Homepage > News & Media > Starting with FPGAs is hard. To have a successful deployment, you want to start in software.

Starting with FPGAs is hard. To have a successful deployment, you want to start in software.

Magmio Podcast. To have a successful deployment, you don’t want to start with FPGA. We challenge the speed of light. Crypto is not there yet. 

We are thrilled to share our very first podcast we have done at Magmio, featuring a discussion on FPGA technology and high-frequency trading led by our friend Mike Bell and our business manager, Michal.

In this insightful discussion, Mike and Michal delve into the intricacies of FPGA (Field-Programmable Gate Array) and how Magmio is pioneering low-level hardware programming to help their clients succeed.

❗️Key Takeaways:

⏩️ You can achieve only a certain level of latencies with software, and if that's not good enough, the FPGA is the logical next step forward.

⏩️ The FPGA journey starts in software.

⏩️ You can be up and running in months, not years.

⏩️ There is no viable use case for FPGAs in crypto yet, but when there is, we will be ready.

⏩️ Our business model is like Netflix. You pay every month, and you can stop at any time.

This podcast episode is a must-listen for anyone interested in the intersection of technology and trading. Whether you're a professional trader, CTO, or simply curious about these topics, our discussion provides valuable insights.

👍 Our team has done terrific work making this premiere episode, and we would be happy if you listened to it, shared it, and gave us some feedback.

⏩️ SwitchGearsNow | magmio.com


Mike Belt: Hello friends and welcome to our podcast today. We thank you for joining us. My name is Mike Belt and I am very happy to introduce my friend Michael from Magmio.

Michael: Hello Mike, thank you for having me.

Mike Belt: Thank you for being here, so we're going to talk about some interesting things today. One of the things that Michael is going to talk about with us is HFTs and the application of some FPGA cards in HFT. Without further delay, let's just go ahead and get into this. Michael, I would like to ask you what, you know, before we talk about Magmio or anything that Magmio does as a company, can we talk about these FPGAs? Can you explain those a little bit, please?

Michael: Sure. FPGA stands for field programmable gate arrays. Basically, it's hardware which can be programmed. FPGA chips more specifically can be programmed to carry on specific operations and custom requirements and in our case, what we're going to be talking about, is executing the trading algorithms more specifically.

Mike Belt: Okay, so since you said that, how can these FPGAs be used in trading?

Michael: Well, you can offload basically your algorithm inside. You can put an algorithm inside the chips, the FPGA and the FPGA chips can execute, as I said, the algorithm itself completely independently on the software. That allows you to reach those high frequency ultra low latency nanoseconds latencies and by the end of the day to win more trades and make more money. That is the typical use case that we are at the Magmio we're specializing in.

Mike Belt: Okay, so basically you could say that there's sort of a nitro boost put into this card.

Michael: Yeah, you can say that.

Mike Belt: Okay, I understand better. So let's go ahead. Let me ask you a little bit about Magmio. Who is your customer, what are they looking for, what kind of cooperation do you offer them?

Michael: Magmio is as a company, as a product, we are FPGA based ultra-low latency framework and where you put the tick-to-trade pipeline inside the FPGA basically and then, therefore we allow our clients, even if they have no expertise whatsoever, to use this technology in the trading and have a successful FPGA deployment. We're allowing them firstly by, like I said, offloading the critical part, trading critical part inside the FPGA inside the chips. That means market data decoding – you basically decode the data from the exchange as it does book building. You're building the book inside the FPGA if that's a part of the project and of course the order entry as when you're firing orders back to exchange that everything is happening inside the FPGA as a first part. The second part of course is that we provide the API, the C++ API, the interface, which allows you to communicate with the card which allows you to do the implementation and everything you need to know with kind of oversight about everything. This is how we help our clients. To your question what the client is? Usually it is the prop Trading Company who don't really have the FPGA expertise because this is really still a very, I would say, niche thing. Most people don't really know anything about it.

