Senior Vice President of Corporate Marketing, Western Digital Corp
Dave Tang, Western Digital & Martin Fink, Western Digital l | CUBEConversation Feb 2018
Dave Tang, Sr. Vice President, Western Digital, & Martin Fink, Chief Technology Officer, Western Digital, sits down with Jeff Frick for a CUBEConversation at theCUBE Studios, Palo Alto
https://siliconangle.com/2018/02/14/r...
RISC-V plans to fulfill open-source architecture innovation dreams
Digital transformation and the proliferation of big data are driving a renaissance in software development, requiring new advancements in hardware and processors. With a range of needs from a variety of users and platforms, standard instruction set architectures are no longer fulfilling all use case...
About David Tang
In a February 2018 CUBEConversation, David Tang, Senior Vice President of Corporate Marketing at Western Digital, discussed the company's commitment to adopting the RISC-V open-source instruction set architecture. Tang stated that Western Digital ships approximately one billion processor cores annually across devices such as USB sticks and hard drives, and that the company plans to migrate all of those cores to RISC-V. He described the move as a way to bring scale to open architecture, allowing Western Digital to modify and optimize instruction sets for specific storage and data-centric applications. Tang said that the open-source nature of RISC-V enables the company to develop specialized processors for applications including artificial intelligence, machine learning, and IoT, which he argued require tailored architectures beyond general-purpose designs. He also noted that the collaborative ecosystem around RISC-V encourages contributions from partners and developers, which he said fosters innovation and enables niche applications. Tang characterized Western Digital's strategic shift as a response to the need for scalable, flexible processing solutions in a rapidly evolving data landscape.
Transcript (35 segments)
J
Jeff Frick0:10
Hey, welcome back everybody. Jeff Frick here with theCUBE. We are in our Palo Alto studio. The conference season hasn't really kicked off yet into full swing so we can do a lot more kind of intimate stuff here in the studio, for a CUBE Conversation. And we're really excited to have a many time CUBE alum on, and a new guest, both from Western Digital. So Dave Tang, Senior Vice President at Western Digital. Great to see you again, Dave.
D
David Tang0:30
Great to be here, Jeff.
J
Jeff Frick0:31
Absolutely and Martin Fink, he is the Chief Technology Officer at Western Digital, a longtime HP alum. I'm sure people recognized you from that and our great machine keynotes we were talking about it. So great to finally meet you, Martin.
M
Martin Fink0:44
Thank you, nice to be here.
J
Jeff Frick0:45
Absolutely, so you guys are here talking about and we've got an ongoing program actually with Western Digital about Data Makes Possible, right. With all the things that are going on in tech at the end of the day, right, there's data, it's got to be stored somewhere because you guys are a rising tide, lifts all boats, kind of company and really enjoy watching this whole ecosystem grow. So I really want to thank you for that. But now there's some new things that we want to talk about that you guys are doing to continue really in that same theme, and that's the support of this RISC-V. So first off, for people who have no idea, what is RISC-V? Let's jump into that and then kind of what is the announcement and why it's important.
D
David Tang1:35
Sure, so RISC-V is, the tagline is, it's an open source instruction set architecture. So what does that mean, just so people can kind of understand. So today the world is dominated by two instruction set architectures. For the most part, we'll call the desktop enterprise world is dominated by the Intel instruction set architecture and the mobile and embedded world is dominated by ARM. And so both of those are great architectures but they're also proprietary, they're owned by their respective companies. So RISC-V is essentially a third entrant, we'll say, into this world, but the distinction is that it's completely open source. So everything about the instruction set is available to all and anybody can implement it. We can all share the implementations. We can share the code that makes up that instruction set architecture, and very importantly for us and part of our motivation is the freedom to innovate. So we now have the ability to modify the instruction set or change the implementation of the instruction set, to optimize it for our devices and our storage and our drives, etc.
J
Jeff Frick2:53
So is this the first kind of open source play in microprocessor architecture? There have been a lot of attempts in the past to create other open architectures to kind of rally around this in a meaningful way has really been a challenge. And so I'd say that right now, RISC-V presents probably the best sort of clean slate, let's take something new to the market out there.
D
David Tang3:27
So open source, obviously we've seen, you know, take over the software world, first in the operating system which everybody is familiar with Linux but then we see it time and time again in different applications, Hadoop. I mean, there's just a proliferation of open source projects. The benefits are tremendous. Pretty easy to ascertain at a typical software case, how is that going to be applied do you think within the microprocessor world?
M
Martin Fink3:54
So it's a little bit different. When we're talking about open source hardware versus open source software. You still have to create a physical design and you still have to call up a fab and say, will you make this for me at these particular volumes? And so that's the difference. So there are some differences between open source software where it's, you know, you create the bits and then you distribute those bits through the Internet and all is good. Whereas here, you still have a physical need to fabricate something.
