The event was opened by Alexander Bratkov, Graviton's product trainer. Our company's portfolio includes network and server equipment, software and hardware complexes that include data storage systems and a software-defined environment, the so-called hyperconvergence. In addition to heavy-duty solutions, we also have peripherals. Our peripherals include printed products and monitors. There are client solutions – all-in-one PCs, laptops, PCs, mini PCs, and workstations.
Today we will talk about server solutions. So, achievements of 2024.
These are the fourth and fifth generations of Intel Xeon Scalable, and server solutions based on these processors are in the register of the Ministry of Industry and Trade. These are special industry products related to medicine, namely the all-in-one. This is a completely channel product. It has special color rendering, special screen coating, and brightness. This is printing equipment. Also industrial network equipment. At the same time, this is the first full year of the distribution business. We have doubled the number of employees in R&D, that is, we have more engineers and more programmers. We have increased the groups that monitor quality control at contract sites, of which we currently have about 10.
We have a project for the industrial design of client solutions "AUTHOR".
We do not take a standard catalog solution from some Chinese developer, we independently develop our own solution from scratch by our industrial designers. We have a new website with an online configurator and personal account.
Now briefly about marketing related to "Graviton".
This is a new structured product catalog, convenient navigation through sections, flexible search, and fast response and page loading. The UI was redesigned, the UX was redesigned. This is adaptation for mobile devices.
And, if we talk about the functional parts, this is the launch of a partner's personal account, where you can view the design of projects, get access to the configurator, certificates, personal manager, price list, and information about the balances in the warehouse.
The company "Graviton" is a manufacturer that makes registered equipment.
We really do everything ourselves, this is a full-cycle development from the purchase of a reference to writing documentation, editing documentation, preparing all the necessary documents, then the purchase of a component base for our products takes place, after that the component base together with the documentation goes to the contract site. The contract site is located strictly on the territory of the Russian Federation. That is, surface mounting takes place there.
Currently, we have three main levels of server equipment.
The first, entry level. This is the first and second generation Intel Xeon Scalable Gen 2. Servers still meet all the needs of government agencies. The middle level is represented by the third generation Intel Xeon Scalable and, accordingly, AMD EPYC. The upper level is represented by the fourth and fifth generations. We even have a prototype of the sixth generation. The only thing is, there are nuances regarding soldering in Russia. So far, let's just say, there are questions about this.
Let's consider the entry level.
First and second generation Intel Xeon Scalable Gen 2. Our solutions are presented in four chassis. Two of them are a 1U solution with LFF and SFDisc, and two solutions are 2U with LFF and SFDisc. These solutions are built on the basis of two motherboards. The first motherboard is a single-socket "Taiga", 8 DIMM, everything is quite simple, a simple network, there are two full-fledged X16, PCI Express slots. And the second option is "Tundra". It comes in a slightly larger form factor, this is Extended TX, here there are already two processors, and it is possible to install 16 memory modules. Here there are already five PCI Express X16 physical ones. The grid remains the same. The amount of RAM increases due to the number of expansion slots. Therefore, it is possible to install more cards, if we are talking about a 2U solution. These solutions have a limitation on the number of expansion cards installed.
If we talk about the application of these products, this is the entry level, this is small and medium-sized businesses, small government agencies. This is a virtualization solution that provides stable performance for simple resource-intensive applications. Such as simple data analytics, entry-level work with artificial intelligence. Most likely, this concerns face recognition, nothing more. That is, this is not training. There are special APU servers for such solutions, and there are servers on the fourth and fifth generations of processors. Accordingly, a fairly simple scalability of solutions that is suitable for most corporate and government infrastructures and IT systems that do not require any very deep knowledge and special increased capacities.