They don't even know that this can be used in the trading and we're trying to change that. So even if they really don't have any expertise in-house because hardware designers are hard to find, they're very expensive usually, but even then, we are allowing them to use it because you only need to know C++, software language and so even if they only know that it's perfectly okay to use our solution, our framework. The second use case could be even the bigger companies, even if they do have the FPGA in-house team, they still could use our services because they want to utilize their existing resources and very often, their FPGA team is occupied completely, swamped with the work and then they want to use, it's logical only, to use some kind of a third party too or it could be a case to use a third party as the ways to speed things up.

Mike Belt: Right.

Michael: Because you have an idea, you have some kind of algo which is latency sensitive which means that if you were to get faster you would win more trades and make more money.

Mike Belt: Which is the ultimate goal.

Michael: Which is the ultimate goal, by the end of the day. And if you're not fast enough in software because there is only so much the software can manage, basically you can achieve only a certain level of latencies with software and if that's not good enough, the FPGA is that logical next step which will get you over the line. You know which will get you into that nanosecond latencies, I would say areas. The whole Magmio is like, I would call it as a software development kit for better imagination, but instead of CPU your algo is running on the chip, on the card, which you plugged into the data center of the exchange itself.

Mike Belt: And then it's just super fast.

Michael: In some cases you are quite literally challenging the laws of physics. The speed of light and that kind of stuff. It's all very technical. It sounds cool like, yeah, it's really like that.

Mike Belt: It does sound amazing and you're talking about time sensitivity in trading and, of course, the goal is to be effective, to make money. So let's just say that when we talk about time sensitivity, I'm a customer, I meet you, I'm introduced to Magmio, getting this thing started and seeing some results at the end are very important to me, so what kind of time frame are we looking at and getting this cooperation going and seeing some results?

Michael: Well, it's all about the priorities, I would say. At the very beginning, it is always full of conversation between us and potential clients, because we need to have this kind of conversation at the beginning to really understand what they're trying to achieve and if the FPGA is a good way forward. Because it may not be, and it's only logical to say sooner rather than later just to find out. After that in general from the big picture side, it's pretty straightforward, you have to get the card, you have to plug it inside the data center of the exchange because you need to be as close as possible to the exchange to get the lowest possible data ideally direct market access to be directly co-located as they say to the exchange and then it's all about getting your algo inside the card and get started but the low level data is really important. That's because if you don't really have that in place, if you don't really have that infrastructure in place you would lose the race even before you enter it. And that doesn't really make any sense. That’s really important. What I would like to say at this point is, you don't really want to start with an FPGA, you don't really want to start with the card itself.

Mike Belt: Okay, your focus and Magmio's focus is the FPGA but you're saying don't start with the card.

Michael: Yeah, that's what I'm saying. I know it sounds weird, but you can imagine that this is a very difficult project to be deployed and to have it and to be successful in this industry and you need to be really careful and you need to test a lot. This is a big investment which takes a lot of risks. So to ease that risk, what we've done and how we advise our clients to start, and what is the typical way how to get started with us is with evaluations, we call them evaluation. The very first is software evaluation, which basically means that we've built the software model which is a C++ representation of our FPGA framework so everything is there like a market data decoding, order enter but everything is just like software which can be sent to the clients quite easily. It’s just a link. This is the very first step, how they can start playing with it. But it's still software, you don't really need a card, you don't really have to have it at this point, but you can get very easily familiar with the interface with the API, with the implementation. Everything you need to know you can get familiar with that as a CEO of the company or your team, your project manager, especially, if you don't really have any expertise with FPGA that's very, very important because there are going to be a lot of things which are completely new and sometimes it's very very difficult to wrap your head around it and you will have a lot of questions and we're trying to help the clients as much as possible. So this software evaluation has two parts. It's first the model, as I said, for the interface implementation that kind of stuff and the second is the HLS compiler. This HLS compiler, it's a well-known tool, I think, made by Xilinx, and it's a compiler for C++ code into the VHDL or Verilog - hardware languages. It's a very, very important part because basically it tells you two things. The first thing is, if you put your algo and have it translated, converted, into hardware languages, it will tell you first the resource consumption, basically how big, how complex your strategy is and whatever it can be fitted inside a card because it may not be. It might be too complex or too big if you have too many symbols, too many orders, too much stuff which is very common when you trade in software because in software you can quite literally decide how much memory you want to use on the fly but you can't do that in an FPGA. There is no dynamic allocation memory or anything like that, so you have to know beforehand if it can fit. If not, you have to do things differently.

