Fujitsu’s New Chip Beats the World’s Fastest Supercomputer

Author By Vytautas Palubeckas
Updated :Dec 25, 2023
Why You Trust Glozine

Our editors independently research, test, and recommend the best products; we may receive commissions on purchases made from our chosen links. You can learn more about our rating and review process here.

If you are keen on following the history of computing, you may have noticed the main issue has been computer speed. From one generation to the next, speed has been the main factor followed by others like size, storage capacity, etc.

The world’s fastest computer is clocked at about 33.86 petaflops (A petaflop is equivalent to one thousand million floating point calculations per second).

Fujistu’s new chip, SPARC64 XIfx processor can be used in supercomputers capable of delivering a speed of up to 100 petaflops. This is roughly three times the current fastest computer.

This new chip is the successor to SPARC64 IXfx currently used in PrimeHPC FX10 servers which are sold by Fujitsu. These are high-speed high-performance servers.

SPARC64 XIfx processor will deliver performance of over 1 teraflop, about four times that of SPARC64 IXfx processor.

More on this new chip is expected from Fujitsu in the Hot Chips conference which will take place next week in Cupertino.

As science and technology develops, many manufactures are looking to design supercomputers that can help in solving complex problems in areas like science, economics, national security, engineering, health, etc. They have to do this in order to remain competitive.

Researchers in meteorology, space, energy, biological sciences, nanotechnology, statistics, artificial intelligence, etc also need supercomputers with high speeds to help manipulate and analyse the huge amount of data they have.

Japan is known the world over in delivering high speed chips. The new chip is two generations ahead of the SPARC64 VIIIfx processor, an 8 core processor used in K computer. This is the fourth-fastest computer in the world. It is located in Japan and is able to deliver 11.28 petaflops of peak performance.

According to Fujistu, this new chip boasts an array of memory and throughput technologies plus 32 processor cores which ensure data transfer between servers and within the system is fluid. It will fit into a 2U chassis, giving the server a total of 384 main cores.

In addition to the 32 main cores, each chip will have two assistant cores whose functions are not yet clear.

It will have an on-board support for Hybrid Memory Cube as one of the throughput improvements. This is an upgrade from DRAM in today’s servers. The Hybrid Memory Cube has memory modules that are stacked on top of each other rather than lying flat next to each other on a motherboard.

Hybrid Memory Cube is 70 more power-efficient that the conventional DDR3 DRAM modules and provides 15 times more throughput; this is according the to the memory maker, Micron.

Its new Tofu 2 interconnect will enable the CPU and other components to communicate with each other efficiently. This is a great improvement on data transfer.

The Tofu 2 interconnect has a maximum speed of 12.5Gbps, almost two-and-a-half the speed Tofu 1 used in K computer.

The servers having SPARC64 XIfx will also support optical interfaces. This will allow these servers to network using fiber optics which are faster than copper cables.

It is interesting to note that that first computer, the ENIAC had a clock speed of only 100,000 cycles per second. This show how computer technology has developed over the years.

Vytautas Palubeckas