucrative part of Intel's business, the company is looking to pivot entirely from being PC-centric to what the company calls "Data-cetric", meaning an even greater focus on server (Xeon) processors, as well as things like non-volatile memory, compute accelerators, and FPGAs. This event, in turn, serves two purposes for Intel: reassure investors that Intel is on the right track for continued growth, and to outline the technologies and products that will get them there.

11:44AM EDT - We're sat about 15 minutes early

11:44AM EDT - Already seen some interesting things

11:45AM EDT - Looks like Raja Koduri and Jim Keller are here

11:45AM EDT - I saw their name badges

11:45AM EDT - but

11:46AM EDT - That's probably backwards and upside down

11:46AM EDT - But I'm sat behind him

11:46AM EDT - There's an Intel rep near to me who says there is a Cascade Lake system on stage

11:46AM EDT - The one on the left, apparently

11:47AM EDT - Here's a run down of what's going on today

11:48AM EDT - I'll be live blogging the morning schedule for sure. So sit tight, we should start here in about 10-15 minutes

11:50AM EDT - For those that are interested, today's Live Blog will be done using the Lenovo Ideapad 330

11:50AM EDT - The one with Intel's 10nm Cannon Lake chip

11:53AM EDT - I have no worries that the laptop will be OK for live blogging

11:55AM EDT - It's the battery life more than anything. There's a discrete GPU in here as well

11:55AM EDT - Last time I got the best part of 3 hours, just writing notes

11:56AM EDT - So we're in battery saver mode and low brightness. Fingers crossed

11:58AM EDT - OK, Koduri has been spotted.

12:00PM EDT - And Keller

12:00PM EDT - Looks like we're going to start in a few minutes

12:01PM EDT - Lisa Spelman to the stage

12:02PM EDT - VP of DCG, GM of Xeon and Datacenter marketing

12:02PM EDT - Just going over the goal of today is

12:02PM EDT - A chance to show how Intel is servicing is customers

12:03PM EDT - Going from the core datacenter out to the cloud and further to the edge

12:03PM EDT - Silicon foundation covering the portfolio will be discussed

12:03PM EDT - Nevin Shenoy to the stage

12:04PM EDT - Intel has long been the foundation of the world's major announcements

12:04PM EDT - Today will be laying out the plans for the strategy for the future

12:04PM EDT - Covering a broad set of products: Xeon, FPGA, ASIC,Connectivity, Optane

12:04PM EDT - Updates on roadmaps

12:05PM EDT - On how Intel will continue to win

12:05PM EDT - Data defines the future of the industry

12:05PM EDT - Today, half of the Intel revenue is data centric

12:05PM EDT - (basically non-consumer PC)

12:05PM EDT - Data is the new oil, the new power

12:06PM EDT - 90% of the world's data has been created in the last 2 years

12:06PM EDT - Only about 1% is being used to create meaningful business value

12:06PM EDT - Driving down the cost of computing, and driving the value of data

12:07PM EDT - 41x perf increase since 2006

12:07PM EDT - The greatest data collector example: the car

12:07PM EDT - Solving problems with technology

12:07PM EDT - Intel's first autonomous test fleet in Santa Clara and in Israel

12:08PM EDT - It's an end-to-end compute problem that exercises all the assets of intel

12:08PM EDT - Vehicles of the future run on data as much as gasoline today

12:09PM EDT - (Now he's describing how autonomous vehicles work)

12:09PM EDT - 4TB of data per hour generated in a car

12:09PM EDT - it's being send to the cloud which is training the models

12:09PM EDT - Once the models are then deployed, then blending the intelligence with the cloud

12:09PM EDT - The car will need 10x intelligence compared to today

12:10PM EDT - By the end of 2019 (?), Intel and Mobileye will have 2m smart cars on the road

12:10PM EDT - The car can only carry 30-40km of data at once

12:10PM EDT - So the need for continuous updates

12:10PM EDT - It will take an industrial approach

12:11PM EDT - Cloud, edge, networking

12:11PM EDT - It requires tight integration of all the resources

12:11PM EDT - Edge, Vision, AI, Cloud

12:11PM EDT - Intel believes that it sees a big and expanding total available market

12:12PM EDT - Intel is set to address $160B TAM in 2021

12:12PM EDT - That has now increased to $200B

12:12PM EDT - 9% CAGR 2017-2022

12:12PM EDT - Biggest opportunity in the company history

12:12PM EDT - Only have 20% of the market today

12:12PM EDT - Set to grow market share

12:13PM EDT - Time to talk about 3 of the growth drivers - cloud, netowkr and AI

12:13PM EDT - Easy on-demand access to resources

12:13PM EDT - 2/3 of cloud is TAM expansion

12:13PM EDT - Increasing need for custom CPUs

12:14PM EDT - Intel Cloud SP CPU Volume has increased from 18% to 50% from 2013 to 2017

