Intel Ice Lake and Hyperscalers - CTO Advisor Hybrid Infrastructure Phase 3
Transcript
>> Hey, it's Keith Townsend, Principal of the CTO Advisor. And if you've been following the series, we're in phase 3 of the CTO Advisor Hybrid Infrastructure. For those of you just joining the journey, we're on the journey to show the work behind building a hybrid infrastructure. And as that we have a special treat sponsored by Intel, is the VP of Hyperscale Strategy at Intel, Rebecca Weekly. Rebecca, welcome to the show. >> Rebecca: Thank you so much, Keith. >> So we both promise that we're not going to get too geeky and get too into the weeds.
But you guys had an exciting announcement a couple of weeks ago now. As the recording of this video, 3rd Generation Intel scalable processors, also known as Ice Lake. We've been waiting on these processes for quite some time. What are some of the most important takeaways technically from that launch? >> Well, there's a lot of exciting things with Ice Lake. So obviously, performance is critical and we are delivering 46% higher performance versus our previous generation processor. We've specifically really focused on acceleration for the kinds of workloads that matter at the network, at the edge, and certainly in cloud environments.
And part of that is absolutely AI. So you will see up to 74% higher performance for AI workloads on the Ice Lake processor. And that's critical because so often, AI recommendation systems, natural language processing, this is happening in the midst of a workload. So it's you've got to collect the data, you've got to process the data, then you infer something from the data. So that pipeline is where there's an opportunity for general purpose to really shine. So that is incredibly exciting for us.
Some of the other areas are really around security in the Ice Lake processor. So obviously, Intel SGX, this is its debut in the scalable processor line, which is really exciting. It's obviously been a part of both our client products and our products E3 products for awhile. But being in the SP line means you can have so much more performance, so much more memory in an enclave, and that is really exciting. And then the other area where we're most excited or I'm very excited, is in the crypto space.
So we've brought new crypto acceleration to our processor and it's going to help deliver innovations up and down the stack to make sure that you are getting the most widely deployed cryptographic algorithms running as fast as possible on Intel. >> So thank you for that recap of some of their improvements of this generation over previous generations. Obviously, pertinent to your role, how does this benefit the hyperscale? >> Oh, so much. So in hyperscale, when you're operating at hyperscale, it's all about availability, reliability, and the ability to highly leverage and utilize the processor.
So being able to take advantage of the fact that 93% of the world's data is running on Intel processors with consistency, with the ability to run anywhere, that is a huge aspect of delivering more performance in our latest generation, but consistency across your existing infrastructure, that is huge for cloud. You need to be able to deliver actually at scale all the time. I also think, again, I mentioned a little bit about the acceleration both for crypto as well as AI. There's so many recommendation systems that are fundamental to how the cloud works, whether it's your e-commerce feed and deciding what to serve to you when you search for a book for your six year old, or if it's something more along the lines of your social network and what kind of newsfeeds you might want to look at.
That is the entire process of a lot of our interaction with digital services online. And that is just inline crypto, or in the sense of authenticating an end user. And it is actually recommendation systems for everyday use that work really well on general purpose compute. So that is probably the two key areas that I have found from my day job to be most exciting with Ice Lake, where we're really making people... SSL is running so much faster. CDN workloads can run so much faster when you're using this kind of accelerated crypto and the ability to accelerate AI.
>> So the overall premise of the CTO Advisor Hybrid Infrastructure and the CTO Advisor advice to clients is that the hybrid infrastructure is going to be a huge part of the enterprise IT data center manager's job from day to day. We're going to manage on premises assets, off premises assets. And as we're looking at things such as Ice Lake, we're trying to figure out how does that play into our overall landscape? As background for you, in the CTO Advisor Hybrid Infrastructure, what we did was we partnered with Oracle Cloud Infrastructure, and went out and expanded our hybrid infrastructure via high-speed networking, to bring kind of that monitoring goodness inside the data center.
Oracle Cloud Infrastructure and all hyperscalers fall under your purview. How do you help customers like me who aren't quite ready to go into the enterprise data center upgrade phase, leverage some of the goodness vest in Ice Lake? >> Well, certainly we are working to make sure that we are in every single public cloud, but we also work very closely with all of our different primary ISVs across the ecosystem. So whether that's VMware, or SAP HANA, or any workload, Oracle Database, to make sure that it's taking advantage of the newest and greatest features in our Ice Lake, that definitely enables those workloads, whether they're running in an on-prem private cloud or in a hybrid methodology to a public cloud, can have that experience and that performance.
