Discussion AMD Acquires Xilinx

[DisEnchantment]

AMD to Acquire Xilinx, Creating the Industry’s High Performance Computing Leader

Browse AMD’s company-wide and financial press releases.

ir.amd.com

Straight from the Horse's mouth

Strategic transaction strengthens AMD’s industry-leading technology portfolio —

• Expands AMD’s rapidly growing data center business
• Xilinx, the No. 1 provider of adaptive computing solutions, increases AMD TAM to $110 billion
• Immediately accretive to AMD margins, cash flow and EPS
• All stock transaction with combined enterprise value of approximately $135 billion

All stock. No Cash. Could not have been a better way to make use of that inflated market valuation.
Lots of people were questioning AMD why they are not capitalizing on their Market valuation. But this is it.

Management and Board of Directors
Dr. Lisa Su will lead the combined company as CEO. Xilinx President and CEO, Victor Peng, will join AMD as president responsible for the Xilinx business and strategic growth initiatives, effective upon closing of the transaction. In addition, at least two Xilinx directors will join the AMD Board of Directors upon closing.

They got decent advisors too.

Advisors
Credit Suisse and DBO Partners are acting as financial advisors to AMD and Latham & Watkins LLP is serving as its legal advisor. Morgan Stanley is acting as lead financial advisor to Xilinx. BofA Securities is also acting as a financial advisor and Skadden, Arps, Slate, Meagher & Flom LLP is serving as legal counsel.

Ex AMD rejoins the family bringing Xilinx with him

“We are excited to join the AMD family. Our shared cultures of innovation, excellence and collaboration make this an ideal combination. Together, we will lead the new era of high performance and adaptive computing,” said Victor Peng, Xilinx president and CEO. “Our leading FPGAs, Adaptive SoCs, accelerator and SmartNIC solutions enable innovation from the cloud, to the edge and end devices. We empower our customers to deploy differentiated platforms to market faster, and with optimal efficiency and performance. Joining together with AMD will help accelerate growth in our data center business and enable us to pursue a broader customer base across more markets.”

 

[Hitman928]

I just hope that it doesn't end up causing them years of financial ache like the ATi acquisition did 😢

It's an all stock deal, so no need to worry.

 

[Saylick]

Really? I always had the impression the good working relationship was there first and that's why they considered the merger to begin with. So I'm surprised anybody would consider that a "big surprise". But I agree it's certainly a good sign for the future working of the combined company.

I'm not surprised at how they have a good working relationship. Tons of companies have good working relationships. I'm surprised that they've been planning something jointly the whole time the acquisition was under review. This type of collaboration normally doesn't happen until after the acquisition is confirmed, so they either had the utmost confidence that the acquisition would go through OR they knew that they could mutually benefit from a collaboration even if it didn't. The slide deck showing Xilinx integrating their software suite under the same platform as AMD alluded to this but I didn't get my hopes up.

 

[moinmoin]

This type of collaboration normally doesn't happen until after the acquisition is confirmed, so they either had the utmost confidence that the acquisition would go through OR they knew that they could mutually benefit from a collaboration even if it didn't.

The latter was apparently the case. As your article stated (and you even quoted):

"Apparently, AMD has had a long-term development agreement with Xilinx and was able to work together even as the deal was getting approved."

So first this long-term development agreement was there, and while that was underway a merger seemed like a feasible goal for both sides. Makes sense to me.

When searching for Su and Peng in the years before there are plenty events where both had talks like SEMICON West, #AIDesignForum etc. so there was a mutually shared vision at least between the two even back then. Then there was also this in early 2018:

Adaptive Computing

www.xilinx.com

 

[Vattila]

When searching for Su and Peng in the years before there are plenty events where both had talks

After the announcement of the deal, they also disclosed that merger talks had been going on since 2018 2019, I think. Check out the SEC filings. Those include records and summaries of meetings, rival offers, terms negotiated, etc. All good material for the hypothetical book.

Edit: I dug up the SEC filing for those who are interested in the background story:

"From time to time over the past few years, AMD and Xilinx have had informal discussions regarding potential strategic partnerships, collaborations and a potential combination between the two companies. In connection with Xilinx’s approach to AMD about a potential business combination of the two companies, on August 13, 2018, AMD and Xilinx signed a mutual non-disclosure agreement."

SEC Form S-4 filing

 

[Asterox]

I just hope that it doesn't end up causing them years of financial ache like the ATi acquisition did 😢

Xilinx is FPGA market leader, and this video is from year 2014.

