 So we're here in the Taipei 101 building right here with Gigabyte and who are you? Hi, I'm Peter from Gigabyte. So this is the Thunder X2 right here and hi. Hi, nice to meet you again. And who are you? My name is Hoshino in charge of product planning in Gigabyte. So what are we looking at right here? Now we see this is the new product for the SoundX2 solution. It's the ARM64 solution. So is this ready now for shipping? Yes, some of we have three systems here. You see this is the 2U4-0 high-density solution. There will be the 1-0 of that. So this is high-density that means you can fit 4-0 in a 2U4 reactor. 2U4-0 systems. Very dense. Very high-density computing ability for the HPC. What's connected here? This is the storage 2.5-inch SATA SAS HDD. So it could be a bunch of SSDs maybe, 7.15 terabytes? Yes, in the beginning we just used the standard SATA SAS HDD. But in the future we will have some NVMe U.2 SSD. And how is the NVMe U.2 SSD going to be great? It will be six times performance than the traditional hard disk icing. And that's great for cloud? It's great for cloud, great for HPC storage and for a lot of new usage and applications. So and here you have some different systems too? Here's the 2U compactor. We call it R280 board system. It's also a sound x2. And we use the common design you can see here. These two systems use the same motherboard. That's the very optimal design for 1U2U boards to use the same motherboard. How's that possible? Because we use the common design in the beginning of the development stage and we develop the system. Not only consider about the motherboards, also we consider about all the total solutions. The server solution and the storage solution. Rear IO flexibility and a lot of power consumption we'll consider all. And make the common design to make the best, optimal design. But for example the cooling fans right here are smaller than there. But it's the same motherboard, but how's that going to work? That's different usage because it depends on region demand. Some regions may be in the United States, in Europe they want a 1U solution usually. But in China they have some similar concerns usually. So they want the 2U4 compactor mainly. Does this mean that this is the same density as that one? Because you can put two of them in the same space as one of those? No, this is only one node, but that's four nodes. But you can put two on top of each other, right? Yeah, but this is only one node, two sockets. This one in a 1U, it's two nodes, two sockets. So four CPU in a 1U, this is only two CPU in a 1U. So how soon is this available? So these two systems will be launched in the next months. Next month? We used the final version of the CPU, B1 SoundX2 CPU. So this is the general availability chipset, right? The final performance, the great performance? Yeah, great performance similar to the Intel PURLY solution and the AMD APEC solution. So how good is the performance? Because I just did an interview with the guy from Serve the Home. He used a Gigabyte ThunderX2 production machine and he compared with AMD and Intel. How's the performance difference? So usually for the, maybe your old imagination of ARM solution, the ARM is just the low end of the entry server market. But now they are performance for basically GCC spanking score. I think it's the similar number to the Intel PURLY solution and AMD APEC solution. It can be also used for the mainstream of the IT enterprise and HPC applications. So lots of application compatibility. But the performance also, the single-thread performance, how does that compare? Similar, but you know, for every applications in the server market, it's a little different from the consumer image because it depends on all case by case. Different applications use different solutions and provide different performance in the TCO. Because at Gigabyte you are experts in having many, many different server solutions all around here and there. So basically you have Intel, AMD and ARM. We have all the principle solutions. We are the most completed solution in the world for server market, I think, because we have the Intel main solution and also we have the AMD APEC solution. Even the Qualcomm centric solution, we have our system. And Sandex from Cavian is also our main product. And Sandex, the mobile generation, they shipped a lot of systems to customers. Yeah, a lot of them. Can you say how many? Oh, secret. But so far it's not millions, right? I mean, so far it's not a whole data center full of arm. As you know, I think it's just the beginning of the ARM solution because for the ARM solution current main demand is from the research and some data center for the sampling only. And to the 2020, there will be a very big change because in Japan the next generation of the supercomputer will use ARM architecture. And as you know, in China they are very concerned about the security engine so they want the open source like ARM solution. So government now pushing the ARM solution very hard in China. Because, I mean, we're not in China, we're in Taiwan, which is not China, but there's so many people. Everybody wants to do smartphone full speed. They want to do games. Everybody wants to have amazing