Mike Belt: Right.

Michael: But if it can be, that's good news. The second thing which it will tell you, the report which will tell youm is the latency. How fast basically you are going to be with this specific algo which you have. So that is the software evaluation. In an ideal world, what we are telling our clients and once you get the ball rolling in the software evaluation and you have maybe a month and you like what you see and this is good and this might be a good path forward. You order the card at the same time so while you're testing in the software you're still waiting for the card so you're using the time in a smart way and once you get the card because there is some kind of lead time before you get it. You should be ready to just take your work which you've already done in the software and basically put it inside the card and, just don't take me literally, copy-paste the stuff from the software and then put it on the hardware because the interface is going to be the same so there's nothing new about that so it's going to be ninety-five percent the same not one hundred percent because there are still differences between hardware and software. Something which cannot be simulated.

Mike Belt: Right, it makes sense.

Michael: Yeah, but most of it will be the same and that's very good news and once you get a card, like we talked about, you plug it inside a data center with direct market access with the data and you start testing, still testing, with the card see how it behaves and what algo is doing, because this is all very important. This is how you test and once you like what you see and you think already the next logical step is production trading. I mean live trading on the exchange. 

Mike Belt: Okay, so I think a lot of people are probably waiting to ask this question. We're talking about the stock exchange so what exchange do you actually work on, work with?

Michael: Well, we work with various exchanges all over the world. Historically we've started in the USA on NYSE and NASDAQ and then we've added CME and BATS and now we basically support all the major exchanges in the USA. We've added a long time ago EUREX in Frankfurt in Germany, now we're working closely with NSE - India National Stock Exchange of India. We're working in Turkey in Borsa Istanbul, so it's all over the world. We're just about to add Thai Exchange in Thailand, we're very excited about that, and we're looking closely at Brazil, the B3 exchange, the third biggest commodity Exchange in the world right now. I think that's going to be a very nice exchange to have. We're looking at the Singapore exchange as well. We're usually very client driven so if somebody comes to us and says ´Hey I want this exchange,´ you know, I’d say sure yeah, even if we don't have that for us it's not that big an issue to add a new exchange. Everything is in templates and everything is very generic, we made it that way. So with us even the time to market is much more swift because setting up all of that by yourself, that would take years this is just a matter of weeks and months. 

Mike Belt: That's where you and Magmio made it easier for your clients, this is your expertise.

Michael: Yeah, sure, like the market data protocols and low level programming and that kind of stuff you don't really want to focus on that. If you're a trader and if you're out there trading every day fighting to make money, you don't want to care about some low level FPGA programming that nobody really knows anything about and some weird protocols which come from the exchange. You want to have it covered completely and this is what we do. You want to focus just on your algo, because by the end of the day, this is what it's going to make you money, this is your bread and butter.

You want to have it as good as it possibly can be and this is like what we're saying to our clients. You take care of your algo which you don't really share with us at any point. I think that's very important to mention that we made it that way, we understand that because they are really secretive when it comes to that.

Like I said, which is understandable. You focus on that and we just focus on everything else. The low-level programming, the IP cores, market data decoding, book building, that kind of stuff. We take care of everything and you just focus on the business and that's very important, plus it's not going to take two years to be deployed, because you have the algo now, you have the algorithm which is suitable, which can be good right now and a lot of things may change in two years for example, that's very important. 