What is worth noting about the entry level. There are three main levels, and I would say that there is also a zero level. The zero level is runrate systems. The trick of these systems is that, unlike channel products, they are in the warehouse already in a pre-assembled state, ready for shipment. And the delivery times are minimal. The cost of these solutions is also minimal, because the servers are priced at self-assembly. At the same time, this is a register. But any runrate has its limitations. And there are also a number of limitations that affect availability and cost. There is no NVME there, only SAS drives can be installed. There are limitations on power supplies and RAM. This limitation is not physical, but it will be so upon shipment. That is, you can't assemble anything interesting in runrate. You simply won't be able to do it.
Next we have the main level. These are the most popular positions.
The main level is divided into two categories. These are standard Extended TTX motherboards and one proprietary version. This includes solutions on the third generation Intel Xeon Scalable Gen 3 and AMD EPYC. AMD EPYC has a distinctive feature, which is a larger number of cores per processor, and a larger number of threads. That is, from an economic point of view, this will be more interesting than, for example, the same Intel. Well, plus again, the solution is quite well protected in the competition.
Also about AMD EPYC, it can be said that, unlike Intel, it has a larger number of PCI Express lines per processor, which allows you to make assemblies with a larger number of expansion cards, or install a larger number of NVME drives.
All mandatory indicators are power systems according to the N + 1 redundancy scheme. That is, if one fan fails, we can easily replace it "hot", and the system does not stop working, and continues to work under the same loads, at the same pace.
The second part of the main level is just a proprietary motherboard.
The motherboard is called "Valdai". Here it will be more interesting both in terms of layout and in terms of the capabilities of this motherboard. First of all, here is already a proprietary form factor. There are more cases and different back panels for these cases. There are more expansion cards that you can install. If I'm not mistaken, you can install 6, 7 and even 8 expansion cards here.
Wide expansion capabilities with support for 2 PCIe 4.0x16 and 2 PCIe 4.0x8 allows you to install network cards, GPUs and other expansions for virtualization and cloud computing tasks.
Power management with redundant power supplies (from 800 W to 2000 W) and pull-out rails with CMA ensure stable operation in critical environments.
Here there is already Dual BIOS, microCD. There are 32 DIMMs, which is enough, you can assemble a server very well, it will be quite dense. There is a limitation - this is 270 Watts for the processor. This solution, probably, can already be used for artificial intelligence as an entry-level solution. Because here you can, if you wish, put two dual-slot full-size video cards.
Next comes a slightly more advanced solution for GPU.
If you have engineers who work in computer-aided design systems, or if you have designers or engineers who work with fairly strong 3D models. A feature of this solution is that it was originally designed and positioned as a GPU server. The trick is that you can already put three dual-slot top video cards here, or it can be three top Quadro. At the moment, these servers are assembled with two types of motherboards.
The first motherboard is "Tundra", dual-socket, and the second version is "Arctic" on AMD. The case is placed both on the table, as a tower, and in a 19-inch rack. It has lids that open from above and below, and a special article of rails is easily installed there. That is, it can be easily put in a cabinet. According to the disk subsystem, here we have two backplanes, four disks each and two backplane options. The first is 8 NVME, and the second option is 8xSAS/SATA. The baskets by default are 3.5 inches, but with the possibility of mounting 2.5-inch disks. That is, a good storage system can also be implemented here. Another feature of this server is that it has a special power distribution board that allows you not to accumulate a huge bunch of wires, but to connect only those cables to those video cards that you need and that will be used. This has a positive effect on the cooling system.
Next we have information from the category "new". At the moment, this is not in the register of the Ministry of Industry and Trade. That is, this is still a developed product.
A product that meets modern requests, modern requirements. Namely, this is work with artificial intelligence, with large databases. A feature of this product is that it is a first category protection, namely a server motherboard on "Baikal" server processors. Support for up to 8 GPUs. And all this in a 4U case with CRPS power supplies. There are two CRPS power supplies with four modules each. Two modules in each.
And now the most interesting thing. What to do before purchasing a product? How to make sure that you really get the product that you want, with which you will be happy to exist in the future, and you will not experience any problems with this product?