cloud performance. So maybe it's very important for China to have a policy to say we need to consider maybe to be greener or to have better multi-core solutions that are more custom for saving power. You're correct, because for the ARM device, now more and more ARM devices are used for IoT and AI. So if they use the same ARM architecture, then you need some performance to translate the 086 solution to ARM solution. But if we can use the ARM solution directly for server in the backhand, then that program will be solved. And there's no any waste of the performance to translation. So ARM solution will be very good. I think it's a second source to get some market share from 086. That's why we think it will be crucial. But so far those designs compared to some Intel, they're not lower power in terms of one of these, you call it U2, right? The overall power you put in and all the coolers and everything is kind of like the same infrastructure, maybe the same cooling stuff that's in an Intel solution. Same amount of power. So it's not lower power, but it's a different architecture so it might consume less power per task maybe. So first for the design view, we use the common design. As you see, this is our R181 series. And you can see there are two systems. Both the R181 system, that's the Intel solution, the party solution, and that's where the other one will be, the AMD APEX solution. You can see we use the common design, the same... The fan to the same fan and some real design and the same CRPS power supply. That means we already make ARM solution be the mainstream solution in the world. So they use the same design to provide the same performance. And the same materials for the connector, same blue, same storage, use the same design. Only the chip and some PCB stuff is changing, right? Yeah, maybe you just can see the socket design is different, but really for our engineering side, you need to change a lot of things in PCB and a lot of schematic. It's very technical things, but we do the same design for all the architecture, that's true. What did you change in PCB? Can you say without saying any secrets, right? What was the biggest challenge in the PCB, for example? Yeah, for the X2 solution, I think they have some special things. Like they use the BJ type CPU firstly and so it's mounting on the PCB motherboard. And you know, Cavian also have their own LAN chip set solution called Q-Logic. So we use their own LAN solution Q-Logic and with the storage on the motherboard. So for this kind of view, so we have the optimized design for ARM solution to provide the best performance not only for the computing, also for the connecting and storage. And there are some other things you can say about the PCB design. What's going on here? This is similar, but how about over here? Yeah, so for this generation, you know there is a trend. So for the storage, customer want the high-performance things like the NVMe U.2. And the SoundX2 also can provide NVMe solution. So we will have NVMe solution for this new SoundX2 generation. So it's also the new design for our SoundX2 servers. The high-density design right here, is this something you can only do with Cavian Thunder X2 or you can also do with Intel on AMD? This kind of 4-in-1? Yeah, they are also the common design. You see there is also the 2-in-1 model for the AMD and Intel use the common design. So like this and 4 of them go inside? Yeah, there is a very high voice from the market that customers want to have the ARM solution. They want to have the second source solution. So for our target, for our business strategy, we provide more and more flexible solutions for customers, not only the Intel solution. So customers can choose by themselves. And one more thing is very important that in server market, they are very difficult to the consumer solution because to the different applications, you need to provide different solutions to make the best performance and best CTO. For some ARM, especially for some HPC and ARM device-talking applications, I think SoundX2 will have better performance than Z80X. For what? For the web searching and some storage side. Sometimes the ARM solution will have better CTO. TCO for that. So Google should buy millions. I'm joking, but I don't want to know what's the secret or anything. They are very interested in that. The cloud companies are very attentive to what's going on here. I cannot tell you too much secrets, but I can tell you Big 7, a lot of Big 7, they have already started to do the validation with the ARM solution. So that means they've been testing together with you in the last year. They've been testing, right? But potentially now next month. So I'm just to have an idea. The company like Google, Facebook, Amazon, do they buy stuff from Gigabyte and themselves, the server, usually? I think they do both. Because you know they are a very big company and as I said, they have a lot of different applications. It's depending on the applications and they want to get the best performance and the best TCO. And for this kind of view, they want to try a lot of different systems and currently our system can hit some application for them. And one more thing is to the future