12:14PM EDT - New use cases like AI are coming to the public cloud

12:14PM EDT - Intel views that as expansive

12:14PM EDT - Diversity of needs of companies has led to a deeper intimate relationship with customers

12:15PM EDT - Further customized products for specific customers

12:15PM EDT - Highly customized products for their needs

12:15PM EDT - All about strategy and growth

12:15PM EDT - Now networking

12:15PM EDT - $24B TAM for 2022

12:15PM EDT - Intel has 20% of the market today

12:16PM EDT - Moving off of proprietary equipment to dedicated servers

12:16PM EDT - Looking to benefit from the same economics

12:16PM EDT - Now applying through the core, to the edge, to the access of the network

12:16PM EDT - Requires top to bottom approach

12:16PM EDT - Low power through high perf

12:16PM EDT - Instruction set optimizations

12:16PM EDT - Software and ecosystem

12:16PM EDT - One example on stage

12:16PM EDT - 5G baseband reference design for edge compute

12:17PM EDT - Optimized for space and cooling

12:17PM EDT - Reference design includes Xeon-D

12:17PM EDT - Intel NIC

12:17PM EDT - Intel SSD

12:17PM EDT - As the world shifts from 4G to 5G

12:17PM EDT - There's 1.5m base stations sold every year - we see the market getting more compute intense

12:17PM EDT - (there was a small air cooler on that system)

12:18PM EDT - Customers are responding

12:18PM EDT - Also AI

12:18PM EDT - 30% CAGR expected

12:18PM EDT - $2.5B today, $10B in 2022 (TAM)

12:18PM EDT - Building purpose built products for AI

12:18PM EDT - Building AI into a lot of the portfolio

12:19PM EDT - New era of data center technology

12:19PM EDT - Need to move data faster, need to store more, need to process everything

12:19PM EDT - Workload accelerated products

12:19PM EDT - Starting with moving data

12:20PM EDT - Data traffic within data center, moving data rack to rack is growing

12:20PM EDT - Connectivity is increasingly goriwng as a bottleneck

12:20PM EDT - Intel is growing R&D

12:20PM EDT - Omnipath is succesful, 10G eth is #1 in the market

12:20PM EDT - Moving into SmartNICs

12:21PM EDT - Driving infrastructure offload

12:21PM EDT - Many cloud customers want to drive offload

12:21PM EDT - Combining assets FPGA and Ethernet portfolios

12:21PM EDT - The biggest opportunity comes in optical

12:21PM EDT - Coming 2019: Cascade Glacier SmartNIC

12:21PM EDT - Moving data around rack-to-rack with optical

12:22PM EDT - Silicon Photonics is looking at lowest cost per bit, lowest power per bit

12:22PM EDT - Plugging at it for 18 months+

12:22PM EDT - Working with top cloud providers and OEMs

12:22PM EDT - Now storage

12:22PM EDT - Re-architecting the memory/storage hierarchy

12:23PM EDT - Hot tier (DRAM), warm tier (SSD), cold tier (HDD/Tape)

12:23PM EDT - Insert Optane DIMMs and Optane Storage

12:24PM EDT - and 3D NAND

12:24PM EDT - Annnounced QLC NAND

12:24PM EDT - Introduced Optane SSDs a few quarters ago

12:24PM EDT - 40x lower latency, higher endurance than a NAND SSD

12:24PM EDT - At the top is Optane DIMMs, coming with Cascade Lake

12:25PM EDT - Enabling new workloads that were not possible before

12:25PM EDT - Solving bottlenecks in the DRAM tier

12:25PM EDT - Opportunities are huge

12:25PM EDT - $10B market

12:25PM EDT - Intel is uniquely positioned to deliver

12:25PM EDT - Research for over a decade

12:26PM EDT - Revamp the IMC on the CPU

12:26PM EDT - develop the software and ecosystm

12:26PM EDT - Performance is 8x in certain use cases, such as SparkSQL compared to DRAM only systems

12:26PM EDT - Enable 9x read transactions, 11x users per systems

12:26PM EDT - Delivering the promises of the hardware

12:26PM EDT - Such as persistence

12:27PM EDT - Enables high-availability systems

12:27PM EDT - From 3x9 to 5x9 availability

12:27PM EDT - Intel is super excited

12:27PM EDT - Bart Sano,VP of Platforms from Google Cloud, to the stage

12:28PM EDT - Multi-year partnership

12:28PM EDT - Help customers adopt new technology faster

12:28PM EDT - Google was first to deploy Skylake

12:28PM EDT - Working hard with Optane deployment

12:28PM EDT - Engineering engagement with Google

12:28PM EDT - Early adoption of Skylake was important to Google

12:28PM EDT - Demonstrates how you can transform the cloud

12:29PM EDT - Working with Intel directly allowed Google to deploy at scale quicker than usual hardware cycles

12:29PM EDT - Using AVX for 20-40% or even 100% perf improvements

12:29PM EDT - millions of hours of product use time

12:30PM EDT - (Raja just took a photo of me)

