 With me on stage is a wonderful person who will talk to us about CPU development power open source and how this all fits together and how this Will influence hopefully influence our future CPU architecture and development So please give a very warm round of applause to Matteo Michel Okay Thanks for having me here today And I apologize in the beginning for my strong German sex and mixture of German and English so so for that because I am coming from Saxony actually living in Chemnitz still born in Chemnitz grew up in Chemnitz or Kamarstadt whoever knows about that That's why the accent is still mixture of sex and German But still want to show that are still smart people living in Saxony Not just people in Tristan waving some strange flex all the time. I want to say Thanks I just want to give a brief introduction today about open power. I got inspired I think two years ago by a talk by David Kaplan who was talking about CPU development all together from scratch to the final product and all the steps for it So I also want to touch that briefly today, but I want to go more into open power and Just tell you something about it because I'm not really sure if everybody knows about that yeah foundation and Might also be a little off topic in this conference here, but maybe is still of interest This is my agenda for the day So first of all like to just talk a little bit of about me then As mentioned about the CPU there have been a throw at all and why open power About myself as I mentioned still living in Chemnitz, but I'm my desk is located here This is the building in left of IBM Germany I'm there in the verification department of CPU development It's not just located there It's a worldwide distributed team virtual teams. So I'm working together with people in Bangalore US France England and so on so basically Yeah, all the continents are included My main focus is verification I'm also not an expert of open power But I'm just still involved in some topics of it. That's why I think I can at least talk about some steps here And that's all about me Let's go into the technical agenda This is the CPU development for as we are still working it today at IBM So we basically it's yeah a simple overview you start at the left by Either coding or generating or somehow specifying some hardware description Maybe VHDL maybe be a lock whatever you think might work This is also this Large arrow pointing to it the source that I'm using usually for verification There are also multiple different versions of HCL that are possible But just assume it's basically one thing that is specifically specifying the hardware how it should work and Yeah, there are some kind of tool stream that basically converts all the different things to the final product on right So out of the HDL you will get a nest that list with all those logic components being generated There are also some other specific verification teams working on that one Then you can generate the schematics out of it. Yeah, you will see some my trend resistors in there Yeah, basically Forming those resistors into the layout and with the layout you basically get to the final product The I see at the end out of the fabrication process I will also have some numbers later today, but assume this process you can go through Yeah on a daily basis, but from a first line of HDL code until the final product. Yeah, just assume like two or three years until you have a real working product Now there's some nasty Flowchart from some of my colleagues that somehow just should illustrate How the overall flow is somehow working? So you just saw the the version coming from hardware description language into the product now This is the workflow inside the company itself. So you have basically some kind of product You really what is your target or really the previous version? You have some customer that somehow has some urgent need for certain things to run on that And to basically give some kind of specification You have some teams working on high-level design and then this is basically your input source Sometimes very specific sometimes very loose And then you have different teams working in parallel trying to to get this product put together So we have some kind of system component teams that are looking at the boards and all the electrical Components on it. You have a team that's looking at this black box. We had to put it at the end You have firmer teams You have basically design teams that writing that reach the L code or very low code and then you have basically at the end Verification team that is basically making sure That this also works at the end before you go to have and get it produced And then you have also other teams. Yeah, looking at timing and looking at circuit design integration So that's all going more into physics Which is really important at the process of design because you want to have the maximum performance coming out of your product Yeah, when you're done, basically you push the button get it fabricated then there are some tests going on and If that goes well, then you have a good machine Sometimes you find bugs. That's why this small red dollar sign is coming in If you find bugs you basically want to find them as early as possible To basically turn and turn around your logic until it really works before you get to the fab If you miss that point, you have to spend much more dollars on it and then you yeah Have to spend multiple cycles on that Just be sure you don't want to be that person being responsible for having to re-spin through the fab because that process going to the fab is like two months two and a half months and You don't want to be that person first of all spending the money and also causing that delay for the customers at the end And if you are really bad person then at the end you get into manufacturing all those machines are being put to the customer side they push a button and nothing works and Yeah, let's don't talk about that, but I didn't see that in my life so far This is another picture somehow trying to illustrate the same thing again, but now trying to decide to the left more to design verification on the right side more to Yeah Physical stuff that is being done. So you see a physical science into this placement routing stuff that you Know it needs to be done But basically I personally don't want to be in that area. It's always critical and you will always be at the end of the process Everybody's waiting for you to push the button