Intel Data-Centric Innovation Day: A Live View

Posted on April 2, 2019

9:35 AM: We’re getting a bit of a late start. Bob Swan, Intel CEO, is giving us an introduction before the main event starts.

He’s discussing the transformation of Intel from a PC-centric company to a Data-centric one, which explains the name of today’s event. Data-centric computing represents a $1.25T opportunity this year alone. It will only increase. Swan indicates that Intel continues to grow its market here and that its success is tied to that of its partners.

To get there, Intel will provide

Full portfolio of CPUs
FPGAs
Memory
and more

Intel is absolutely everywhere these days and they’re going to be in even more places tomorrow.

9:30 AM PT: Navin Shenoy, Executive Vice President/General Manager, Data Center Group, Intel

9:40 Shenoy is now on stage and indicates that this is the first full portfolio launch in the company’s history. He reiterated the message around innovation as Intel transformations from being a CPU provider to one that has CPUs, memory, storage, FPGAs, and more.

Over 50% of the world’s data was created in the past two years and this will accelerate. AI and ecommerce are drivers of this trend, but there’s a lot more happening as well. Even with all of this data, only 2% of the worlds data has been analyzed, so there’s massive opportunity.

Industry mega-trends:

Shift to the cloud
Growth of AI and analytics
Cloudification of the network and the edge

My opinion: It’s clear that we’re at an inflection point in computing as compute shifts from monolithic data centers to the cloud and to the edge.

Intel has expanded it’s data-centric infrastructure

Intel is introducing seven new products today, including CPUs, Optane memory, network adapters, and more.

Intel’s 2nd Generation Xeon Scalable Processors are seriously impressive. Here’s a screenshot of the high level stats. Scales to 56 cores or 112 cores in a dual socket system. That’s… a lot! 12 channels of native DDR4 memory.

Challenging the Xeon 9200. They are now going to try to set a world record on stage today. This should be interesting. They have a 2 x 56 core system on stage. They’re going to run a SPECrate 2017 floating point benchmark. The current record is 282 fp_rate. The Xeon 9200 achieved 522 fp_rate. It’s safe to say that this beats the old record.

And now we’re moving from the beast Xeon 9200 to the bread and button. These mainstream CPUs offer 1.33X average performance gain over the previous generation of CPUs.

They are also innovating in AI performance, which mains AI become even more mainstream. It will be a $10B market by 2022. The first gen Xeon scalable provided 1 Mbps of inference throughout and was 5.7 Mbps by December of 2018. With built-in inference acceleration, Intel now achieves 14x deep learning inference improvement from the gen 1 numbers.

Intel is now showing a video demo to show how DL Boost improves workload performance. And here’s the net of it.

Intel just brought out an executive from AWS to talk about their partnership.

And now we’re getting an overview of the cloudification of the network. AT&T is a key customer here and they’re well on track to achieving their goals using Intel.

1.5x+ performance over 1st gen (measured in mpps throughput) without new features. With SpeedSelect technology (new), this jumps to 1.7x+ performance. This technology locks some cores at a higher clock rate. New CPU generation is showing really great results.

New topic: Data has to be able to get to the CPU fast enough for the CPU to be able to maximize its efforts. Next on stage, we’re hearing from Twitter (Matt Singer) to talk about system-level optimization efforts. Twitter has to support over 1 trillion events per day. A single tweet will have a lot of individual events. There’s a huge Hadoop backend that supports the whole thing. Twitter worked with Intel to try to dial areas of performance challenges. The result – removing a storage IO bottleneck by adding Intel SSDs at key points – allowed Twitter to reduce racks of equipment and reduce the number of hard drives per cluster.

Moved from 4 core to 24 core Intel G2 Xeon processors. Between this and reduction in storage devices reduces energy cost by 75% and improves performance by 50%.

Next problem: DRAM doesn’t support application developer needs and isn’t persistent. Flash doesn’t provide the performance needed. The answer: Optane Data Center Persistent Memory.

Google Cloud (now a guest on stage) says that Optane is great for in-memory database workloads, which makes sense given the Optane form factor.

My opinion: Optane could be a game changer in memory and storage. However, I do wonder if Intel’s desire to hold the technology so close could work against them. NAND flash is a massive market. Optane is Intel-only. It’s the very definition of vendor lock-in. I don’t know yet if Intel is or will license 3D XPoint to other vendors or not, but it could accelerate adoption.

10:30 AM PT: Lisa Spelman, Vice President/General Manager, Intel Xeon Processors

10:38 AM – This session is now underway. Lisa is going to take us to the edge. Latency, bandwidth, security, connectivity are key edge drivers.

More processing is happening at the edge due to industrial IoT and other factors. By 2021, there will be 7 million service edges in play. CDNs are one of the biggest. Think Netflix, etc. Some estimates place these at 60% of Internet traffic.

Edge with DRAM alone vs. edge with Optane demo. The demo, as expected, shows more latency in the DRAM alone side. Comcast is seeing a reduced TCO and better performance using Optane.

Intel FPGAs

The speeds and feeds are coming fast and furious now!

Microsoft Azure GM now on stage. He says “customers still want hybrid” which is clear from all of ActualTech Media’s research as well. On-prem isn’t going anywhere anytime soon.

Windows Server 2008 and SQL Server 2008 – customers can move to Azure to get extended support or they can migrate. This is a brilliant move by Microsoft. “Want to keep 2008? Well, you have to move it to Azure if you want it supported.” For many, it will be less painful to lift and shift these workloads than it would be to upgrade the underlying OS and database.

AzureStack HCI is also being discussed. Learn more about that here: https://azure.microsoft.com/en-us/blog/announcing-azure-stack-hci-a-new-member-of-the-azure-stack-family/

With Intel: Double the IOPS of previous infrastructure. 13.7 million IOPS with 25% fewer servers. With SQL Server persistent memory, “much faster.” He didn’t provide specific numbers here. New announcement: Azure is delivering some of the largest SAP bare metal instances available. SAP with up to 24 TB of memory. They are also refreshing their virtual machine instances with the latest Intel technologies.

And now we’re on to security. Intel has a lot to talk about here. There are security mitigations in the new CPUs to enhance protection. They are offering encryption for data at rest, a trusted platform service, and even hardened full-stack solutions.

“Security is the number one CIO imperative.” There are layers of software controls from perimeter to server, but this puts CIOs and CISOs in a reactive posture. Intel is partnering with Lockheed Martin and HPE. Establishes a hardware root of trust up the full stack.

My opinion: This is interesting and could prevent a lot of attacks that are the result of unpatched operating systems and applications.

VMware is now on stage to talk about Intel and VMware storage. One of the engineering collaboration is the rethinking of storage and memory. They are seeing higher VM density and improved storage performance.

11:25 AM: And we’re wrapping up! Thank you for following along!