12:30PM EDT - Not only talking about innovations

12:30PM EDT - Going for very demanding workloads

12:30PM EDT - Bringing it out as soon as possible

12:31PM EDT - Partnership with Intel and SAP HANA running on GCP VMs powered by Optane Persistent Memory using Cascade Lake

12:31PM EDT - D in DRAM is a problem - need persistence

12:31PM EDT - Reducing downtime

12:32PM EDT - The larger the data sets, need to scale

12:32PM EDT - A lot of pull and a lot of interest

12:32PM EDT - Google is awaiting the first Optane Persistence memory - the first retail units just shipped for Google

12:33PM EDT - First production module of Optane DIMMs

12:33PM EDT - (Despite the announcement a couple of months ago... ?!)

12:33PM EDT - Broad industry support

12:33PM EDT - ISVs are getting ready for Optane DIMMs

12:34PM EDT - A lot of engineers working 365 days a year for this

12:34PM EDT - Only scratching the surface

12:34PM EDT - Now for processing everything

12:34PM EDT - Investing for decades on processors

12:35PM EDT - 20th anniversary of Xeon

12:35PM EDT - Consistently delivered advancements

12:35PM EDT - Shipped over 220m Xeons

12:35PM EDT - over $130B of revenue

12:35PM EDT - Intel has been *so* successful in the data center

12:35PM EDT - Xeon is tailored to the workloads that our customers care about

12:35PM EDT - Support for scale up workloads

12:36PM EDT - Lots of new technologies

12:36PM EDT - Recently, TSX, QuickAssist, AVX-512

12:36PM EDT - Have confidence we can tailor Xeon for a broad set of workloads

12:36PM EDT - Confidence in the ability to deliver leadership for Xeon

12:37PM EDT - Google was an early Xeon-SP sampling plan

12:37PM EDT - Fastest ramp to 1M

12:37PM EDT - Q2, Intel shipped 2M units

12:37PM EDT - Rate of adoption has been accelerating

12:38PM EDT - Leadership performance vs x86 competitors

12:38PM EDT - up to 1.48x per core, 1.72x for L3 packet forwarding, 3.2x Linpack, 1.85x database, 1.45x memory caching

12:38PM EDT - Ultimate flexibility

12:38PM EDT - 8+ socket support

12:38PM EDT - 60 SKUs (official, numerous off-roadmap)

12:38PM EDT - Reinventing Xeon for AI

12:39PM EDT - From Haswell to Skylake, improved inference by 277x, training by 240x times

12:39PM EDT - From AVX-512 and software optimizations

12:39PM EDT - 5.4x since July 2017, launch of Xeon-SP, inference perf

12:40PM EDT - well received by customers

12:40PM EDT - Need a platform with ultimate flexibility

12:41PM EDT - Step function increase in performance has a significant impact in revenue

12:41PM EDT - Generated 1B of revenue for Xeon in 2017

12:41PM EDT - Now Cascade

12:42PM EDT - Adding new instrucntions for AI, AI0Boost

12:42PM EDT - AI-Boost for INT8

12:42PM EDT - So this is AVX512-VNNI

12:43PM EDT - (My LG phone just FREAKED out when I needed to take a VITAL photo. I want to throw it across the room)

12:43PM EDT - Now for a Cascade Lake demo

12:43PM EDT - Comparing Skylake-SP vs Cascade with VNNI

12:44PM EDT - 11x speed up

12:44PM EDT - >VNNI was Knights Mill related

12:44PM EDT - Intel Select Solution

12:45PM EDT - Reducing time to deploy

12:45PM EDT - Program to verify configurations with workload optimized performance

12:45PM EDT - Making it easier to deploy

12:45PM EDT - Worked with 30 industry leaders

12:45PM EDT - Announcing an expansion of the portfolio

12:46PM EDT - Select Solutions for AI, Blockchain, and SAP HANA

12:47PM EDT - The Intel differentiation, using all the assets of the company year after year after year

12:48PM EDT - Jim Keller to the stage

12:49PM EDT - Going through why Jim joined Intel

12:49PM EDT - the main thing is scale

12:49PM EDT - Met so many engineers, good enthusiastic engineers

12:49PM EDT - Working with people who wants to change the world faster

12:49PM EDT - Diversity of technology of Intel is exciting

12:50PM EDT - CPU, GPU, AI, FPGA, Software, Storage, IO

12:50PM EDT - Tech is an endless train

12:50PM EDT - Great new products on 14 and 10

12:50PM EDT - Getting up to speed with the fab guys on 10

12:50PM EDT - Technology is good, but getting it out is hard

12:50PM EDT - focused on performance and yield, getting very hands on

12:50PM EDT - Bringing an outside perspective

12:51PM EDT - Each company has a unique culture

12:51PM EDT - Jim is a culture nerd

12:51PM EDT - Understanding how Intel works and being able to contribute to it is exciting