Mike Belt: Obviously, a lot of exchanges are supported and you're open to more but there's another big question, a big topic today and it's crypto. So what about crypto? Is that available for exchange or with this program?

Michael: Well, the short answer is no. We don't really do crypto. The reason being of course, I think it's logical, because we have talked about exchanges and direct market access and stuff like that, none of this is in this exact sense is with crypto, the FPGA and HFT is really about centralization. You have one exchange, one market data and you want to be as close as possible to achieve these latencies, but crypto is kind of the opposite. Everything is happening everywhere. It's really about decentralization. That was the whole idea about it, to be decentralized. But once there's going to be a suitable use case for it, we're definitely gonna be looking into it and we're gonna be supporting that. It's not like we're against it or closed to it but there we don't really see the viable use case now but, like I said, when there is the case, we will be there. 

Mike Belt: Yeah, well. I think that makes sense. When you said it's decentralized and you're working with centralized, it's very, like you said, it's the opposite.

Michael: Yeah, yeah.

Mike Belt: Okay, all right, so working with somebody that can participate in this and wants to participate in this, how does your business work? How does it work? 

Michael: Yeah, we like to keep things simple so it's basically we just license our IPs every month, so I often like to say it's like Netflix, you just have to pay every month and you can stop at any time and it's not really the one-time thing payment. Nothing like that. It's just a licensing structure and the part of it, of course, is not just IPs but also our time customer support, new features the clients want to add, and the stuff like that. Because quite frankly any long-term contract anything like that doesn't really work in this business because a lot of things can change, like, quite fast and you don't want to be stuck in a contract which doesn't really work for you. 

Mike Belt: So, if my strategy changes, I can stop, I can give you notice within a month.

Michael: Sure, yeah.

Mike Belt: And stop my payment.

Michael: Yeah, of course. We will do everything that we know. We will do our best, not to happen, not to let it happen if you know what I mean like, so of course, but it could happen, of course, it could happen that the use-case which was successful yesterday will not be successful tomorrow and then it's really important to have this freedom to say not just I want to cancel it in order to not to use it ever again but rather I would like to take a pause just to re-focus. I want to focus on different strategies, just back to the drawing board and I will come back to you. That would be preferred actually. 

Mike Belt: Yeah, okay. That makes sense and I think businesses would like to hear that. So working in all of this and with all this, what is the cooperation like between the client customer and Magmio?

Michael: I think that the crucial part, the crucial period of time is still the testing, what I was talking about, about the software evaluation, hardware evaluation. This is where most of the communication is happening and for us it's a crucial part of the whole project because, as I said, it's difficult to wrap your head around it, especially, if you're coming from the software  world and this is suddenly a hardware world and a lot of things are different. We would like to work on this partnership cooperation basis. When we tend to, you know, we call ourselves partners. We're really in this together because, like I said, this is not a one-time thing, it's not like we built something and sent it. Now it's on you and you're off you go but no. We're trying to be in touch with clients every day on Slack or Google chat, just messaging all the time, exchanging ideas, solving problems because this is, by the end of the day, this is all about solving problems, solving problems together, having the weekly calls, monthly calls. But I want to emphasize that period of testing which might take two months, three months, four months, there is no specific, we don't really tell the clients how long this is supposed to take. That's a very crucial idea when we are quite in touch every day, all the time, just to try to navigate them all the time and we're just trying our best to show them the ropes at the beginning and once they really understand the product, the framework and everything. That's the most important thing for us to happen.

Mike Belt: Exactly, that's when you know you've done well.

Michael: Exactly. We're just trying to be as supportive as we possibly can be and it's not really that difficult, because the guys love what they do, they really love working on it and they're really enjoying it.

