The biggest tech war, Intel and NVIDIA battle in the AI market
In 2019, a statistical report revealed that NVIDIA has absolute control in the AI market. According to the latest statistics, NVIDIA’s market share in the Cloud Services, Artificial Intelligence and Data Centers sector has exceeded 97% .. which is a frightening percentage. Extremely.
The reason is that it is a frightening percentage that NVIDIA achieved this by using only graphics processors (GPUs) … and in this it faces competing GPUs from AMD, and competing artificial intelligence accelerators TPUs from Google, Intel and Xilinix .. in addition to CPUs from Intel and AMD.
In addition to that, NVIDIA graphics processors now work on more than one architecture for CPUs, of course they support X86 architecture from Intel and AMD, and support IBM PowerPC architecture, in addition to ARM architecture .. which means that NVIDIA graphics processors work under any processor Central CPU is available in the market. This cannot be said about the rest of the other artificial intelligence accelerators .. whether from Google, AMD, Intel, or Xilinix.
NVIDIA competes in this sector with a number of solutions .. The first is the Tesla V100S robust card, which is the most powerful option available in the computing and artificial intelligence market. The card comes with Volta architecture and provides the user with global computing power equal to 130 billion arithmetic TFLOPs (130TFLOPs) with 16-bit (FP16) … with a Memory Bandwidth data exchange rate exceeding 1.1TB per second.
Then there is the Tesla T4 card, which also offers 130 billion tightly packed 8-bit AI computational TOPs (130TOPs) (INT8) … with 65 billion (65TFLOPs) for 16-bit tight general arithmetic (FP16).
There is also a Titan RTX card, which offers 260 billion TOPs, with 8-bit AI (INT8) which is a very frightening rate! With 130 billion TFLOPs (130TFLOPs) tightly 16-bit global arithmetic (FP16).
All of these are central solutions for Data Servers and Super Computers, .. Even in the ordinary consumer market, RTX cards represented a strong spearhead with which the company invaded the market for artificial intelligence in games.
But NVIDIA did not stop at this front only, but also pursued peripheral uses, such as mobile devices, self-driving cars, and autonomous machines, in addition to 5G network accelerators and the end device market that uses Edge AI, such as surveillance, imaging, and automated factories … etc. . And for all this, NVIDIA introduces the Tegra microchip … the latest of which is the Xavier chip, which contains 512 graphic cores from Volta with 64 Tensor Core artificial intelligence arrays … to provide 11 billion arithmetic operations 11TFLOPs tightly (FP16) and 32 billion AI operations tightly (INT8).
On top of all that, NVIDIA has the most powerful software platform for public computing and deep learning, which is the CUDA platform, which is widely used in most levels of computing, from universities and research centers, to small and large companies, Super Computers and Data Centers, the platform has become the first choice For everyone because of its ease of use, its absence of errors and defects, in addition to its extreme efficiency. It works on the vast majority of laptops, desktops, personal PCs, etc., as well as all computing architectures: PowerPC, X86, ARM, etc.
With these solutions, NVIDIA has become the benchmark by which computing power is measured in the fields of artificial intelligence and supercomputing. NVIDIA has comfortably swept these markets and became the first dominant and the primary criterion without the slightest resistance.
But this is about to change. Intel has seen NVIDIA dominate these markets, and feel the negative impact of this on them. Every NVIDIA graphics card replaces a number of Intel core processors, all of these are potential profits that Intel will lose and NVIDIA will gain. Intel has seen this and is determined to change it at any cost. Intel has deemed this to be its fateful battle to win. She wins it knowing full well that she has only the CPUs that NVIDIA GPUs are crushing on as easily as a bulldozer traverses the eggshell.
And because the battles are only decided by armies, Intel prepared to form its own army, bought several emerging companies in the field of artificial intelligence, and spent billions of dollars to have its own army of processors and custom accelerators that directed them like bayonets towards the heart of NVIDIA .. The biggest contemporary technology battle is not being fought In the world of CPUs, not GPUs, but AI. Investments worth $ 35 billion and five companies led by Intel .. Intel has mobilized an army to confront NVIDIA in the artificial intelligence market … and is preparing for a fierce battle in 2020 and beyond.
This behavior reminds us of what Intel did with its computing cards project (cards containing arrays of microprocessors), which Intel called Xeon Phi, with which Intel tried to compete with NVIDIA but failed to achieve any success, then quickly closed it in favor of Focus I have to buy other companies with more successful accelerators. This didn’t become Intel lately.
Intel also has Altera, which bought it for $ 17 billion, which provides private FPGA computing arrays .. Intel has directed these accelerators to the AI market to compete with NVIDIA, Intel’s most powerful accelerator is Stratix 10, which provides 136 million artificial intelligence processes per watt (136GOPs / W) .. but he loses hard against Xilinx’s most popular accelerator called Virtex UltraScale, which comes with a capacity of 277 million artificial intelligence operations per watt (277 GOPs / W) .. But both of them lose heavily in front of the Tesla T4 card, which provides 1850 million artificial intelligence operations Per watt (1850GOPs / W) !!
Xilinx is preparing to update its FPGAs for the new Versal platform, which will allow it to achieve an average of 176 billion AI processes, TOPs, tightly 8-bit (INT8) … (176TOPs) with its Versal AI Core accelerators, achieving an average of 203 billion operations. Artificial Intelligence (203TOPs) … with its Versal Premium accelerators, which are rates – for both accelerators – clearly outperforming Intel, but both are less than NVIDIA’s Titan RTX cards with 260 billion operations as mentioned above.
But all this talk about Intel’s non-competing solutions is not important. The big problem in all of these solutions is that they are very late. Jumping over all this with its new generation, which it has already announced .. On the side of separate cards, the new generation of NVIDIA cards is expected to provide the equivalent of 800 billion artificial intelligence 800TOPs .. which is an increase also estimated at seven times the previous generation Volta, which It means that NVIDIA will make a pivotal leap that will put it on the front lines again in front of all the concerted efforts of Intel.
As for the microchips, NVIDIA has officially announced the new generation of it already, which is the generation that will follow the Xavier mini chip, and it will be called the Orin mini chip, which will come with a frightening increase in artificial intelligence performance up to seven times the capacity of Xavier, to give Orin nearly 200 One billion artificial intelligence operations (200TOPs). Which means that the battle will never be in Intel’s favor, nor will the battle be in the interest of Intel, even the rest of the competitors, such as AMD, Xilinx and Google. All this, and NVIDIA has a qualitative edge in software and tools for indoctrination and learning thanks to the CUDA platform.