So we do that work directly with software partners, we do that work directly with hyperscale cloud providers, and we do that work throughout the entire spectrum of hybrid use cases. So whether that's OpenShift and some of the work that's happening in that domain space, or Cloud Paks, or exascale, or something within those, Arc, HCI, we do work across the ecosystems to make sure that the newest and greatest features in our processors are available in their deployments. So what I like about that is, of course we want to make sure you have that option and opportunity, but it also still runs consistently.
So you will get more performance if you're running on Ice Lake, but it'll still work if you're running on CLX, or Skylake, or any of our previous generations. And again as much as I love my new processor, when you are a CTO or running your company, you want to have your work consistently running. So there may be a large installed base that is not the latest and greatest yet, and that's very important to be able to leverage and have consistency across. So I very much look at this as we're all on a journey.
People who come to the public cloud often for disaster recovery services, business continuity planning, backup services. It's kind of how they start, is just to make sure that they have that capacity available when they need to. And that may not be the latest and greatest bleeding-edge processor of the day. And that's okay. What we work on as Intel is to make sure you have a reliable, secure, fabulous experience, regardless of which product you will land on. And then we always try and make sure we're building in the best and greatest workload behavior so that you want to use our newest processor, of course.
And I really think that the area where Ice Lake for end users in the hybrid cloud is going to be most exciting, is in that domain space of confidential computing. A lot of- >> So- >> Oh, sorry, please continue. >> Oh, I'm sorry, you were about to hit on a topic that I can- (laughter drowns out speaker) But before we get to confidential computing, I just want it to tell the audience we did exactly what Rebecca was talking about. We extended our VDI platform out of Oracle Cloud, leveraging VMware, so we could keep our operational strategy consistent.
We can leverage Optane, which we'll get into in a few seconds, and as well as new newer generation Intel processors compared to what we have the data center. One of the big changes in Intel processors over the years, has been the focus on crypto, and specifically on secure enclaves in the form of SGX. There's one API move where from a development and data process, we can secure data within a system. However, when I'm reading about some of the latest, I don't want to call them hacks, they're more security breaches that are kind of user error, we have a hard time with securing S3 buckets, how does Intel help surface off the goodness that's in crypto and SGX to the end-user via the hyperscalers?
>> Absolutely. It's a great question. We do a lot of work in the Linux community. It starts at the kernel. I mean, we have to make sure that things like SGX get upstreamed into Linux so that they can be exposed at a higher application level. SGX does require application level changes. It's not going to be transparent to an end user that you can just have your S3 bit bucket work and you can pull through it as a secure enclave without doing that work.
So I think we still have work to do. I know I want to overstate. But we are working actively on it. So we're partnering with VMware. Again, we're partnering with Oracle to bring those into capsulated data sets into an SGX mindset and make it easy for application users to choose security. I really believe the trade-off between performance and security has been so significant for users for so long. We've kind of trained people, well, you can go to the cloud but it's not going to be secure.
That's your choice, right? And so we use virtual private networks and we use other methodologies for encapsulation to try and isolate end users or isolate data, not really security, not really trusted foundational security. What SGX really will enable us to do now is to make confidential computing an easy choice, a ubiquitous choice if you want to. Again, it's going to take some changes to your code. We've done some great work with Graffen, with Fortinet. There's a huge number of ISVs out there who are already offering some form of service and cloud, right?
We have great work with Azure offering confidential computing. IBM is offering confidential computing. Ali Cloud if you're in China and across the world is offering confidential computing with SGX already, and more are flooded to come out over the next few months. So for me, what I see is we put out a technology and the world of software is starting to enable it. It's upstreamed in Linux. Now we go through the different application stacks. We help people choose secure VM and have the ability to encapsulate their data at scale with performance.