12:51PM EDT - Getting up to speed is a monumental task, then contributing to it

12:52PM EDT - Another Ray Kurzweld reference

12:53PM EDT - The half-a-watt to mega-watt problem

12:53PM EDT - Putting it all together in a plan is exciting

12:54PM EDT - Roadmap

12:54PM EDT - Cooper Lake 14nm

12:54PM EDT - Ice lake 10nm in 2020

12:54PM EDT - Cooper Lake in 2019 with BFLOAT16

12:54PM EDT - Cascade Lakeshipping in Q4

12:55PM EDT - Push on all fronts

12:55PM EDT - Including softwware

12:55PM EDT - Security Mitigations in Cascade

12:55PM EDT - Expect Intel to build on the heritage

12:56PM EDT - Through 2019 to 2020

12:56PM EDT - Maintain leadership position

12:56PM EDT - $200B TAM in 2020 is largest in Intel's history

12:56PM EDT - Unparalleled range of assets

12:57PM EDT - Intel is hungry to go after the market

12:57PM EDT - Now Navin Rao to the stage to talk AI

12:58PM EDT - evolving strategy

12:58PM EDT - (Keller and Raja just left the room)

12:58PM EDT - Making AI more applicable

12:58PM EDT - AI has been in the hands of the elite few so far

12:59PM EDT - Today has a strict line between training and inference, future might be more blurred

12:59PM EDT - AI is THE emerging critical workload

01:00PM EDT - 30% CAGR 2017-2022

01:00PM EDT - Computing is in this new architectural phase

01:00PM EDT - The core of computing is changing, Intel is well positioned to lead

01:00PM EDT - AI is evolving

01:00PM EDT - Proof of concepts are now unlocking real value

01:01PM EDT - (for anyone taking notes, my 10nm Cannon Lake laptop is now at 47% battery after one hour live blogging)

01:01PM EDT - Drug discovery, conversational translation

01:02PM EDT - AI is becoming a general purpose paradigm

01:02PM EDT - unlocking the ability to find value in data

01:02PM EDT - AI is expanding

01:03PM EDT - Developing hardware with Intel Compute inside to deploy from datacenter and edge

01:03PM EDT - Will expand as IoT grows

01:03PM EDT - Combination of dense compute and general purpose compute

01:03PM EDT - Managing complexity at scale

01:03PM EDT - Comprehensive AI portfolio

01:03PM EDT - Moving AI where it is needed

01:03PM EDT - One size does not fit all

01:04PM EDT - Expanding from mW to kW

01:04PM EDT - Optimization occurs across the spectrum

01:04PM EDT - Investing heavily in software that binds them

01:04PM EDT - Where the compute resides is down to where the data resides

01:04PM EDT - Need flexibility and a tool chain that works

01:05PM EDT - Like to think about this as 'Xeon Plus'

01:05PM EDT - Embracing the heterogenous world

01:06PM EDT - Partners winning with Intel AI

01:07PM EDT - Talking about AI dev lifecycle#

01:08PM EDT - Only 30% is training

01:09PM EDT - Intel works with customers across the entire process flow

01:09PM EDT - Building tools to make life easier

01:10PM EDT - Once AI goes into general purpose computing, we have achieved scale

01:10PM EDT - The goal is to evolve the platform

01:11PM EDT - Added instructions to help

01:11PM EDT - 64-bit is the defacto standard for HPC

01:12PM EDT - AI is an interesting as it is tolerant of low precision

01:12PM EDT - Invested a lot of time and energy making low precision models work on Xeon Scalable

01:13PM EDT - Intel Nervana NNP L-1000

01:13PM EDT - Commercial availability coming in 2019

01:13PM EDT - bfloat16

01:13PM EDT - changes what you can compute as it removes bottlenecks

01:14PM EDT - Based on Intel measurements on limited distribution SDV (Lake Crest) compared to Intel measurements on NNP-100 simulated product

01:14PM EDT - At scale you care about efficiency and 5% performance increases

01:15PM EDT - Software is essential

01:16PM EDT - Aim to enable everyone who wants to use AI

01:16PM EDT - going from low level to high level

01:16PM EDT - continual abstraction

01:16PM EDT - Data scientists work at the library level

01:18PM EDT - nGraph

01:19PM EDT - See the benefits from the low level to developers that are working on the topology regardless of hardware

01:19PM EDT - Defining APIs for inference as well

01:21PM EDT - Larger images such as drug discovery requires CPUs

01:22PM EDT - Using TensorFlow, optimized libraries, Omni-Path

01:23PM EDT - Reducing TCO

01:24PM EDT - AI Builder's Program

01:24PM EDT - Making it easier for customers to access and leverage the technology