view, I think there will be a cruise for the ARM architecture for the 2020. So you know in servers, there will be a very long leader time to do this kind of validation in the temporary stage. So we provide our first very early sample to Zen to make the leader time shorter and then they can have the real order after they do the validation. So how was this year, last year, how was it for you? It was very busy, very fun. How did it, challenging? I should say that in this way. For the last year, we do the sound x2 solution and other solutions for the sampling and for the sampling stage, engineering and RD resource will be very busy. So for next month, we hope our sales will be very busy. So next month, you hope you can go on holiday? I'm joking. But maybe, are you have plans? I'm joking, you don't have to say. In the last year, maybe there was no time for holiday? As I said, there is a very hot topic and a hot case now in different regions. I think I'm busy in this month and next month I would go different regions to talk about that, especially the ARM solution. And you're from Japan, right? Yeah, I'm from Japan. Originally. So that means you can have very good conversations with the potential Japanese customers, right? Yeah, I can speak Japanese, so I talk with SoftBank, you know, there's the owner of the ARM solution now. So they are very excited about this SoftBank, I'm sure. Yeah, they are very excited about that, I think. Maybe that's why they buy ARM, because of this company? Yeah, they see the growth, they see the potential of the ARM solution and as I said, especially in Asia, Asia now is big green and for Asia, I think ARM will be very hot solution in future. So, and of course, when I did the interview yesterday with the Cadmium, they said that in the other parts, because it's similar, kind of like the same price, but the CPU, potentially, they will have a competing lower price, potentially. That was going to be exciting, but maybe it's also something where you stand by, it's up to them how they're going to do the business, they're going to give a discount or no discount. Sometimes Intel gives discount sometimes, AMD maybe will try to do, it's interesting what's going to happen. I know about the pricing, a little competition, but I can tell that for the MSRP, the SoundX2 solution will be very competitive than Xeon 86, whatever Intel solution and AMD solution. And one more thing is, we are maybe the biggest partner with Cavian side for their SoundX solution. So, their pricing is very aggressive and our pricing will be more aggressive because they will provide special price for us. We will have the Gigabyte price. And Gigabyte is selling millions of servers every year? Yeah. In total? Yeah. So, you have all the infrastructure, you know exactly how Taiwan works and all how to make everything in China and everything. That's for them. Our business style is very feasible, so we not only just provide our server to the AMD user, we also can provide our server to some SI customers, distributors, so customers can choose what they want. So, we can do this kind of business. We can do the total solution for what we call level 10. And you see this is level six without memory, CPU and hard disk. And even for motherboard, you can see we have our motherboard right now. But these motherboards are AMD. But do you have motherboards like this for the KVM? We were, but for the beginning, usually we do the optimized design for system because now we are more focused on system because it has a high revenue. And one more thing is, if we do the optimized design for the system, then it can provide the best performance and optimize for different applications. But for motherboard, it's not for AMD user, it's just for some SI customers. They need to combine the motherboard to their services for their solutions. Wouldn't it be cool to have maybe some kind of development boards, something like that? Would you consider doing that or is it not your job to do that where maybe something kind of affordable for developers to get Thunder X2? Yup. That's also in our plan so you see we do a lot of servers. Now we also have our priority and the trade-off for the resource. And in the beginning, we just do the optimized system for customers. And if customer want like a research company or some engineering professional guys, if they want to standard motherboard to do the validation with the motherboard, we also can talk with them. And if I just turn around, so this one is not for next month. Is it a little bit later? When is it available for this one? Kick here. So this is not for next month? This is not for next month. So September? I think in Q3 maybe. Q3. So what's the still remaining challenge here? The challenge, we have two things. Now we are under starting and under construction. One thing is, you know for this server, there will be shared power and the shared thermal solution in this 2U. It's very high density you know. And for this case, in this case you see the CPU will have very high power consumption. Then we must solve much more difficulties than this 1U2U solution because it has more CPUs inside this 2U4 factor. Totally 8 CPUs. So the thermal solution