J
Jeff Frick4:33
Now, how much more flexibility can you do then for the output when you can actually impact the architecture as opposed to just creating a custom chip design, on top of somebody else's architecture?
D
David Tang4:45
Well, let me give you probably a really simple, concrete example that kind of people can internalize of some of our motivation behind this, because that might sort of help get people through this. If you think of a very typical surveillance application, it never changes and you really want, only know when stuff changes. Well, today, in very simple terms, all of those frames get routed up to some big server somewhere and that server spends a lot of time trying to figure out, okay have I got a frame that changed? Have I got a frame that changed, and so on. And then eventually it'll find maybe two or three or five frames that have got something interesting. So in the world what we're trying to do is to say, okay well why don't we take that, find no changes, and push that right down to the device? So we basically store all those frames, why don't we go figure out all the frames that mean nothing, and only ship up to that big bad server the frames that have something interesting and something you want to go analyze and do some work on? So that's a very typical application to really focus on the data that matters, and get some intelligence.
J
Jeff Frick6:14
And that's critical as we get more and more immersed in a data-centric world, where we have realtime applications like Martin described as well as large data-centric applications like of course, big data analytics, but also training for AI systems or machine learning. These workloads are going to become more and more diverse and they're going to need more specialized architectures and more specialized processing. So big data is getting bigger and faster and these realtime fast data applications are getting faster and bigger. So we need ways to contend with that, that really go beyond what's available with general purpose architectures.
M
Martin Fink6:51
So that's a great point because if we take this example of video frames, now if I can build a processor that only does that, the processor is much smaller, it uses much less power. So the cost of the processor that I put into the device where we put it, is a tiny fraction, but the cost savings of the overall solution is significant. So this ability to customize the instruction set to only do what you need it to do for that very special purpose, that's gold.
J
Jeff Frick7:27
So I just wanted to, Dave, we've talked about a lot of interesting innovations that you guys have come up with over the years, with the helium launch. Which I don't know, a couple, two, three years ago, you were just at the MAMR event, really energy assisted recording. So this is really kind of foundational within the storage and the media itself and how you guys do better and take advantage of evolving land space. This is a kind of a different play for Western Digital, this isn't a direct kind of improvement in the way that storage media and architecture works but this is really more of, I'm going to ask you.
D
David Tang8:08
Yeah, it is a different play, but at the same time, we've been really focusing on, in the last few years, our mission and what we're about is enabling people around technologies that really help the world extract more value from data as a whole, right. So it's way beyond storage these days, right. We're looking for better ways to capture, preserve, access, and transform the data. And unless you transform it, you can't really extract the value out of it so as we see all these new applications for data and the vast possibilities for data, we really want to pave the path and help the industry innovate to bring all those applications to reality.
J
Jeff Frick8:38
It's interesting too because one of the great topics always in computing is you know, you got compute and store, which has to go to which, right. And nobody wants to move a lot of data, that's hard and may or may not be easy to get compute. Especially these IoT applications, remote devices, tough conditions and power, and then on top of that, you've got applications coming up like autonomous driving and extracting information from a lot of video data. So what's interesting here, where does the scale come, right? At the end of the day, scale always wins. And that's where we've seen historically where the general-purpose microprocessor architectures is dominated but used to be a slew of specialty purpose architectures but now there's an opportunity to bring scale to this. So how does that scale game continue to evolve?
D
David Tang9:37
So it's a great point that scale does matter and we've seen that repeatedly and so it's a significant part of the reason why we decided to go early with a significant commitment was to tell the world that we were bringing scale to the equation. And so what we communicated to the marketplace is we're going to go big and go big and that translates into a billion cores that we ship every year and we're going to go on a program to essentially migrate all of those cores to RISC-V. It'll take a few years to get there but we'll migrate all of those cores and so we basically were signaling to the market, hey scale is now here. Scale is here, you can make the investments, you can go forward, you can make that commitment to RISC-V because essentially we've got your back.
J
Jeff Frick10:37
So just to make sure we get that clear. So you guys have announced that you're going to slowly migrate over time your micro processors that power your devices to the tune of approximately a billion with a B, cores per year to this new architecture.
D
David Tang10:54
That is correct.
J
Jeff Frick10:55
And has that started?
D
David Tang10:57
So the design has started. We're in the process of building out the new engines. We have already developed some RISC-V based cores for use in our development. So the design has started and we're working on it.
J
Jeff Frick11:10
Okay, okay. But that's a pretty significant commitment and again, the ideas you explicitly said it's a signal to the ecosystem, this is worth your investment because there is some scale here.
M
Martin Fink11:19
That's right.
J
Jeff Frick11:20
Yeah, pretty exciting. And how do you think it's going to open up the ability for you to do new things with your devices that you before either couldn't do or were too expensive with dollars or power.
