AMD dreams of being the king of server hybrid components

6

A few days ago, AMD hosted a financial analyst conference and made some announcements regarding its consumer offerings. We were also able to learn more about the server roadmap and brand ambitions for 2024.

It’s been 5 years since AMD made a big comeback in the consumer processor space with its Ryzen that turned the market on its head and put Intel in difficulty. The goals were then achieved and the company must now drive the point home.

AMD and Intel are no longer alone in servers

And it intends to do so, doubling down on ads in recent months to impress its partners and investors: Zen 4 is coming, and new Radeons will follow. The growth in the next few years will come in part from its efforts in the server space, which is a growing market, with an estimated potential of at least $42 billion over the long term.

Server Market, an Opportunity That Doesn’t End: $26 Billion in Zipper

AMD targets a market of over $42 billion. © AMD

And if the share won by EPYC processors isn’t as strong as it is now (about 12%), they’re still in the minority despite all the work the AMD teams do. The inertia of this sector is even stronger, and the entire super-ecosystem of Intel has played a huge role in its favor, both at the hardware and software level. As Pat Gelsinger, the new head of the Santa Clara giant, declared it loud and clear: he intends not to let go and will return to the result.

But for the old competitors, the threat is now elsewhere: ARM solutions are gaining momentum, with high energy efficiency, very good density, and certain competitive pricing; The strengths that have so far been those of AMD against Intel.

And so we see companies like Ampere Computing gaining ground, with NVIDIA ambushed on their CPUs. At Computex, the latter was clear: by 2024, he intends to position Grace (Hopper) as surrogates that will allow him to accomplish the mission set by Jen Hsun Huang: to reinvent the data center.

Advertising, your content continues below

Just like Intel, the father of GeForce realized that with a full view, one convinces the pros. In addition to its own CPUs and GPUs, it has campaigned for years to adopt next-generation network (DPU) solutions to migrate some accounts and lighten the central processor. It now offers its own switches, servers, and even data centers. All with a good dose of software that’s (almost) ready for any situation.

A new era for EPYC, the age of diversity

To all of these threats AMD must respond now, and not move in a few years from the ambition of getting close to number one to not finishing on the platform.

So a battle plan was drawn up, with a strategy on several levels. The first is to review the EPYC processor offering to adapt it to a wider range of needs. Strategy completed One die fits all From the beginning, we adhere to customer needs and competition.

Advertising, your content continues below

Thus Zen 4 will be available in four divisions. First of all, EPYC Classic (Genoa) processors expected for the fourth quarter, with both classic and full cores. As a reminder, it’ll reach 96 (compared to 64 in the EPYC 7003), engraved at 5nm with support for DDR5, PCIe 5.0, Compute eXpress Link (CXL) and instructions like AVX-512 or AI acceleration.

AMD reports an 8-10% improvement in IPC over this generation, but performance is up to 35%, and 25% on the power efficiency side (for Ryzen 16C/32T). Developments that allowed to improve engraving accuracy and frequencies.

Zen 4 will have twice as many skins… and Zen 5 (Taurine)?

AMD architectures roadmap for server scale CPU.  © AMD

AMD architectures roadmap for server scale CPU. © AMD

Advertising, your content continues below

In the second step, version “X” will come with 3D V-Cache to improve performance in very specific cases, and it is already in the current offering. In 2023 we will discover Bergamo (Zen 4c), AMD’s answer to ARM chips: up to 128 cores can be used in the same socket. Thus it will be reduced, even if the manufacturer promises that it will run the same instruction sets, avoiding different levels of optimization on the same generation.

Finally, this conference was an opportunity to discover Siena, the codename for processors that should target the telecom and embedded market. You will therefore be limited to 64 cores, and more easily accessible, with an emphasis on the performance / watt ratio. An interesting strategy with the emergence of “edge” uses but also of “software-defined” network solutions, running on traditional processors, especially around 5G.

Consolidation of supply after the external growth stage

Those choices take on more meaning when you look at them from the two recent acquisitions of AMD: Xilinx, the giant of FPGAs and AI-ready solutions, and then Pensando, which offers smart NICs and other DPUs. In order to speed up more and more things directly through the network.

From now on, AMD no longer offers only CPUs and GPUs, but an ecosystem

AMD intends to offer a complete ecosystem of its partners.  © AMD

AMD intends to offer a complete ecosystem of its partners. © AMD

That’s enough to open up markets for the manufacturer that were previously closed to it, by offering a highly integrated offering where each brick is designed to work with the others, and rivals Intel and NVIDIA are gearing up too.

Advertising, your content continues below

It now remains to be seen who will get the most convincing offer, but above all the best execution. Since having an interesting proposition on paper is one thing, you still have to unite customers and developers, pick and win big contracts. Everything still has to be done, especially since the local sales team isn’t really AMD’s strong point.

The future will be a hybrid, and the MI300 will be an APU

The last part of the strategy, but not the least: cross-breeding architectures. This is also a movement we see elsewhere, but it seems to be a clear trend for the market for the coming years: not only will different CPUs, GPUs, and accelerators be mixed within the motherboard, but also by different chipsets interconnected in the same SoC.

AMD says it’s already ready for this little revolution, because it already offers processors made up of various Chiplets And it reaches the fourth version of the internal threading (Infinity architecture), with a cover that allows it to expand to CXL 3.0 and UCIe standards.

MI300: Zen 4, CDNA 3, and HBM in One Package, Is the Competition Collapsing?

With the MI300, AMD prides itself on providing the best comprehensive data center solution.  © AMD

With the MI300, AMD prides itself on providing the best comprehensive data center solution. © AMD

We can thus imagine seeing tomorrow’s components with the CPU, GPU, AI accelerator and why chips are not made by third parties, side by side or stacked (3D stacking).

Until then, the first testing will be done with the MI300. This computing solution for servers will be an APU cored at 5nm, consisting of a CPU (Zen 4) and a GPU (CDNA3) to access the same HBM memory, all in the same package.

AMD promises eight times performance over the MI250X in AI-like workloads, with a performance-to-watt ratio five times between CDNA 2 and CDNA 3. See you next year to see what it’s all about. will actually return.

Advertising, your content continues below

Advertising, your content continues below

#AMD #dreams #king #server #hybrid #components

Leave a Comment

Your email address will not be published.