The future of Intel: “The PC is not dead”

Vision 2022 is Intel’s first full-fledged post-pandemic event, held in Dallas, Texas. It aims to bring together partners, customers and the press to present the company’s vision for the products and technologies of the future. While the restrictions are gradually being relaxed in this country, almost nobody in the USA wears a mask anymore and one might think that the pandemic would be completely overcome. According to the plans of Intel CEO Pat Gelsinger, this event is to take place every six months and thus follow the legacy of the IDF, the Intel Developer Forum, of previous years.

Intel’s “Vision”: A software ecosystem

Intel is a rapidly changing company for a number of reasons. The market is demanding an unprecedented speed of innovation and the ability to renew from companies, and the same is being demanded of Intel. For the American company, this also goes hand in hand with the need to overcome a phase of apparent difficulties with the so-called “execution”, ie the ability to respect internal roadmaps leading to the release of specific products in well-defined periods.

Intel’s all-round blow: Numerous chips for supercomputers at the in-house exhibition (4)

Source: Intel

The best example of this is the Alder Lake generation, both mobile and desktop, which would certainly not have made this leap in innovation and performance if Intel had remained the quasi-monopolist in consumer CPUs without serious competition (from AMD). would. When it comes to desktop CPUs, Intel has been facing significant headwinds for a long time with the AMD Ryzens, which has now also blown away significant market shares in mobile CPUs. There is also the Apple competition. Not only that Apple “broke up” with Intel and with the Apple M1 not only appears as a hardware competitor, but has also set a milestone in efficiency. Pat Gelsinger said last year that Intel will return to Apple notebooks, you just have to “just develop better processors” than Apple. This self-confidence was again underlined: “It will take some time, but we will return to the Macbooks”, says the new boss of the Client Computing Group Michelle J. Holthaus self-confidently.

Much work has been done since Pat Gelsinger returned as the new CEO and the American company has once again proven its ability to maintain “execution” as the fundamental objective of all projects. Intel is now a company that must work effectively on all fronts to meet the technological commitments of its public roadmaps while also doing a gargantuan job of expanding production capabilities, both for its own products and those of third-party customers.

If you look at Intel as a classic “chip company”, i.e. as one whose core business is the production of processors, you lose sight of the essentials of the company. In fact, the semiconductor market is increasingly connected to the software ecosystem, the speakers point out, without interaction and dialogue between the different hardware platforms being impossible. According to Raja Koduri, companies like Intel find it much more difficult to develop software solutions than hardware solutions, even though Intel employs around 17,000 software developers.

“The PC is not dead”

“No one works eight hours with their smartphone, right? But we still sit in front of a PC,” emphasizes Michelle J. Holthaus, who has been in charge of the Client Computing Group at Intel for almost four weeks and succeeds Gregory Bryant (GB), who long ago Keynotes for Intel as a face. The momentum created around the world by the pandemic has brought even more focus to technological innovation and in particular the PC platform. It is enough to visualize how changes that would have taken a decade on paper, such as the indispensable shift to smart and flexible work, have taken place in literally a week.

According to Holthaus, the pandemic confirms that the PC is the reference platform for being continuously operational, especially compared to smartphones: Without a PC, it would not have been possible to carry out work productively in many areas. Holthaus does not only mean the home office including the booming video conferences, she also mentions video consultation hours with the doctor, which are more widespread in the USA than in Germany. According to Holthaus, this requires more than just smartphones.

Alder Lake-HX: 16 cores in notebooks and a view of Meteor Lake

PCGH has already reported on new high-end CPUs for notebooks. The 16 cores of the new HX models are again performance and efficiency cores – in this case eight each for the i9-12950HX and i9-12900HX. The previous Alder Lake top model i9-12900HK had six performance and eight efficiency cores. Intel advertises the use of the HX models primarily in workstations, i.e. professional applications such as Autodesk or Blender. It is a CPU “for engineers, scientists and enthusiasts”. The target group for high-end gaming notebooks should probably also be addressed with the latter. OEMs have already announced some gaming notebooks with HX CPUs. Because of the low gaming added value, cheaper Alder Lake CPUs will certainly continue to be used more often here.