M
Martin Fink11:33
So we're going to step and iterate through this and one key point here is a lot of people tend to want to start in this processor world at the very high end, right. I'm going to go take on a Xeon processor or something like that. It's not what we're doing. We're basically saying, we're going to go at the small end, the tiny end where power matters. Power matters a lot in our devices and where can we achieve the optimum power consumption? So if we can reduce the power consumption of our devices, that's a huge win for our customers, you know. If you think about your laptop and if I reduce the power consumption of that SSD in there so that you have longer battery life and you can get you know through the day better, that's a huge win, right. And I don't impact performance in the process, that's a huge win. So what we do, what we're doing right now is we're developing the cores based on the RISC-V architecture and then what we're going to do is once we've got that sort of design, sort of complete is we want to take all of the typical client workloads and profile them on that. Then we want to find out, okay where are the hot spots? What are the two or three things that are really consuming all the power and how do we go optimize, by either creating two or three instructions that can bundle that down.
D
David Tang13:12
We're in a unique position I think in that the technologies that we develop span everything from the actual media where the bits are stored, whether it's solid-state flash or rotating magnetic disk and the recording heads. We take those technologies and build them all the way up into devices and platforms and full-fledged data center systems. And if we can optimize and tune all the way from that core media level all the way up through into the system level, we can deliver significantly higher value, we believe, to the marketplace. So this is the start of that, that enables us to customize command sets and optimize the flow of data.
M
Martin Fink14:11
We're not going to develop our cores for all applications. We want the world to develop all sorts of different cores. And so for many applications somebody else might come in and say, hey we've got a really cool core. So one of the companies we've partnered with and invested in for example, is Esperanto. They've actually decided to go at the high end and do a machine learning accelerator. Hey, maybe we'll use that for some machine learning applications in our system level performance. So we don't have to do it all but we've got a common architecture across the portfolio and that speaks to that sort of open source nature of the RISC-V architecture is we want the world to get going. We want our competitors to get on board, we want partners, we want software providers, we want everybody on board.
J
Jeff Frick14:57
It's such a different ecosystem with open-source freeing up a lot of possibilities that you can address a customer by their specific need, not by big general-purpose middle anymore. That's not a great place to be, there's all kinds of specialty places where you can build the competence and with software and you know with, thank goodness for Moore's law decreasing prices of the power of the compute and now the cloud, which is basically always available. Really a exciting time to develop a myriad of different applications.
D
David Tang15:34
Right and you talked before about scale in terms of points of implementation that will drive adoption and drive this to critical mass but there's another aspect of scale relative to the architecture within a single system that's also important that I think RISC-V helps to break down some barriers. Because with general purpose computer architectures, they assume a certain ratio of memory and storage and you need it for scale. So that's another great benefit of these new architectures is that the diversity of data needs where some are going to be large data sets, some are going to be small data sets that need high bandwidth. You can customize and blend that recipe as you need to, you're not at the mercy of these fixed ratios.
J
Jeff Frick16:30
Yeah and I think you know it's so much of kind of what is cloud computing. And the atomic nature of it, that you can apply the ratios, the amount that you need as you need, you can change it on the fly, you can tone it up, tone it down. And I think the other interesting thing that you touched on is some of these new, which are now relatively special-purpose but are going to be general-purpose very soon in terms of machine learning and AI and applying those to different places and applying them closer to the problem.
Martin, I want to turn to you because you were looking way down the road for a long time. So you came out, I'd looked at your LinkedIn, you retired for three months, congratulations. Hope you got some golf in but you came back to Western Digital so why did you come back? And as you look down the road a ways, what do you see that excites you, that got you off that three-month little tour around the golf course and I'm sorry I had to tease about that. But what do you see? What are you excited about that you came back and got involved in an open source microprocessor project?
M
Martin Fink17:40
So the short answer was that, I saw the opportunity at Western Digital to be where data lives. So I had spent my entire career, we'll call it at the compute or the server side of things and the interesting thing is I had a very close relationship with SanDisk, which was acquired by Western Digital. And as I looked at that, I said, you know, all data flows onto and off of Western Digital devices, taking that from a real position of strength in the marketplace and say, what could we go do to make data more intelligent and rather than start kind of at that server end and so that I saw that potential there and it was just incredible, so that's what made me want to join.
J
Jeff Frick18:32
Exciting times. Dave good get. We're delighted to have Martin with us.
M
Martin Fink18:36
Thank you.
J
Jeff Frick18:39
All right, well we look forward to watch it evolve. We've got another whole set of events we're going to do again together with Western Digital that we're excited about. Again, covering Data Makes Possible but you know kind of uplifting into the application space as a lot of the cool things that people are doing in innovation. So Martin, great to finally meet you and thanks for stopping by.