is the concern we are now under validation. And one more thing is, for our internal validation we need more CPU than these two of case. You know, if for the 1U2U we just have two CPUs then one system can be validated. But for this one need 8. So we are talking with Kevin about to provide more CPU samples for us. It will be fast enough it will be faster than our little time of the validation. So it depends on the CPU numbers and our thermal solution study status. Then we can decide the run timing of this system. But when I did the video with Kevin yesterday he said they have 40 different SKUs. 40 different available SKUs. Are they all compatible with all these three? Or you only choose specific one only for these products? Or you only take the highest one. What do you do? If you say the coverage we have the biggest coverage of SoundX2 because we do this this is our Gigabyte standard channel product. 1U2U and 2U4U. That will be the mainstream product in the market. But we also do the CRB product for Kevin. We make their CRBs for 1U2U even the workstation CRB for them. So they do a lot of applications we know that and we do a lot of the coverage I think we are the biggest partnership to do SoundX2. What is CRB? CRB is reference board. Is this the 32 core 2.5 GHz or is it a 16 core 1.6 GHz or something like that? Which one is this? This time Kevin provides almost more than 40 skills of CPUs. They have 32 cores 28 cores and even the 16 cores. So they want to have the same coverage of the Yingtiao Perley solution like Skylake and the AMD APEC Napper solution. Because you know even for the server applications they will have HPC, enterprise mainstream storage entry. Different segmentation in the server market. You cannot just have one skill to do so you need to have the different TCO skill and the performance skill to fulfill different requirements. So some of them have a different thing on the SoC for networking or different thing for another market or the similar SoC design is just a different clock speed only. No this time they change the strategy. In the SoundX1 generation they have different application skill they just use the same strategy like the Yingtiao Perley and the AMD APEC just to provide different performance and application usage for the skill. So there is no big category between the CPU skill. So the design of the SoC is similar just different clock speed. Yeah it's the pin-to-pin design. And the cooling will be different if it's a 16 core 1.6GHz or 32. The TDP were different. So different. And here it's easy for you to fit there's not so much space to fit the fan or the cooler. The fan is too high fan something like this but for in our 2U Fono you can see this one. I cannot open the cover. So in our 2U Fono we have an advantage for the thermal solution because we use the double fan wall. Use the double fan wall then we can provide more performance for the thermal solution. Compare with our competitors they only have the single fan wall but we provide the double fan wall. So the fan but how about the heatsink? The heatsink will be the 1U high heatsink. And that's just fine. The 1U height. 1U high heatsink for both every node and the difficulty is you know they were sharing the 2U double fan solution. So that means we can provide all the skill supported in this server. But one thing is for the full spake usually we will have some limitation of the CPU skill but if customer don't want to use the full spake that means they maybe don't need too much real I.O. or they don't want full storage then I think all the skill can be supported in this server for some next year. And you could also potentially work with startups, new companies that are doing new hosting solutions that want to provide bare metal or companies like packets maybe they will be very happy You have all kinds of share. I cannot mention the name. The whole range right? Yeah very big range. From the distributor S.I. research company and the cloud company and the big 7. University even yes they are very interested in arm solution they are now studying. From China to USA to everywhere. All regions worldwide. I'm in charge of worldwide product here. For all the different solutions. Something happening in Africa are they having some arm servers because maybe they don't have one? It's a good question. For the African maybe they would save some power maybe not here. Depends. So this has been a very busy year and maybe we'll see how much it gets next year. I hope it gets revolutionary. I've been doing arm server video for 8 years. So this is the moment I've been waiting for right? Thank you for your question. Before it was kind of like testing or slow. I think it's very important for us to do this kind of things. In this market we need some pioneer to do different solutions because I'm basically for the car company in Paris many years ago and you know in cars different brands like Toyota Volkswagen and GM but for server there's no option and no choices. I think it's a little boring. Maybe we need to do some interesting job to engage a lot of different solutions for different applications. So we are the pioneer to do so. Gigabyte is the pioneer to do so. We are the