The Chinese and Russians produce their processors and graphics – let’s check the most interesting proposals

As you well know, most of the electronic equipment is produced in China, which gives a very large technological base to the local companies. Although they cannot directly use patents and solutions from companies such as Intel, AMD or NVIDIA, they copy them and use them to a greater or lesser extent to produce their own graphics cards or processors. While in Russia they focus mainly on creating a powerful processor, in China they have no qualms about competing in the field of both CPU and GPU, as well as extensive motherboards. Although the latter look cheesy, it must be remembered that without a dedicated socket it would be difficult to embed any processors from the three companies on boards intended for Intel or AMD.

The Russian Elbrus processor did not live up to expectations

Let’s start with the fact that the Russians, thanks to the government that reigns there now, want to function fully independently in the world, as a great power that everyone must take into account. To this end, they guard not only their natural resources, reluctantly (and dearly) sharing them with their European neighbors, but also would like to keep technological and equipment capabilities in check. Many years ago, work was started on Russian microprocessors, which, in principle, would have an equal fight with Intel or AMD units. The most advanced chip is produced by MCST (Moscow Center of SPARC Technologies) processor called Elbrus-8S, which is a natural evolution of the 4S chip. The number in the name is not accidental. It means the number of cores contained in the CPU. It is worth knowing that the first MCST processor saw the light of day in 2008. It was proudly and convincingly called Elbrus 2000. Its production was assisted by TSMC factories – the largest Taiwanese supplier of microcircuits in the world.

Elbrus 2000 did not impress in any way. The core developed a maximum of 300 MHz, which in 2008 caused a smile of pity at best. It only had 256 Kb of second level cache and 75.8 million transistors as well as one core. His successors looked a bit better, although they were far from the achievements of Intel or AMD. The Russians, however, did not lay down their arms. In 2015, they started work on the Elbrus-8S system, which was intended to finally engage in an equal fight with the American proposals. After five long years of research and design, mass production began. The processor can only be used on dedicated server boards that can accommodate up to 32 such units. The system is equipped with 8 cores (and 8 threads) operating at 1.3 GHz and made in 28 nm lithography. It has 16 MB of third level cache and supports DDR3 RAM with speeds up to 1600 MHz.

For the needs of one of the largest Russian banks, Sber, an even more powerful variant was prepared, which accelerated to 1.5 GHz and supported DDR4 RAM with a speed of up to 2400 MHz (in 2021). Unfortunately, the hardships were for nothing, as the processor did not exceed the demanding expectations. As many as 44 different functional parameters were tested and in each of them Elbrus was losing to Intel’s proposals, sometimes reaching up to 26 times higher response time. It is a really crushing comparison. It is true that SberTech, the bank’s technological unit responsible for implementing and keeping everything in check, was very surprised that the platform prepared by the Russians works at all. In addition, he is managing somehow. At the moment, its implementation does not make any sense, but our eastern neighbors are not going to lay down their arms. The Elbrus-16S model has already been announced, produced with the use of 16 nm lithography, equipped with 16 cores and accelerating to 2 GHz. Although it’s a long way from MCST to engage in an equal fight with Intel or AMD, in the next few years they may catch up with what the competition offered in 2010.

China is doing boldly in the graphics card market – take a look at this

What is worth knowing? In the last few years, Chinese companies have started to produce x86 processors, DRAMs or even NAND chips, which may soon establish an equal fight with the products of the largest competitors. They could not let go of graphics cards, because in the end this is probably the biggest money, and at the same time the greatest opportunities. In September last year, the Chinese company Jingjia Micro tested GPUs with power similar to the NVIDIA GTX 1050 or AMD RX 560 models and the NVIDIA GTX 1080 or AMD RX VEGA 64 models. The cards reportedly performed well, although they had obvious driver problems. The most important thing, however, is that local, huge corporations are already able to provide products technologically advanced enough to constitute real competition for the tycoons. Innosilicon and Xiandong are also carrying out advanced work. They made a joint effort to deliver to the local market a system called Fenghua 1. Interestingly, the card supports many advanced technologies, including such a common API as DirectX (the first Chinese card fully compatible with Microsoft’s proposal), Vulkan or OpenGL. Since it is a product intended for the consumer market, it also has the appropriate connectors – HDMI 2.1, DisplayPort 1.4 and DVI.

The card is equipped with Micron GDDR6X memory, and the graphics processor is based on one of the Innosilicon Innolink chipsets. Moreover, the chip complies with the latest PCI-E 4.0 standard, works in Windows, Linux and Android environments, and its theoretical performance in FP32 is approx. 6 TFLOPs. By comparison, the NVIDIA GTX 1660 SUPER graphics card has just over 5 TFLOPs. However, I believe that this is still a great achievement of Chinese companies, because referring to such giants as AMD or NVIDIA is quite an art. After all, Intel will join this race in a moment. Fenghua 1 should be available in stores in late 2022 or early 2023, and may even gain in performance by then. The first force show took place in December and featured the smooth running mobile demo of the GFXBench T-Rex. Unfortunately, not very reliable, and without a frame-per-second counter. Systems with 4, 8 and 16GB of VRAM will be available on the market.

Equally interesting are the reports from the Chinese company Shanghai Tianshu, which last year planned to start the production of graphics cards made with the use of 7-nanometer architecture. The schedule is intended, unfortunately, only for the tasks of workstations and servers. A processor with the working name “Big Island” would be responsible for generating the image, which has a theoretical performance of 37 TFLOPS in FP32 at a TDP of 300W. These are truly stunning results. All the more so by comparing them with the competition. The AMD Instinct M100 pulls out 23.1 TFLOPS, and the NVIDIA A100 stops at 19.6 TFLOPS, although the Tensor cores that can jump to the level of 156 TFLOPS make a huge difference there. All this makes it possible to look at the local suppliers with great optimism. Innosilicon announced in December more graphics chips, this time intended for players, which will reach up to 10 TFLOPs (shown in the photo above). I do not think that in the next few years they will even come close to the AMD or NVIDIA proposition, but their presence on the market may affect the pricing policy or the approach to the customer among American corporations. At the moment, we can only wait for the next fruits of Russian and Chinese engineers.


About Alex Marcell

He likes dogs, pizza and popcorn. Already a fanboy of Nintendo and Sony, but today throws anything. He has collaborated on sites and magazines such as GameBlast, Nintendo World, Hero and Portal Pop, but today is dedicated exclusively to Spark Chronicles.

Check Also

Motorola Frontier promises to be a premium smartphone

It seems that Motorola wants to expand its smartphone offer with a model that is …

Leave a Reply

Your email address will not be published. Required fields are marked *