That's why I chose the left side being more into verification front Also dealing very close with the design teams And you see it's it's two circles Circling because you are always working somehow in the same source different times and that's basically what you do all the time We are going more into a kind of continuous integration Okay, and here are the numbers that I promised some will try to compare the numbers from Building a larger ship Compared to a processor And that's quite interesting because some more at the end when comparing you for both things you need at yet Need at the end for the whole system about three thousand people Also about the construction time. It's similar. So we're like three years for the ship like four to five years for the whole Processor and system. It's not four to five years for just one shot It's basically a multiple turns of the processor until the machine is basically ready to ship to the customer And also You see the cost at the end. So it's more than one billion dollars that needs to be spent for one generation of those systems Just to mention there are two Architectures currently still being worked on at IBM. It's the power architecture and DC architecture Those numbers are for the C processor But you can compare those pretty similar from a cost point of view So that's already it from the CPU development aside, so let's jump in into open power and Yeah, looking at that That's why I called wide open power because sometimes it looks strange Why should something change in that kind of hardware stuff? It's always be there So you see you work on your PCs. You see somehow getting smaller you have your phones But somehow similar kind of yeah stuff in there Also from the server part of it. Yeah, it's somehow in some building in the cellar and it should just work But there's some commercial component to it Over years this kind of process worked very well for players like IBM or others But basically over time the market shifted drastically So you have a Moore's law which basically provided over years a very constant flow of Performance improvement, so you didn't have to do much You just had to move to the next generation of your technology and then surprisingly you got 10 20 Whatever percent out of new performance without changing much in your design Nevertheless you added more features to it to satisfy customer needs But that is more and more Never true anymore Also on the other side you have other trends like new kind of workloads are coming up just thinking about some kind of Bitcoin mining, which is somehow trending today or these days, but also kind of consumption models of those server farms are changing and in previous years it was like Somebody from some company called. Yeah, I need the new machines then the customer presented move there and yeah Sold something which was staying there for the next couple of years But these days cloud is something very drastically changing the operation model of our sellers at IBM and That also basically was something that we felt because at the end just less and less products were sold The cloud infrastructure was not up and running from our point of view because we didn't bet on that Other companies like Amazon were very fast on bringing it up So as you can see today most of the cloud infrastructure is x86 based and I think it's hard to find a power servers for the public public cloud providers On the other side you have seen also the software ecosystem changed a little so the To more open and mature open providers That's one part of it and on the other side you have some kind of of strategy changes that you Need to do now, but more to that later on Just back to the most law On that chart you see it's basically compared performance per dollar. That's something The customers look at so they look at how much money they need to spend for which kind of performance And that's the number they compare between the different providers and then only if you can Match or be below the the other companies than you you are able to sell stuff to them And as I mentioned before that was working quite well for the past couple of years But now basically more slow slowing down and doesn't help anymore So what you are now trying to do is you need to somehow enhance the performance by other means So what could that be so you could plug something to your system So you can't move everything onto the processor because you just can't fabricate it So you need to do something close to the processor. You can block some x-ray letters on it You can somehow work on the storage systems to be faster You can look at the IO if you can move to a new generation there or you can look at the memories If they can provide you something But at the end everything comes together at a data point So data is that thing that somehow is told to be the issue over the next couple of years because of Different trends you have in other things providing Much or sometimes too much data and you just somehow need to handle it So every device is somehow creating data at the end. You need to have some backbone Working on it and basically some compute centers that can deal with all those data and you see what kind of data The greatest of those charts basically envisioned coming from different fields Coming more to open power and basically Yeah, most law is something I've already talked about but there were basically two other things that forced this kind of creation of this Open power community The second thing was Google Google was basically known to be a large provider of something like search or other stuff And Somehow they were not really happy with their x86 infrastructure. So somehow they were talks between other companies and also with IBM and they basically ran some stuff on the back then power 8 machines and were surprised for different patterns that were surprisingly good and on the other side, they also Basically we're looking right in the eye of a monopoly of Intel. So that was a second thing Google was not being afraid was afraid of and basically Seeing that this was creating a very costly infrastructure. So basically being bound to a monopolist and then Not open to to any movement somewhere else. That's why they Basically felt to do something else and also at the end I don't really know but at the end they Want to build their own computers and that's why they they searched for some partnering and at the end came to IBM and Yeah, look at us if we can do