Meteor Lake is Intel’s next but one generation of notebook CPUs. Curiously, only the Intel Vision is on display and not the 13th generation, codenamed Raptor Lake, which is due to be announced later this year. During the keynote, Michelle J. Holthaus held a Meteor Lake die for the camera. Shortly before, it was already reported that Meteor Lake was already successfully booting in the laboratory and an original image of the compute tile was also doing the rounds on the net at the trade fair. Meteor Lake is technically interesting because it uses Intel Tiles to piggyback the GPU onto the CPU, similar to what AMD does with the level 3 cache on the Ryzen 7 5800X3D. The tiled architecture allows Intel more freedom in assembling the die than with the current Alder Lake or the next Raptor Lake generation. The P cores (Redwood Cove) and the efficiency cores (Crestmont) can be seen on the compute tile, but not the graphics unit or interfaces. The respective vertically stacked tiles or chiplets, as AMD calls them, are to be manufactured by TSMC in their node with three nanometers.

New Xeons: Sapphire Rapids introduced

Intel’s new HEDT CPU Sapphire Rapids

Source: PC Games Hardware

The first versions of the fourth-generation Intel Xeon Scalable processor, codenamed Sapphire Rapids, are being revealed and are expected to be available later this year. So Intel hasn’t given up on the HEDT segment, although it has lost a lot of its relevance on mainstream platforms, not least because of Ryzens with 16 cores. Nothing else has been released since Cascade Lake-X from 2019 – after all, the many cores of the Ryzen and Threadripper CPUs were a formidable competitor.

The Sapphire Rapids now offer current standards, such as DDR5, PCI Express 5.0, CXL 1.1 and feature new integrated Xe graphics accelerators that offer up to 30 times the performance of the previous generation through software and hardware optimizations for AI workloads should offer. The Intel Xeon processors with the code name Sapphire Rapids and the High Bandwidth Memory (HBM) will significantly increase the memory bandwidth available to the processor and thus once again become an alternative for supercomputing.

HPC monster “Ponte Veccio” and other server GPUs

The monster “Ponte Vecchio” GPU contains over 100 billion transistors and uses EMIB and Foveros to connect a whopping 47 tiles in a single package. As with Meteor Lake, the tiles are connected both vertically and laterally. The chip was designed for the Department of Energy’s Aurora Exascale supercomputer, but will also be sold to customers working in HPC and AI. This complex chip will feature a memory structure with a bandwidth of over 5 TB/s. With the presentation of its chips, the company wants to reassure the public and investors that it continues to make progress. Intel’s CEO recently showed off an 18A SRAM wafer at an investor meeting in February and said it is currently six months ahead of schedule.

Raja Koduri with Intel’s Datacenter GPU (Arctic Sound-M)

Source: PC Games Hardware

Intel’s data center GPU, codenamed Arctic Sound-M (ATS-M), is the industry’s first discrete GPU with an AV1 hardware encoder. ATS-M is a versatile GPU with high transcoding performance that is expected to reach 150 trillion operations per second (TOPS). Developers can develop for ATS-M via oneAPI thanks to an open software stack. ATS-M will be available in two formats across more than 15 partner projects including Dell, Supermicro, Inspur and H3C. The GPU for data centers will be launched in Q3 2022.

The Habana Gaudi2 is announced as the next-generation AI processor for training and inference. Intel acquired Habana Labs in 2019 and the Gaudi2 is now ready as the next-gen AI accelerator. Intel touts that the Gaudi2 offers twice the AI ​​training performance of current NVIDIA A100-based offerings. Gaudi2 is manufactured in 7 nm, goes from 8 to 24 TPCs for media decoding and processing, has 96 GB HBM2e memory and 48 MB SRAM cache, offers 24 x 100 GbE networks and has a TDP of 600 watts.

Habana Greco was also announced as a new offering optimized for deep learning inference efficiency. The Greco will be delivered to customers in the second half of the year. The move from the existing Habana Goya to the Greco means 16nm to 7nm manufacturing, 40GB/s DDR4 memory up to 204GB/s LPDDR5, various new hardware features and a single-slot HHHL card with a TDP of 75 watts compared to 200 watts for the Goya.