For this we have "Testing as a service". It implies the following. You provide mandatory project documentation, provide a technical task that describes the «hardware» that you plan to use. After that, it goes along the following path of development. We either take the hardware that we have in the demo fund and assemble it from it, or, if it is something unique, then we can assemble it ourselves from our warehouse and provide it for testing.
We can provide this testing in two ways. The first way: transferring this hardware to our circuit. And the second option is remote connection. For example, if you are testing hyperconvergence based on special software, then, of course, only remote connection is possible here, because it will be quite difficult to drag a cabinet with four or five servers and three or four switches, of a decent cost, for example, across Russia. Here, of course, it is better to use remote access.
We also have system testing and certification of specific specifications, if this is necessary for the customer. This is usually necessary when you are communicating with a large customer who has their own requirements and their own approvals that cannot be bypassed. You either present your testing methodology to our engineers, specialists, and they conduct testing, or we develop this methodology ourselves, or request it from vendors.
Well, and you can't ignore, of course, our standard warranty packages. We currently have four of them. These are Standard, Standard Plus, Extended and Extended Plus.
You can see the differences on the slide. First of all, this is the response time. If we are talking about simple solutions, then it is 24 hours. If we are talking about extended options, then it is 4 hours. There is also an engineer's visit with spare parts, there is a repair time in working days. If we are talking about Standard, then it is 22 working days. If we are talking about Extended Plus, then it is one working day. If we are talking about some hard-to-reach place, for example, about Sakhalin or Kamchatka, then you need to understand that the repair time will in no way be one working day. Here you will need to discuss all these nuances with the service workers in advance. Either it will be a specially trained person on site who lives nearby, or something else. All this can be discussed, and here lies a certain uniqueness in our entire packages. It lies in the fact that they are not finalized. You can always come to us and say what exactly you would like to add or change. And we can go for it. Of course, this may lead to an increase in price. This, of course, will be the time that you spend on agreeing and working out these points, because this should somehow be fixed in the documents legally in the future, but we can do it. We have already had similar cases, there have been similar cases with the Ministry of Internal Affairs and with some large companies, when it was necessary to extend the term and add unique features.
On the next slide - AMD EPYC 9005.
There are quite a large number of cores, and even more threads. A large number of PCI Express lines. The same two M2 keys directly on the motherboard itself. The cases that we use in the fourth and fifth generations. Here are 4 expansion slots. Here you can install up to two full-fledged dual-slot video cards, whether it's 4090 or H100, or the top Quadro. And if we started talking about EPYC 9005, here, unfortunately, there is no slide about the 4th and 5th generation Intention Scalable, which is currently both in the register and in our warehouse. The same cases are used there. These are servers based on the "Ural" motherboard. This is a large amount of RAM, these are high-performance processors. This is again the possibility of installing up to 8 and even up to 9 cards, plus the OCP slot is included there. In the fourth and fifth generations, we can install from two to three dual-slot full-size video cards in the cases.
Let's move on to data storage systems. At the moment, we have two main types of data storage systems.
The first is PAK and DSS, the second option is storage servers. Both options are in the register. If we are talking about PAK, then this is necessarily software plus hardware. That is, software is an integral part of the hardware here, which is also in the register. There are three software options: Baum, Raidix and Argo. And this is what distinguishes us from our competitors, because we do not impose any specific product and we do not have any fixed configuration. That is, you can choose a configuration for your own budget and the performance that you need, and you can choose the software that you need.
If we are talking about storage servers, it's all the same, only without software.
For example, you need to expand, and you need to expand to registered servers. But you use some of your own software, for example, open-source, you like it, it suits you in terms of level, you have been living with it for a long time, and in general you don't mind using it in the future.
This is where storage servers will come in handy. Since they do not have a software component, you can install anything on them. And, of course, it is advisable to see in advance, if there are pilots, whether this software is suitable for our servers, whether everything will work well.
Data storage systems are divided into two types.
These are single-controller and dual-controller solutions.
The shelves are essentially compatible with almost all. Our shelves are not in the register.
Illustrations provided by the press service of GRAVITON