01:25PM EDT - Hardware doesn't matter without software

01:25PM EDT - engaging with open source

01:25PM EDT - AI Academy and AI Dev Cloud

01:26PM EDT - 110K developers, 150k users each month

01:26PM EDT - Summarizing Xeon for AI now

01:27PM EDT - Driving the next phase of AI

01:27PM EDT - Example: Using ML for image rendering

01:28PM EDT - Bringing realism into rendering

01:29PM EDT - Next up is Cloud

01:30PM EDT - Raejeanne Skillern

01:30PM EDT - VP DCP and GM Cloud Service Provider Platform Group

01:31PM EDT - Worked hard over the last decade to learn about our future strategy to service the growth

01:31PM EDT - One size does not fit all - customization is critical

01:31PM EDT - It's not all about perfomrance and efficiency - it's about product differentiation

01:31PM EDT - Need to support the infrastructure at scale

01:32PM EDT - (24% battery left)

01:32PM EDT - Cloud is everywhere, but we're at the beginning at examining the possibilities

01:32PM EDT - Delivering extreme performance at scale efficiently

01:32PM EDT - Driving new service opportunities

01:32PM EDT - Cloud is the foundation of the digital transformation

01:33PM EDT - None of it would be possible without power perf eff of the cloud today

01:33PM EDT - Growing in company revenue percentage

01:33PM EDT - Will this growth continue

01:33PM EDT - Will be a high gross segment

01:33PM EDT - This growth is 2/3 TAM expansive

01:34PM EDT - Coming from next gen cloud service providers up to the large cloud

01:34PM EDT - There is a wave of digital services driving the industry

01:34PM EDT - 70% of main cloud comes from Digital Services such as eCommerce

01:35PM EDT - Driving by content on digital creation and media

01:37PM EDT - Intel is out accelerating demand for cloud services

01:38PM EDT - Intel offers the ability to customize SKUs

01:38PM EDT - First optimized SKU for google in 2008

01:38PM EDT - Xeon-D was built out of joint innovation with facebook

01:38PM EDT - Nearly 30 SKUs are custom off-roadmap uniquely for the service provider

01:39PM EDT - They want to keep it off roadmap

01:39PM EDT - Amazon is a big customer

01:39PM EDT - z1d instance that offfers sustained all-core turbo of 4 GHz

01:40PM EDT - (So, custom = custom turbo? so binning?)

01:40PM EDT - Also semi-custom by integrating third-party IP

01:40PM EDT - using packaging technologies

01:40PM EDT - Ability to scale up and scale down

01:40PM EDT - If they don't want to pay for certain things, can be taken out

01:40PM EDT - Can also make custom ASICs for customers

01:41PM EDT - Also connectivity can be optimized

01:42PM EDT - Using the wide Intel portfolio and system level design

01:43PM EDT - Intel talks a lot about software for a hardware company

01:43PM EDT - It's the key to creating new services

01:44PM EDT - 200 engineers are engaged with cloud service providers for hands-on side-by-side engineering

01:44PM EDT - Already working on 150 cloud service projects this year already

01:45PM EDT - Oath (merger of Verizon, Yahoo, AOL) has custom CPU

01:47PM EDT - workload optimization

01:47PM EDT - (12% battery left)

01:48PM EDT - Wechat is 82% of Tencent's ecommerce revenue

01:48PM EDT - So AI perf is critical

01:48PM EDT - Optimized inference perf by 16x

01:48PM EDT - such as voice-to-text translation

01:50PM EDT - Intel working with Pixar to update the code base to improve performance

01:50PM EDT - 2-4x perf

01:51PM EDT - SGX on Azure for security

01:52PM EDT - Market Acceleration as well

01:53PM EDT - Cloud Insider program

01:53PM EDT - 650+ members in the program

01:54PM EDT - Creating the pipeline for future services

01:55PM EDT - Using sandbox development to get access to new hardware

01:55PM EDT - Customers can reach Intel engineers

01:55PM EDT - (moving into super low power mode just put some of my tabs in super unresponsive mode)

01:56PM EDT - 30% revenue uplift from engagements

01:56PM EDT - Selling up the stack, better TCO

01:57PM EDT - (I'm in 400 MHz mode. huh)

01:57PM EDT - Summary

01:58PM EDT - Broad set of silicon optimization

01:58PM EDT - Offering differentiation

01:59PM EDT - Robust marketing and sales programs and services

01:59PM EDT - OK, 15 minute break time

01:59PM EDT - Be right back

02:19PM EDT - OK here we go. Using my HP Spectre x360 Core i7 7th gen backup

02:19PM EDT - Now up is Raj Hazra talking about the future of enterprise

02:19PM EDT - CIOs have been stalling about spending on infrastructure

02:20PM EDT - Potential thought and question about if this was systemic

02:20PM EDT - Geopolitical effects in 2015 caused macroeconomic uncertainty in demand

02:21PM EDT - It confused many in the industry about long-term use about clouds

02:21PM EDT - Would everyone move to public cloud?

02:21PM EDT - Would corporate datacenters be obsolete?

