A train of new processors with features that are each more surprising than the next. In Dallas, on the occasion of its Intel Vision 2022 show in person, Intel returns to the good times of the IDF, this annual event hosted by Pat Gelsinger in the early 2000s and getting attention at every opportunity. computing towards new technological horizons. Ten years later, fifteen years later, Pat Gelsinger is back, he is now the CEO and he has the intention of bringing Intel back to the role of its machine.
“Intel’s challenge now is to launch the great stuff we have in our drawers as soon as possible. Previously, we were limited in our production capacity, in our ability to provide the expectations of our customers. We have had management problems, we do not supply the best products in certain markets. To solve all this, we reorganized Intel into six divisions. Each in an IT sector, each has its own accounting and each aims to gain as much market share as possible against specific competitors, ”the CEO launched during the event.
“Honestly, how can we explain that we have nothing to offer in 10 years in the data center accelerator card market? This era is over, we are now coming up with a very aggressive offer!”, He lost at a press briefing.
The announcements are a processor for portable stations, the Core HX, which has 16 cores and runs at up to 5 GHz. A 52-core Xeon Sapphire Rapids, approximately 10% faster in tests than the 64-core AMD Epyc Milan. An Artic Sound M GPU designed to render server-side graphics and should enable all the immersive environments that the cloud will soon offer, including Facebook metaverse. A Gaudi 2 accelerator that speeds up artificial intelligence algorithms, with a score of 5425 on the ResNet-50 benchmark, while Nvidia’s A100, the ultimate in the field, only reaches 2930.
And so are IPUs – the equivalent of DPUs at Intel – that should make it possible to generalize smart networks to 200 Gbit/s per Ethernet connector.
A new industrial strike force
As for the sixth activity, it’s pretty simple the factories, the ones that Intel has modernized in Ireland and Arizona, such as the ones that should go off the ground in 2027 in Germany and Ohio.
“We have raised the technical level of our factories, by changing suppliers for equipment or by investing in these suppliers so that they can improve their products. But also by changing the way of directing this activity. We learned more on production lines in a year than in ten ”, swears Pat Gelsinger.
“There, we need more automated equipment, we’re releasing it on our sites. Between the lines, he remembers that he was released at EMC during this famous ten years and Intel asked him to back when he became VMware’s most powerful boss.
“What else is left for us?” Change our industrial approach to our customers. Our new approach is to bring the ecosystem where our customers are, in France, Italy, Poland, Spain. We will build the chips they need on site, “the CEO added.
In this case, some European countries should see the emergence of an industrial structure on their territory. In France, this will be the R&D center where the next data center chips will be developed. Officially, this center will focus on the design of Xeon and GPUs that Intel will sell around the world and, at the same time, will welcome car manufacturers or any other industrialists to codesign the special parts in mind. they put in the heart of their equipment in the future.began. Unofficially, hyperscalers – AWS, Azure, Google GCP – will go to this center to order custom Xeon builds.
“Hyperscalers want to develop the technologies they use on their own. But their infrastructure is still processor-based. So they expect us to give them the eco-responsible and elastic processors that they have. will be deployed in their datacenter and which will make a difference to serial servers that companies can install in their own computer rooms ”, argues Sandra Rivera, the boss of Intel’s Datacenter division.
“By starting from a base of standard Intel processors, they can benefit from lower prices thanks to an economy of scale. We will make the same circuits regardless of the product or the customer, but we will do differently. theirs. “
“We will work together to innovate on a circuit and this circuit will end up being integrated or not in some of our processor models that are now built by chiplets. In fact, we have more options to innovate than before, ”he explained.
16 cores for laptops
The first day of the event gave pride a place on new processors for portable workstations, a true technological showcase in the race between Intel and its rival AMD. These new Core HX processors will be available in seven models, including a high-end, the Core i9, which has 16 cores, i.e., eight efficient and eight economical. Budget cores can only run one process each, compared to two of the high-performance cores, providing a total of 24 processes simultaneously. The performance cores run at 2.3 GHz with peaks at 5 GHz. Economical running at 1.7 GHz with peaks at 3.6 GHz. Power consumption is from 55 to 157 watts depending on the main usage. The cache memory is 30 MB and the integrated GPU has 32 graphics units at 1.55 GHz.
The i7 and i5 models have more or less reduced number of cores, frequencies and cache memory. At the entry level, the Core i5 has 8 cores (4 + 4), 12 MB of cache, 16 graphics units, slightly better base frequencies (2.4 and 1.8 GHz) and fewer turbos (4. 4 and 3.1 GHz) . Fuel consumption is the same on all models. Three models are available in a “vPro” version, i.e., which primarily incorporates the additional security functions of professional PCs.
The use of this or that type of heart is not very clear. Intel seems to be saying that high performance cores are used by default for foreground applications and economic for those running in the background. An approach different from Apple – which invented the principle of efficient and economical cores in its M1 processors – where economical cores are used by default and only passed to efficient cores when they are full of calculations, typically for graphics applications.
It remains to be seen what the point is of having too many cores in a laptop PC. “More and more applications are being built on multithread. But despite parallel execution threads that, admittedly, are less used in implementing a basic algorithm, the huge advantage of a large number of cores in a portable station is the multiplication of input-outputs. It becomes realistic to use simulation applications that read or write in parallel to several NVMe SSDs, ”explains Daniel Rogers, product manager of the processors division for client devices.
Taking the information, a machine with a Core HX can have four NVMe SSDs identical to its PCIe buses.
These Core HX processors are the laptop version of the Core Alder-Lake for desktops unveiled last fall. Like them, they support PCIe 5.0 buses twice as fast (especially with NVMe SSD access) and 128 GB of RAM per socket. Their threads also execute 17% more instructions at a particular time than previous processor generations.