May 28, 2021 CXL! PCIe 5.0! Xilinx+AMD=Limes and Tacos!

Today I found this comment made by Rand Walker on Seeking Alpha, which is most thought provoking. See link.

{{ AMD stock has suffered since the announcement of the XLNX deal. In part this because of the technicalities of absorbing potentially tens of billions of dollars of shorts from merger arbs hedging XLNX positions, but also the strategic rationale for the deal hasn’t been articulated well. Acquirer companies are often circumspect in describing deal benefits for fear of inciting target shareholders, and anti-trust regulators, but even allowing for this, AMD’s presentation of the deal has seemed muddled. The NVDA/ARM deal seems even more incoherent, so what’s going on?

I think that both deals are to prepare for the coming era of PCIe5.0+CXL in the data center. This technology will allow very close coupling of server CPUs, GPUs, DPUs/smart-NICs and storage devices. Intel’s vision is no doubt to link Xeon CPUs with Xe GPUs and Altera-based smart-NICs; with the XLNX acquisition AMD will be able to do the same with EPYC+CDNA+XLNX-based communications devices. NVDA risks being left out in the cold with no x86 CPU manufacturer incentivized to put in the work support their GPUs and Mellanox products. Hence the ARM deal: it would give them a little leverage to encourage CXL collaboration from AMD and Intel, but also gives them a refuge if they’re frozen out of the x86+DXL Data Center ecosystem.

This video on CXL from Patrick at ServeTheHome finally helped me to see the motivation for both deals. www.youtube.com/…

May 27, 2021. 10:01 AMLink
AMD files Xilinx planned acquisition for EU approval – SA Editor Josh Fineman }}

Here are excerpt from Lisa Su’s presentation at the JP Morgan 49th Annual Global Technology, Media and Communications Conference on May 24, 2021.

Harlan Sur

What we’re hearing about now over the next, let’s say, call it, two to three years is on a unified CPU, GPU architecture, unified memory architecture, high-speed fabric, and actually, the AMD team, your team outlined this last year. And your competition now has announced a similar offering as well. Help us understand the team strategy, AMD strategy for closer integration between the CPU and GPU. I know that the team outlined the roadmap with EPYC combined with your second generation CDNA-based GPU architecture and your third generation Infinity fabric. Is this a 2022 or 2023 launch?

Lisa Su

Yeah. Very good, Harlan. You know our roadmap well. I would say we’ve been thinking about this problem really for the last five years. We look at every problem as an opportunity. The opportunity is, when you have these components sitting side by side in a system, you want to be able to share resources between the components. If you think about a high performance CPU sitting next to a bunch of GPUs that are better for parallel processing, you want to be able to have a very smart interconnect, and we call that our Infinity fabric, to really connect the two in a very efficient way and also share resources. Yes, this has been very much on our roadmap.

Last year, we talked about our first generation CDNA architecture. This year, as I said, we’re putting together our next generation CDNA architecture. This is actually a key component that enabled us to win the largest supercomputer bids in the US around the Frontier Oak Ridge National Labs installment as well as the Lawrence Livermore National Labs installment with El Capitan and many others. But it’s a coherent interconnect between CPUs and GPUs that allow us to fully optimize for HPC and for AI and ML applications. And we will be launching the next generation of that architecture, actually, later this year. We’re very excited about it. I think it’s progressed extremely well. It’s the next big step in sort of innovation around the data center architectures.

Harlan Sur

As we think about the complexity of the compute workloads and requirements for different types of solutions to tackle some of the increasing complexity of computes, you’ve got a great portfolio, but you’re going to be adding to that, right? Adding to that heterogeneous compute portfolio with the Xilinx team. Can you just give us an update on the pending acquisition of Xilinx, their leadership in FPGA and adaptive system-on-a-chip capabilities? And given the requirements for more compute acceleration, how does their technology and portfolio fit with your heterogeneous compute strategy for data center? And what does the Xilinx team bring to AMD from a market expansion opportunity?

Lisa Su

Yeah. Absolutely, Harlan. We’re very excited about the Xilinx acquisition. I think if you think about sort of two companies that are extremely complimentary, that’s really AMD and Xilinx. What Xilinx brings is – frankly, they’re number one in the FPGA market and the accelerated adaptive computing market. When you look at strategically in the data center, our vision of a very heterogeneous architecture includes a world where you have CPUs, GPUs, FPGAs, ASICs, sort of all connected in the same ecosystem. So, bringing that FPGA leadership from Xilinx to the AMD portfolio really gives us a tremendous overall end-to-end portfolio in terms of capability.

So what does these all mean?

Now let’s look at this excellent article written by Patrick Kennedy on May 21, 2021, here.

He also has a great YouTube video explaining everything using the limes, sodas and tacos analogy.

Patrick Kennedy of ServetheHome does a great job here.

Patrick did a superb job explaining everything about the upcoming CXL and PCIe 5.0 technology and how they will vastly improve all the ways CPU, GPU, DPU, FPGA, memory and storage will be connected together.

AMD has been talking about heterogeneous computing, Infinity Fabric 2.0, even 3.0 for the past several years. However, the market barely paid any attention.

Lisa Su also said:

AMD will push the envelope on 2.5D and 3D packaging as we go forward because it’s a key element to unlock that next level of performance. Again, we’ll talk a little bit more about that as we go through the next number of months as we roll out the next phase of our of our roadmaps.

In short the timing is perfect for AMD+Xilinx to combine all their know how and fully utilize what CXL and PCIe 5.0 will bring. It’s no longer just about selling Epyc CPUs or Instinct GPUs. AMD wants to provide the total solution of processors, accelerators, interconnects and likely software solutions as well. Only when AMD reached a market cap of around $100B was Lisa Su able to pull this Xilinx acquisition off without using cash. So AMD bought Xilinx just in time, to be able to fold Xilinx in throughout 2022 and benefit from the capabilities CXL and PCIe 5.0 bring.

Intel will definitely do the same with their Xeon, Xe GPU, Altera DPU and Optane memory.

Nvidia is moving beyond just gaming GPU, and wishes to provide total solutions for the cloud, AI etc. Nvidia already has Mellanox and software expertise. Acquiring ARM will provide the CPU piece of the whole puzzle.

In summary, AMD, Intel and Nvidia will all get into the business of providing end to end solutions for the cloud.

AMD is not about game consoles, PC CPUs and PC GPUs any more. Xilinx is key to the future!

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s