02:21PM EDT - So a stall caused a decline in a trend of increased spends

02:22PM EDT - Actually in the middle of a transformation in serving customers

02:22PM EDT - New ways to do things compared to the past

02:22PM EDT - So Intel transformed to meet the customer expectations

02:22PM EDT - Conversation wasn't about product or TCO, it was about strategy

02:23PM EDT - Compute strategy, Cloud strategy, security strategy

02:23PM EDT - e.g. John Deere goes from automotive to AI-driven agricultural services

02:24PM EDT - IT no longer supports business, it ends up being the business

02:24PM EDT - So three aspects

02:24PM EDT - Legacy Infrastructure is now aging faster

02:24PM EDT - The penalty of not going to latest technology was high

02:25PM EDT - Two: Enterprise would be using private clouds alongside public clouds

02:25PM EDT - Third: AI would drive phenomenal growth

02:25PM EDT - With this as the conviction, Intel invested in strategies

02:27PM EDT - All about easing adoption of private and hybrid cloud

02:27PM EDT - Expanding analytics and growing AI to a broader portion of enterprise use cases

02:28PM EDT - Accelerating time to value through Intel Select Solutions

02:28PM EDT - Make it deployable and consumable in a short amount of time

02:29PM EDT - Working on reference architectures over decades

02:29PM EDT - Make application solutions for end-consumers happen

02:29PM EDT - Intel's conviction is about the transformation and it defined the investment priorities

02:30PM EDT - Doubling of the private cloud adoption on premises since the new strategy

02:30PM EDT - Working on workload mobility across platforms from public/private

02:32PM EDT - CPU deployment running AI on premesis for private clouds has doubled

02:32PM EDT - CAGR is growing

02:33PM EDT - Bringing value from CPU generations as well as the entire goal to set up an infrastructure that can be systematically upgraded and expanded as the workloads expload

02:34PM EDT - Xeon is the biggest revenue generator in the enterprise

02:34PM EDT - Skylake has the fastest ramp since Skylake

02:35PM EDT - Highest mix to the top-end SKUs

02:36PM EDT - Skylake shipments is now more than 50%

02:36PM EDT - (But there's still some Haswell in those shipments)

02:36PM EDT - Higher mix of high-end SKUs

02:36PM EDT - (High-end SKUs, or higher ASPs due to the higher pricing for Xeon-SP vs previous platforms for tier 1 customers?)

02:36PM EDT - It's not just about the processor

02:37PM EDT - Ethernet, Omnipath, Silicon Photonics, FPGAs, SSDs

02:37PM EDT - Offering a level of latency reduction along with capacity per dollar

02:37PM EDT - Unique value delivered to enterprise customers

02:38PM EDT - Also Optane DIMMs

02:39PM EDT - A huge part of SAP's business in having a database up and running for analytics - with growing of dataset sizes means that loading datasets is a key bottleneck

02:39PM EDT - Faster startup times by 12.5x

02:40PM EDT - >3TB memory per CPU

02:41PM EDT - Revolution of the golden age of data

02:41PM EDT - All about the key to delivering performance

02:41PM EDT - Does not happen without key design strategy

02:42PM EDT - This comes about re-architecting parts of the CPU and memory sub-system, and adjusting the software to understand the platform

02:42PM EDT - Accelerating Enterprise AI

02:42PM EDT - Enterprise has a workflow, constrained workflow, processes, validation

02:42PM EDT - but it has to do that over a consistent data architecture

02:43PM EDT - without disrupting operational flows

02:43PM EDT - For AI to be successful, you must run it on the infrastructre of the enterprise and drive value/TCO

02:43PM EDT - e,g. Union Pay using DNN for fraud detection

02:44PM EDT - They did try GPGPU

02:44PM EDT - But the ability to integrate was better with Xeon. First with Broadwell, now with Skylake

02:44PM EDT - Driving innovations hard into the ecosystem

02:45PM EDT - Worked with Private Cloud, Analytics, AI, and HPC

02:45PM EDT - This is the way business will be done and winners will play

02:46PM EDT - Enabling Exascale

02:46PM EDT - A supercomputer for 2021 using New CPU Microarchitecture (Ice Lake?), Advanced Interconnect, Novel Memory/Storage, High Performance Converged Software

02:47PM EDT - Trying to obsolete the term 'battery life'

02:47PM EDT - This is the scope of changes when you bring high performance and innovation to the market place

02:48PM EDT - Enabling 5x perf for 2x power

02:48PM EDT - DoE on this project: #'We are relying on Intel to come through as they have done in the past'

02:48PM EDT - Intel is used to solve business problems and solving humanity's problems

02:49PM EDT - Built and honed based on multiple decades of experience

02:49PM EDT - Infinitely growing TAM

02:49PM EDT - (wait what??)