For global industry market, Xilinix tech or patents no doubt are significantly more important than ATI or AMD "mainstream GPU technology".

Introducing the Industry’s First 20nm Space-Grade FPGA

The new 20nm Radiation Tolerant (RT) Kintex® UltraScale™ XQRKU060 FPGA, enables unlimited on-orbit reconfiguration and machine learning in space for real tim...

youtu.be

 

[13Gigatons]

AMD CEO Dr. Lisa Su joins Yahoo Finance Live to discuss AMD's acquisition of Xilinx and how it will benefit their customers in the strong demand for PCs, as well as the supply challenges in the chip market.

 

[moinmoin]

I dug up the SEC filing for those who are interested in the background story:

SEC Form S-4 filing

Nice read, I recall reading it back when that surfaced. I wonder which company "company A" is? Can't be Nvidia or somebody of that caliber as later said company proposes a merger of equals. Somebody suggested Marvell which looks like a perfect fit indeed. Going by comparable revenue Novatek and Realtek could also be possibilities.

 

[Saylick]

Nice read, I recall reading it back when that surfaced. I wonder which company "company A" is? Can't be Nvidia or somebody of that caliber as later said company proposes a merger of equals. Somebody suggested Marvell which looks like a perfect fit indeed. Going by comparable revenue Novatek and Realtek could also be possibilities.

Yeah, very interesting indeed. This statement from the filing stood out to me:

In addition, Mr. Peng provided an update to the Xilinx Board on Mr. Peng’s recent meetings with the Chief Executive Officer of Company A. After discussions, the Xilinx Board determined that certain recent activity at Company A would render a transaction of the size necessary for a combination with Xilinx challenging from a business execution perspective and directed that Mr. Peng deliver this message to Company A.

I wonder what was happening at Company A at the time that would warrant such a statement. Was anything fuzzy going on at Marvell during the beginning of 2020? I just don't follow Marvell at all to know, unfortunately.

 

[nicalandia]

How is this any difference from Nvidia + ARM? Why one was so accepted and the other so repudiated?

 

[itsmydamnation]

How is this any difference from Nvidia + ARM? Why one was so accepted and the other so repudiated?

Does AMD have the ability to be predatory to existing Xilinx customers AMD directly compete with once the transition is complete. Because that was the case for NV + ARM

 

[HurleyBird]

How is this any difference from Nvidia + ARM? Why one was so accepted and the other so repudiated?

Similarities:

They're both tech companies which costly to acquire.

Differences:

Literally everything else.

 

[nicalandia]

Similarities:

They're both tech companies which costly to acquire.

Differences:

Literally everything else.

Nvidia was trying to acquire a solid CPU business to compliment their GPU/AI Accelerators and build an HPC ecosystem.

AMD acquiring Xilinx will make their HPC ecosystem even stronger, they already have a strong CPU and GPU/Accelerator portfolio

Intel + Altera is also a good move.

I think Nvidia is at a disadvantage against AMD and Intel

 

[waffleironhead]

Nvidia was trying to acquire a solid CPU business to compliment their GPU/AI Accelerators and build an HPC ecosystem.

AMD acquiring Xilinx will make their HPC ecosystem even stronger, they already have a strong CPU and GPU/Accelerator portfolio

Intel + Altera is also a good move.

I think Nvidia is at a disadvantage against AMD and Intel

Nvidia can and already does make ARM cpus.

Them buying arm wasnt about them making arm cpus, it was about them controlling how everyone else uses ARM in the future.

 

[Saylick]

Nvidia was trying to acquire a solid CPU business to compliment their GPU/AI Accelerators and build an HPC ecosystem.

AMD acquiring Xilinx will make their HPC ecosystem even stronger, they already have a strong CPU and GPU/Accelerator portfolio

Intel + Altera is also a good move.

I think Nvidia is at a disadvantage against AMD and Intel

Nvidia can make their own ARM CPUs without owning ARM.

AMDs CPUs are good, but I don't think you can say their GPU/accelerator portfolio is strong in comparison to Nvidia's.

What Intel has done with Altera after they were acquired has been pitiful. Absolutely mismanaged. I wouldn't say that Intel owning Altera jeopardizes Nvidia's success in any way. If it did, it would've had an impact by now.