something together as well as the third point is Extend the reach of power architecture. So as mentioned before In the past there were power PCs power character was somehow a little stronger before power architecture is Getting market share or gaining market share in the last couple of years But the overall unix market is decreasing that fast that it's basically not really working out well for Keeping that architecture how just how it is. So you need to find other means to to strengthen the whole Infrastructure and the architecture over those boundaries and see how you can work with other parties to Get into other different markets And that was basically the root cause Basically get this open power initiative being done in 2013 If you look at the the other part of this chart it back then it looked like all those companies somehow losing market share overall Try to come together and found something But basically looking at its DC days It somehow looks like a good move and let me show you how that worked out Um the first steps that were done We're basically done by ABM First of all basically investing a lot of money into Linux Linux being able to run on power That was a complaint even done by engineers internally over the the last decades basically that even our own applications We're not running really good on our own machines Because we all our own compute pool was mainly based on x86 machines Because we were selling them also before That was a strong complaint, but it didn't help to complain internally and but The customers basically showed us that with power machines. We weren't really able to to beat the market so there was some kind of mind shift going on and basically resulted in investing a lot of money in Yeah, we structuring the code around Linux and basically even a little Indian big Indian Complications to sort them out and get them resolved at the end Second thing is this kind of open source source model for hardware. I think I will talk about that in a minute And basically it's the third step to release Lots of firmware code to really make it available Myself was even surprised sometimes. I just googled my name and some kind of word and yeah surprisingly I One of the first five matches was just some C file that I wrote myself with a colleague So just try to to search for it. You should find it rather easily on the net now This is the mission statement, I think nothing to really read through completely, but The idea behind this is it's not a new company to file found, but it's some kind of partnership or some some open Community That basically is to be installed to get people together that want to improve the power Architecture as well as to to work on the whole stack So that includes basically the basics of the hardware development as well as the the software development flow and basically use Same tooling use same infrastructure use same boundaries That's just a picture to illustrate that on the left So you have the hardware component that basically consists basically of first of all the processor itself and Still the processor doesn't help much You need to have some connection points and which is basically interfaces to Accelerators and not just those interfaces, but I have them rather open that other companies can also you reuse them There's also roadmaps for both of it. That's what I learned also very early in my work That customers really want to see a road map over the next couple of years that they are Can be sure to invest in your product and Can be sure that this continues over the next technology technology cycles As well as the software stack above of it to make sure that this is basically a common and being used by many Different parties and that it basically gets debugged well And that basically gets us to the Ecosystem as of today so the chart before that I showed you have seen there were five parties Founding this community and now you have basically a whole stack that is basically starting with the basics at the bottom now with the thing in IBM as well as well included with the chips Then you have bought manufacturers I always storage companies System integration companies also software But as well you have many research teams included and that's basically something that is Also a key interest of that group to not just do something with companies and try to sell stuff But also to keep people involved in research and trying to get this whole method enhanced To trust some up like 220 members as of today in 32 countries These days I think most of that is still centered around the US because there is basically the headquarters of IBM and the main labs still Located and that's why also many of the universities there joined in those areas but still basically try to get members over the whole stack and even those partners Also started to bring out first products But I think I have one chart for that later as well That's basically Just to illustrate also How open power somehow matches all the other initiative initiatives So we have this open compute group that is I think more concentrating on mechanical electrical synchronization and specifications The open power group somehow is trying to to fill that gap between Whatever is coming from top from software Linux or the stack and Open compute at the bottom Open Cappy is some more part of that open power initiative But it's basically just one factor of it and basically as you can see the screen box in there Basically limits itself to protocols and I owe in the connections internally So that's how it should fit Or how it's intended to be And now what's so special about it? Just to show you that's the the current roadmap. I'm not sure if this is even still accurate But at least it gives you an overview how that is Was being worked out over the last years So as a process a manufacturer you try to to have a real good sense of regular Push of products out into the market So there are different needs If you look at power, there are different circumstances because you have a different Their market that you want to attack if you look at C architecture This will look much different a C processor will come out Yeah, each and every time at the same point in time So there is a very regular pattern by on the power side that needs to be much More flexibility due to the the customer changes or market changes at the end But as you can see in the next years 10 7 nanometers are the products that need to come out and basically will be