02:49PM EDT - The future is bright, Intel is winning, Intel is investing aggressively and smartly

02:50PM EDT - Dan Rodriguez to the stage

02:50PM EDT - VP DCG and GM Communications infrastructure

02:51PM EDT - 'Journey to Network Transformation'

02:51PM EDT - Watching more video per day

02:51PM EDT - Data is exploding

02:51PM EDT - Talking 5G

02:51PM EDT - New use cases and experiences through 5G

02:52PM EDT - The network needs to become something different - it has to be smart in the cloud through general purpose hardware

02:52PM EDT - Intel is partnering to deliver this network transformation

02:52PM EDT - $24B Silicon TAM Opportunity

02:53PM EDT - A Content Delivery Network: Intel can touch every point of the network

02:54PM EDT - Intel's goal is move fixed function appliances for these networking functions onto general purpose hardware

02:54PM EDT - 1: Invest in architectures, IO, memory, FPGAs

02:54PM EDT - To ensure workloads can run on all major areas

02:54PM EDT - Isotropic networks can be deployed and scaled

02:55PM EDT - Partner with the industry

02:55PM EDT - with open source projects

02:55PM EDT - customers and ISVs

02:55PM EDT - Enabling cloud gaming with Kaby-G

02:55PM EDT - wait what

02:56PM EDT - All about growing share in networking

02:57PM EDT - Paying particular interest in comms service providers

02:57PM EDT - Showing CAGRs of driving revenue

02:57PM EDT - percentages of DCG revenue are increasing (that's a super odd metric btw)

02:57PM EDT - All about virtualization as well

02:58PM EDT - Virtualization expected to grow 16% CAGR to 2022

02:59PM EDT - Nearly all of the network virtualized deployments are on Intel

02:59PM EDT - Top to bottom scalability

02:59PM EDT - Perf, Power, Price

02:59PM EDT - Architectural consistency

02:59PM EDT - Reducing overall time to market

03:00PM EDT - 40x packet processing on Xeon compared to a decade ago

03:01PM EDT - Also built on platform technologies

03:01PM EDT - Ethernet, Optane, FPGA, Ethernet, QuickAssist

03:01PM EDT - Dynamic Device Personalization

03:02PM EDT - Silicon Photonics will become a big asset in the era of 5G

03:02PM EDT - Offering the right performance and scalability

03:02PM EDT - Use collaborations to enable transformation

03:03PM EDT - Invest in open source and standards

03:03PM EDT - Open Source since 2010

03:03PM EDT - for DPDK

03:03PM EDT - Collaboration with 300+ memobers, 82+ PoCs/Trias/Deployments

03:04PM EDT - Intel Select Solutions for NFVI (network functions virtualization infrastructure) and UCPE (universal customer premises equipment)

03:05PM EDT - Delivering managed services for enterprise

03:05PM EDT - Now talking about Edge computing

03:06PM EDT - Collaboration with Nokia

03:07PM EDT - Diverse use cases have diverse requirements

03:07PM EDT - low latency vs bandwidth

03:07PM EDT - security and data locality

03:08PM EDT - All about compute at the edge where required

03:08PM EDT - e.g. smart retail store with smart IoT edge node to make decisions locally

03:09PM EDT - Growth opportunities

03:09PM EDT - Transforming central offices at the edge

03:09PM EDT - Businesses with outposts etc

03:10PM EDT - Next Generation Central Office

03:10PM EDT - transforming legacy equipment

03:10PM EDT - A local edge data center

03:11PM EDT - Reducing time to market and R&D

03:13PM EDT - Video time

03:15PM EDT - 2G was about voice, 3G was about data, 4G was about more, 5G is about data services

03:15PM EDT - We believe that network transformation is required to deliver the 5G experience

03:15PM EDT - Intel is uniquely positioned

03:15PM EDT - Driving the enterprise and cloud for years

03:15PM EDT - We have driven with partners 25+ 5G trials

03:16PM EDT - Driven by momentum

03:16PM EDT - tracking to the leading position in 5G infrastructure

03:17PM EDT - Real time video chatting

03:17PM EDT - Reducing cost of backhaul

03:18PM EDT - Winter Olympics trial of 5G

03:18PM EDT - 22 5G links at 10 sites

03:19PM EDT - Intel's strategy has driven revenue

03:19PM EDT - Underlying network transformation strategy

03:19PM EDT - OK time for some Q&A

03:20PM EDT - Navin on the stage

03:20PM EDT - Q: Lot of concern about lithography. Talk about what customers reaction about 14nm / 10nm and competitive intensity

03:22PM EDT - A: I don't talk to customers about nanometers. They care about delivered performance. The need to optimize comes from various factors. Software investment, broadening the portfolio, microarchitectural optimization. Been working through roadmaps with customers: delivering improvements through hardware or software or optimized strategy. For competitive intensity, we expected 2018 to be competitive. Intel has done a very admirable job and outperformed expectations. That's a function of the TAM environment being good, the product put forward and scope, and function of expantion into new domains like AI. We're not scared by competition. Out roadmap gives us confidence.