Mike Belt: Okay, so you're talking about more of the people involved with Magmio, so there's obviously the expertise part of Magmio but if we got away from the expertise what other added value would you say there is in cooperating with or someone that chooses to work with Magmio?

Michael: Well, I would have to say the team because apart from the expertise which is obviously there because we have been doing for almost 10 years, the guys they've been doing this for almost 10 years so there isn't much they don't know about the FPGA, the technology, but there's still a lot they know about the trading itself because it's still the connection of two worlds, the technology, the FPGA technology, hardware technology and the trading and you have to know both. So the expertise is really there but apart from that we're still young guys, we are still a young team, nothing too serious, anything like that. This human approach, I would call it, it's also very important and something that we're making sure that it's always there. I would definitely call this an added value to the whole team and the whole spirit of the company, the still young spirit because I would say that the average age of the company is like 32 maybe 33 years old. Most of the guys are still quite young and, like I said, they love what they do. I think that it doesn't matter what kind of industry it is. It doesn't really matter what you do. You always kind of know if somebody likes what they're doing. 

Mike Belt: I was thinking the same thing when you said that. That’s a deciding factor between working with somebody sometimes and not working with somebody sometimes and not working with somebody sometimes. If they don't like what they're doing I don't know if I want to be involved with them.

Michael: It's interesting because everybody knows the software development, building apps and applications and whatever and not everybody knows about the hardware, that there is some kind of hardware which can be programmed. They teach that at school but most of the people, they really, like, at least what I've heard because I'm not a developer myself, but a lot of guys they kind of like ignoring it and they just want to focus on the software but those few stuck with hardware, they're really into hardware and they really want to like, do that, like they want to really be building the hardware and FPGAs and everything and that makes them even better I think. This spirit, as I call it, really flows around the company and I think it's enjoyable to work with somebody who works that way, feels that way.

Mike Belt: Absolutely, all right. So we get ready to wrap these things up. I think if somebody's listening and they're intrigued by this information, maybe it's new to them, maybe they know a little bit about it but they want more information. How can they go about getting more information, is there somewhere, someplace, some available to download some information easily accessible?

Michael: Sure, I think the easiest way is to go to our website magmio.com and there is a lot of useful information about the stuff I talked about. How to get started, about the product, the individual IPs and you can download a lot of stuff, product brief over there and all the information over there but I think the easiest way how to get the ball rolling is to write us a message and we will be in touch and we schedule a call or something and we'll give some kind of a really good introduction.

Mike Belt: Okay, that's the easiest way I guess. What if somebody wants to run into you in person somewhere, I mean are you guys out? You guys are global…

Michael: Well, we're based in the Czech Republic, in Brno, so if anybody's interested to visit this beautiful city they're more than welcome but apart from that we're trying to attend as many conferences as possible. I just came back from London last month. What we're planning right now is FIA Expo in Chicago where we're doing this basically every year and we're going to have a booth there I'm gonna be there and one of my colleagues will be there so that's the basically the easiest way to meet us, to have a coffee and discuss the FPGA and the hardware acceleration and anything. I think the FIA is the first week in October this year, if I'm not mistaken, but more information will definitely be on our website so you can just check it out. 

Mike Belt: Awesome. Well Michael, I can't thank you enough on behalf of myself but you're on behalf of Magmio, the information that you've given. I want to thank him, Magmio and all of this information about how to advance your trading and your business and that bottom line making more money. If you do want to have more information don't forget magmio.com. You can go there and download some more information and, as Michael mentioned, you may even have an opportunity to even meet him face to face in Chicago and have a coffee. Thank you for visiting with us today, thank you for being here with us, thank you for watching and we wish you all the best.

Michael: Thank you Mike.

Mike Belt: Thank you.




Contact us

Let’s accelerate your trading

By clicking Send, I agree to the Privacy Policy.

This form is protected by Google's reCAPTCHA service.

Ask us anything

By clicking Send, I agree to the Privacy Policy.

This form is protected by Google's reCAPTCHA service.