Hard to say that Nvidia is at a disadvantage without ARM. At least the market doesn't think so given Nvidia's market cap exceeds even that of Intel, AMD, and Xilinx's combined. Their stock price didn't even flinch when they officially gave up on acquiring ARM. Everyone thought it was a long shot, but no one thought that if it didn't happen they would be in trouble. If it did, their stock price would've sank upon it's failure.

 

[Roland00Address]

How is this any difference from Nvidia + ARM? Why one was so accepted and the other so repudiated?

Too many jurisdictions with ARM who could veto.

 

[moinmoin]

Too many jurisdictions with ARM who could veto.

Well, the amount of jurisdictions that could veto is technically pretty much the same for every semi merger.

 

[Joe NYC]

FPGA explained

 

[ThatBuzzkiller]

I think the portability aspect of OpenCL vs SYCL is moot.

As programming models though, they differ vastly. If you go through any presentation on SYCL that mentions OpenCL you will have the advantages spelled out, so I will not go over them here. Go to the SYCL discussion thread. Let me just say, the keywords are single-source and productivity. As a C++ programmer myself, I have never had the slightest temptation to have even a brief look at OpenCL to learn about heterogenous system programming. All I've heard and sensed is that it is a horrible C-based programming model with clunky implementation and driver issues (something you seem to have been deeply scarred by). On the other hand, Microsoft's (i.e. the excellent Herb Sutter's) C++ AMP dialect had me tempted, and SYCL (which in many ways resembles C++ AMP) even more so, to the point I have started to read up on it (ref. the free SYCL introduction book Data Parallel C++). The important keywords for me are modern C++.

CUDA and HIP are both single source and supports advanced C++ features for higher productivity so if portability isn't a priority for SYCL then why should other vendors with their own compute stack implement SYCL themselves instead ?

Funny note about C++ AMP is that AMD created a spiritual successor/extension with HCC but it didn't see any use so it got deprecated ...

In any case, I reiterate that SYCL has been successfully implemented on top of ROCm and CUDA (hipSYCL) and oneAPI/Level Zero (DPC++). So the backend implementations available are already miles ahead of what OpenCL achieved in terms of robustness and support (on Linux, that is). Since SYCL has implementations in terms of the vendors' favoured frameworks, you are much less likely to see the issues of past stemming from poor vendor support and languishing implementations. Also, studies show that SYCL has good performance portability versus CUDA, OpenCL and Kokkos/Raja, while getting top marks on programmer productivity metrics. See the SYCL thread for references.

If you say that "SYCL has been "successfully implemented" with translation layers that don't currently run any applications then that bar must be set very low. As far as productivity is concerned, that is only true if you're willing to target one compiler. If you're going to target multiple compilers like NVCC, HIP-Clang, or DPC++ compiler then your productivity virtually goes down the drain in favour of "performance portability" because you can't make effective reuse of code. You can't have both high productivity and performance portability because dealing with multiple compiler backends ...

Notably, I don't think OpenCL was ever much used in supercomputing, for instance, while SYCL is being adopted at a rapid pace. In time, SYCL will hopefully trickle down into the client space as well, as the SYCL frontend implementers as well as the vendors of the backends broaden their focus (to the Windows platform, in particular).

I think you need to take a wider perspective here (rather than focusing on details such as SPIR-V, drivers and other horrible nerdy nightmare stuff) so that you don't miss the forest for the trees. For programming heterogenous systems, programmers want an open, simple, powerful and productive programming model with robust and well-supported implementations. SYCL is getting there on Linux, and Windows is hopefully next.

SYCL largely still isn't being used because the only hardware that was worth targeting with it has been delayed which was from Intel. If we look at the real world with frameworks such as Tensorflow or PyTorch and applications like Blender's Cycles X, they only have CUDA and HIP backends ...

What exactly do you mean by 'open' ? The implementation ? Many programmers don't seem to really care how the underlying APIs/drivers work so it may as well be just another black box to them. Different architectures ? That might be to be valuable to some programmers but SYCL doesn't really do this in practice ...

If I had anything to say, SYCL would be the cursed antithesis of Vulkan. Many mature Vulkan implementations are closed source but is open to other hardware in practice while is SYCL is the exact opposite for all the wrong reasons hence why it's 'cursed' ...

 

[A///]

I wonder what was happening at Company A at the time that would warrant such a statement. Was anything fuzzy going on at Marvell during the beginning of 2020? I just don't follow Marvell at all to know, unfortunately.

Marvell? Did they ever have the money on hand in addition to shares for what the offer in the disclosure document states?