03:24PM EDT - Q: 10nm and microarch. Ice Lake will be new microarch - along with node shrink (tick+tock), can you explain this different cadence. Can you discuss client vs server changes. What are the perf gains for most workloads between now and 10nm with the two microarch (Cooper and Ice)

03:26PM EDT - A: Over last generations, we've been pushing microarch improvements within a node and from node to node. We added microarch to broadwell even though it was first 14nm chip. We can drive microarch on new nodes. We haven't got into detail about Ice Lake, but will talk about it when it is closer. It's a great product with improvements. In our 14nm, Skylake was a big bump. We called it the biggest improvement in a decade. Our follow on, Cascade, adds new capabilities. They do not always have to be at the microarch level, such as mem controllers for Optane DIMMs. We don't fully know how our partners will innovate over Optane DIMMs. It's a once in a decade change. We expect it to be broadly used as people learn. About client/server - there used be be 18 months lag between client and server nodes. That gap in time is going to reduce. We discussed about client 10nm on shelf in 2019

03:27PM EDT - Q: In the last 5 years you had competition. But AMD did have 20% market share. What's different this time vs now? In the diff segments, where is the most competitive intensity

03:29PM EDT - A: Today, the TAM we're going after is much larger. we see the opportunity more about just microprocessors. It's important to recognize our investments in our portfolio about addressing customers about more than just microprocessors. Optane, FPGA, ASICs etc. It's a breadth of a portfolio to deliver higher levels of perf is a big time difference between 5 years ago and now. Competitive intensity, we are already paranoid about anybody out in the market at all times. We don't think intensity increases, we're always wary about what's in the market. We look about markets and competitors in each market.

03:30PM EDT - Q: Help understand about Optane DIMMs becomes material and revenue. Not a lot of discussion about FPGAs - can you talk about competition there

03:32PM EDT - A: We'll discuss FPGAs in the afternoon. For optane, the TAM is huge. The SSD storage of optane is reported in our DSG segment. That business is ramping and will drive superior gross margins over NAND. The DIMMs business is new, it's a $10B market as a subset of DRAM market - we'll talk about it more as we ramp. We have $0 now as it shipped yesterday, we have a high degree of interest from customers to make it a large scale business.

03:33PM EDT - A: About FPGAs. It's not jsut an NVIDIA story. They're used for a lot of difference places. It's about inference more than training. We'll talk about low latency, efficiency, and ability to react to new topologies. We'll discuss MS Brainwave on Intel FPGAs. Also infrastructure about datacenter. We've grown a lot on the DC already and we expect it to continue. It's not just head-to-head with NVIDIA. FPGAs to lookaside and infrasturcture - GPUs only do look-a-side.

03:34PM EDT - So that looks like a wrap for the morning. Hope you enjoyed it. I need coffee now.

Comments Locked


View All Comments

  • CajunArson - Wednesday, August 8, 2018 - link

    There's nothing Epyc about this dog & pony show at all.

    The only thing that people care about is COARZ Intel. And AMD has 64 of them!
  • psychobriggsy - Wednesday, August 8, 2018 - link

    Come on! You absolutely need AVX-512-VNNI for your everyday uses!

    As Charlie pointed out in his article, Intel will try to dazzle with highly specific and targeted benchmark results today.

    That slide though - Icelake is now Q4 2020 (assuming a 1 year product cadence).
  • CajunArson - Wednesday, August 8, 2018 - link

    Nobody needs "VNNI" because Artificial Intelligence is a dead-end market that nobody cares about.

    Now back here in the real world, I know for a fact that my Grandma needs AT LEAST a 32 Core Threadripper2 to check her AOL email. Let's get real here.
  • yvizel - Wednesday, August 8, 2018 - link

    My grandpa needs 64 cores (+ HT!) when posting on Facebook!
  • NikosD - Friday, August 10, 2018 - link

    AI is for GPUs.
    Everything else is Intel's propaganda.
  • Marlin1975 - Wednesday, August 8, 2018 - link

    So nothing about Spectre and Meltdown yet? Or is intel acting like that does not affect datacenters?
  • dolphin2x - Wednesday, August 8, 2018 - link

    Doubt they'd talk about vulns during a product-focused keynote.
  • Ian Cutress - Wednesday, August 8, 2018 - link

    Cascade with HW mitigations it sounds like
  • psychobriggsy - Wednesday, August 8, 2018 - link

    11:53AM EDT - I have no worries that the laptop will be OK for live blogging
    11:56AM EDT - So we're in battery saver mode and low brightness. Fingers crossed
    01:01PM EDT - (for anyone taking notes, my 10nm Cannon Lake laptop is now at 47% battery after one hour live blogging)

    50% down after 75 minutes when saving power doing a simple web use case when the system should be mostly idle?
  • Ian Cutress - Wednesday, August 8, 2018 - link

    System has RX 540, and is using the Wi-Fi module a fair bit. Display is a 13x7 TN panel with bad viewing angles. Probably very inefficient

Log in

Don't have an account? Sign up now