So we're just on the journey. >> So let's get into this move store process story a little bit. I think one of the core things I heard at the Intel Ice Lake launch was this is not just a processor play. You've hinted to that a little bit around the AI workloads, and the ability to work with ISVs, and attack this from a Linux kernel, and lower level application stack. But a big part of that is just storing the bits, whether that's bits in DRAM or that's bits on disk, and moving it from site to site.
Help me understand something that's been in the news quite frequent, which is Optane, not just the SSDs, but also the PMem. Where does Optane play into this overall strategy, especially when it comes to the hyperscalers? >> Well, Optane PMems specifically, Persistent Memory, really is an opportunity to expand the amount of memory in a system. And fundamentally, why does that matter? Well, when you think about database workloads, it's a perfect example, the more you have to shard that database across a network, the more work is involved in actually being able to access your data and serve your end users.
So the more you can bring it close, the more you can allow yourself to have those faster in-memory database accesses, the more, whatever your revenue generating metric is, can occur. So that might be more users served, more search queries returned, more whatever your KPI, known performance indicator might be. So that to me is a massive opportunity from a memory perspective. Those four terabyte, six terabytes, eight terabyte, massive scale up systems. You can, in a two-socket system, have more memory capacity than you ever had with a four-socket system before.
And when we talk about reliability at scale, that is an opportunity. We really want to be able to deliver massive capacity, ideally with, I hate to say this 'cause there's a lot of value in many cores, but with fewer cores. Because the fewer cores you have, the more opportunity for a very reliable end user experience. When we go to higher core counts, there's more blast radius, especially if you have multiple users on a system. So there's always kind of this tension between having more cores, because then you can do more work and amortize that over all the different bill of materials costs over more cores and more revenue generating entities.
But you also have to balance where you need to have a very reliable experience, and databases are often a mission critical kind of experience. So PMem really gives that opportunity. It also gives the opportunity of persistence in the memory format, and that is unlocking interesting new experiences. We've seen so much in terms of different projects in that domain space. I won't do it credit to just say it, right? It's really about thinking through the differential between accessing storage and memory, and it has just gotten more and more extreme over the last few years between hard drives, the SSDs may get faster.
But even then what we're seeing from a DDR perspective versus what we have seen from PCIe from a bandwidth perspective, they weren't staying super well aligned. I think with DDR5 and PCIE Gen5, were starting to get close again and we'll see some interesting innovation just as that. But as these two formats have gotten further and further apart in terms of latency and speed, we've seen people turn off swap, not be willing to take the hit of going out to a storage medium. And that's why memory has grown so much, so fast, so quickly to the point where for a lot of our customers, it's 60 to 70% of the bill of materials costs.
So being able to reduce that by using a cheaper medium that's slower than standard memory, but has a transparent use case, so that you can actually utilize it like a second level of memory, that allows us to really unlock so much more performance at the application layer. >> Yeah, there's not many applications that do not benefit from warm memory. I think we like to pick on databases, specifically in-memory databases like SAP and even some in-memory stuff from Oracle, and some of the open source stuff are easy kind of examples.
But even in VMware vSphere environments, I always tell customers, if you have to choose between core count and warm memory, I've rarely ever seen an environment that was starved for cores, but I've always seen environments that are starved for memory. So warm memory is always, always good. And specifically VMware vSphere type environments. So Rebecca, I'll leave the last word to you. What did we miss out? Is there any highlights from the 3rd Gen Xeon scalable launch that we missed? >> Oh gosh.
I mean, there's probably a million. But what I would just say is we are so excited to bring this product to market, to be able to have the opportunity to see what confidential computing at scale really can do. To see what that intersection of AI performance with confidential computing can do. What the crypto acceleration that we're bringing out can leverage across the cloud to make you more secure every day performantly. And really, just to bring that move, store process together, whether it's adding from our connectivity portfolio, memory, PMem that we talked about, our SSDs, and all the work.
We didn't call out our wonderful partners in the software side, but all the work from one API to our different cross architectural tools and performance tools that we bring to market, there's so much that we do to try and make sure there's workload optimized solutions from the cloud to the edge, all across the ecosystem. We are your partner and we want to be your partner. We want to earn your trust as a great partner, delivering performance at scale. >> Rebecca Weekly, we really appreciate you joining the program.
com. You can interact with me on Twitter, DMs are open, @CTOAdvisor. Talk to you next, CTODose.