Specified and pushed with the partners But that's the really interesting thing about the the open power initiative that you somehow Try to to connect at the end As I mentioned before External accelerators into the system It's not just to plug them into but to make them part of the system to make them part of the the address map of the processor So it's not just an IO link some PCI also, but it's an protocol layer that makes it really fast compared to other customers also with This power eight enhanced product built together with Nvidia The NV link what it was introduced as an its own link built into the CPU to make that Bent with much faster into those GPUs Not sure if you heard about this announcement quite some weeks ago There are those labs in US. We are basically distributed with those new systems There was some kind of government contract One by Nvidia together with IBM to distribute those systems and basically they consist with the latest hardware of Power processors connected to the data's GPUs and basically built a room full of those servers And basically make possible some better computations and that uses the the newest generation of the in wheeling and Basically that provides bent with that you are not able to get with other CPUs That's the same thing But you see that you have those interfaces not just over a really you still have PCI generations available to get that faster just with the decapi extensions that you have and Here you can get Xilinx products attached also with cappy open cappy Which is basically this new protocol and hardware built into the processor This is just for numbers not for marketing Basically, they were able force web Exxon mobile was able to basically reduce 23,000 servers into 30 open power servers just to get an idea what that makes at the end Just for the numbers that's regarding the community itself The idea is basically to get or they have the possibility to get everybody involved that that means people that write software people that use software or people at the end that use the whole systems and try to basically Get those different in the industry groups together and have a place for them in that overall community And that's why they created different work groups To get more into technical discussions and basically technical solutions at the end and that's basically what we currently have today There are different different 13 different work groups available There are some that I'm even involved with some colleagues And that's basically talking about real problems and solutions at the end Really trying to solve them at different points of the stack and That's basically Should be the last slide regarding what that is at the end It's still some something that costs money That's why companies need to do to spend money to be part of open power But if you are coming from academia Then you are able to spend zero dollars and still be part of the community and there are different levels between them So it needs to be somehow a little bureaucratic that way obviously But you are able to get involved Especially if you are coming from research. I have even checked today to mention should be involved the part of one University I have seen on the chart as well today, so there are already some German universities joining that Community, but most of them were still located in US Um That was all From my side just a brief overview. So it was not intended to be in depth Also, not complete if you have more questions Just come to me and I give you also contacts to other people if you want to talk more details regarding any Interfaces or any contacts into those work groups But at first you should check on power.org. That's the website where you can get more information about that and I Hope that gave you some inspiration and some information about what is going on in your heart for community Thank you very much, Mateo. We do have time for a few questions and someone already lined up Mike one one place Hi, thanks for your talk as a hacker. How can I get a open-power development board if I'm not like a big data center like Google or someone? I think if you have contacts to universities should be possible to get something there Yeah, that would be my advice But I'm sure if you are Hacking activists you should have contacts in that group, but I think you can even Maybe get some some older stuff Since I've seen there are the older sports who are coming with P8 P8 is already The older version now previous version to power 9 is now rolled out Maybe there are also possibilities to get some more machines available The internet woke up and has a question. Yes, so you talked about Companies and academia joining open power and one user wants to know are their plans from IBM site to offer Affordable systems to consumers so they can join I think that's the at the end. It's at least an idea for for companies to get cheap systems available For the real end user. I think it's not the current target since it's also not the market IBM is looking into So it's more a business-related issue at the moment Mike four One question about the firmware you said it's 400,000 lines of code Why is it that big and why couldn't you just reuse Corbwood which already supports open power the older generations? That would be the specific questions I think it's just the stack that is coming out with the machines already At least from an IBM site and I think the same is coming from Google. They have additional lines of code written for their stuff and basically they just released what they already have So it would be a question to the firmware development teams why they don't do reuse as you suspected Mike two Question regarding Moore's law. It's never been about performance. It's been about number of transistors So do you have some numbers for open power regarding that? I don't have numbers here Need to look that up somewhere I guess But basically at the end performance is that what counts for the customers So they look at performance and that performance is that what you can sell And Moore's law was this the thing to improve performance over time and now it basically it's not that easy anymore That's why you need to find just other means Unfortunately, there are quite a few questions lining up. Unfortunately, we are out of time I'm very sorry for that, but I think Mattel you're still around so grab him and ask him in person So, please give another round of applause for Mattel. Thank you very much