Intel Corp. today debuted two lines of processors designed to power workstations, high-end Windows desktops used for tasks such as running simulations and data science software. 

The first processor lineup is known as the Xeon W-2400 series. It includes eight chips designed to power standard workstations. Intel is also rolling out the more advanced Xeon W-3400 series, which comprises seven chips geared toward premium workstations.

“Our new Intel Xeon desktop workstation platform is uniquely designed to unleash the innovation and creativity of professional creators, artists, engineers, designers, data scientists and power users – built to tackle both today’s most demanding workloads as well as the professional workloads of the future,” said Roger Chandler, vice president and general manager of creator and workstation solutions at Intel.

The fastest processor Intel debuted today is the W9-3495X. It features up to 56 cores that can reach a maximum frequency of 4.8 gigahertz. Compared with Intel’s previous-generation workstation silicon, the W9-3495X runs single-thread applications up to 28% faster and delivers a 120% increase in multithread performance.

The W9-3495X comprises four chiplets, or compute modules, that each contain multiple cores. Intel manufactures the chiplets separately and then links them together into a single processor using a technology called EMIB. EMIB enables Intel to make processors in a more cost-effective manner than similar technologies on the market.

To carry out calculations, the chiplets in a central processing unit require the ability to exchange data with one another. Manufacturers usually address that requirement by placing chiplets on a rectangular piece of silicon known as an interposer. It includes wires through which data can be shared between a processor’s different modules. 

Interposers provide fast data transfer speeds, which helps improve processor performance. But they can be difficult and expensive to manufacture. The EMIB technology in Intel’s new W9-3495X flagship workstation chip is designed to address that limitation. 

EMIB uses a miniature interposer that is simpler to manufacture than standard implementations of the technology. The result, according to Intel, is a reduction in production complexity and cost. 

The four chiplets that comprise Intel’s W9-3495X chip are placed on a base layer powered by EMIB technology. The base layer contains an embedded, miniature interposer that allows data to move between the chiplets. It also includes wires that transport power to the processor from the motherboard on which it’s installed.

Intel’s EMIB technology, which reportedly wasn’t available with earlier workstation processors from the company, is only one of the contributors to the W9-3495X’s speed. Intel has also implemented a number of other improvements in the chip. 

The W9-3495X features a 105-megabyte L3 cache that stores the data the chip is processing close to its logic circuits, which reduces data travel times. The result is an increase in performance. The chip also features Intel Turbo Boost Max Technology 3.0, a mechanism that can relegate demanding applications to the fastest cores in a processor.

Several of the other chips that Intel debuted today also feature its EBIM technology. Others support overclocking, the practice of increasing a processor’s frequency above the default top speed. Furthermore, some of the chips can overclock the memory of the computer in which they’re installed to further improve performance. 

The seven chips in the W-3400 series, the fastest of the two workstation processor lineups Intel debuted today, offer between 12 to 56 cores. The lower-end W-2400 series, in turn, offers eight chips with six to 24 cores.

Many workstations feature not only a CPU but also a graphics processing unit to speed up rendering and machine learning applications. Intel chips are most commonly used alongside Nvidia Corp. GPUs. This morning, Nvidia announced that multiple computer makers plan to launch workstations that will combine Intel’s newly announced processors with its RTX 6000 graphics card.

The RTX 6000 includes 18,176 CUDA cores, which have a similar role as the processing cores in a CPU. The chip also features circuits optimized to run ray tracing rendering algorithms. A third set of specialized circuits, dubbed Tensor Cores, can speed up artificial intelligence applications.

Nvidia says some workstations will offer the option to combine the RTX 6000 with its ConnectX-6 Dx SmartNIC chip. The chip is used for networking tasks such as moving data between computers. According to the company, it will enable workstation users to download and share complex datasets more quickly.

Image: Intel