leading pioneer in arm solution and we provide flexible solution and different solution to different applications and customers. Hope they can have second source, third source. I don't want to say anything bad about Intel. You know that's not they are good. I don't want to say that right but they were supposed to have 10 nanometer at Computex but somehow it didn't work. It doesn't work. Maybe arm has already shown 10 nanometer in the phone so I don't want to leak information what's happening next. That's not the confidential information but I don't want to announce this information instead of Intel CEO but actually Intel CEO has already announced that because they were still used 14 nanometer technology but they were transferred to 7 nanometer technology. That's a very checkpoint. Maybe that's a lock for arm because they are risk instruction set. Maybe it's easier to put in a lower nanometer than the complex Intel architecture. Maybe they have some issues with the small nanometer so maybe that gives an opportunity with what happened next year and then year after that. Maybe there will be some opportunities for arm to do something very interesting. I just can say that from the technical view, 7 nanometer will be good performance than 40 nanometers. So that's the truth and so we can see what will happen in next year and that's also one potential for arm solution. And I'm sure they're very busy with all kinds of strategy meetings and all the R&D guys are getting new ideas every day or every week with like oh we should do this and like oh great let's do it. And then they have to design a new chip. Yeah you know now more and more focus is the AI solution cloud. Everything is AI or cloud but what's the AI and cloud? The hardware will be server right? So server market really change very fast and we have our strategy meeting almost every week and sometimes every week we need change the decision or change the direction for the new plan because it's really change a lot and a lot of new technology now coming to the market and we should choose and make the trade of how to use this kind of new technology. And I went to the NVIDIA press conference and you know Jensen Wang he has a very cool style and he always talks very cool about his GPU and it works with GPU right? Yeah we are now working on it. You can plug it in? Yeah currently we also I think both Kavi and us and we are now talking with GPU company such as NVIDIA and AMD but currently the final certification is not finalized in Kavi and we are still on the X2 solution but we will continue to talk with these two company to make ARM server also can be used for AI and HPC. Because last time we did a video you told me that you can connect GPUs over here. Before the design that's why we talk with NVIDIA and AMD because this 2U server for this portion we can support totally 3GPU card there but once you know GPU is not the VGA card you need official certification from vendors so NVIDIA and AMD must do the total test for the GPU card validation and of that they will provide one official approval for this kind of solution so we are now talking about official certification about the ARM solution. Because when NVIDIA showed their Volta 100 it looks like a beautiful GPU Volta 100 can it work with KVM in theory or is it only the GTX 1080? What kind of GPU are you talking about? I am talking about the PCI type not the NVLink NVLink I think still will be using Intel solution now for now it's... SSM2 will be only supported on Intel PURLY and Grand Prix solution currently and for both AMD solution and KVM solution we are working with NVIDIA to talk about how to support NVLink technology but currently we haven't got the official approval but at the NVIDIA press conference they have their Xavi ARM chip together with the V100 and the same PCB so that's the NVLink you talk about right or what's called the faster link faster link it's the different proprietary card it's not used PCI slide it will be have the black baseball under the SSM2 module so it's a special form factor and something new a new category in the last few months I talk on my website is the machine learning the AI accelerators and all that it's so much happening here like Huawei is doing a design ARM is doing a design everybody want to do a new design that's going to be exciting what will happen with this next year in the next few months right? as you mentioned the new technology become more and more and for the AI cloud, HPC and even storage there's a lot of new technology now coming to the market so sometimes it's difficult to make trade off but we had leading position in the server development so we have very good partnership with them and we talk with each other to decide the strategy even the very interesting interview I did with Dr. Shunpei Yamakazi from SEL in Japan and he is an inventor of IGSO and he said IGSO if you can put this in a CPU it will be 100 times less power consumption so it will be exciting future a lot of ideas even in China they use some chipset ARM chipset to do new GPU card I heard so it really changed a lot so it's very fast we are now put our eyes to a lot of new technology and make the final decision to trade off