AMD to Enter the FPGA Market, in Advanced Talks to Acquire Xilinx

[Valantar]

And most importantly telecom and high-speed networking. I think AMD is after that 5G infrastructure sweet pie, since xilinx already has a decent portfolio of comms&accelerator solutions.

That's probably true. AMD also theoretically has a leg up on Intel in terms of integration given that their entire CPU portfolio is based on chiplets - adding an FPGA or two into the mix for an embedded EPYC or similar product series would probably be relatively easy.

As for why they're looking at this in a broader perspective, I think AMD is seeing significant growth but are also seeing the limits of the X86 CPU + GPU/compute accelerator market, and want to diversify. Makes sense business wise, though there is definitely a need to not over-reach when doing something like this. Buying a company worth between 1/3 and 1/4 of your own? That's defintely a big ask. Though that should also mean buying into a solid source of revenue, of course, and as such it should pay for itself in time even if left to its own devices (as long as the interest on the loans needed to buy isn't egregious).

Given the way Intel has been gobbling up companies left and right for the past decade, I think we need someone else of a similar size to balance things out and avoid Intel entirely cementing themselves as the one-stop-shop for all enterprise compute solutions. (No, I don't think NvidiARM+Mellanox quite does that, as ARM's main business is (still) consumer chips.)

 

[R0H1T]

6 vs 30, piece of cake.

Yeah if only the stock markets didn't outperform virtually every other asset class in that time, perhaps with the exception of real estate, not to mention the number of billionaires has probably quintupled as well.

Which is to say that it's (valuation) no big deal, really just look at ARM+Nvidia

 

[mtcn77]

Now you know who designed those huge HBM connectors. First.

 

[Valantar]

Now you know who designed those huge HBM connectors. First.

What?

 

[mtcn77]

What?

From whence they were collaborating with the industry to build Fury Series, the name of Xilinx was right up there.

They were multiple partners. They went over the reticle limit.

 

[Imouto]

Unlikely. They are probably developed on FPGAs but FPGAs tend to be pricier than ASICs and there is no call for them in an emulation style console where costs are generally cut as much as possible.

MiSTer DIY: Accurately Simulating Retro Game Consoles With FPGAs

Thanks to a tight community of talented developers, the MiSTer project brings cycle-accurate emulation and low latency to Terasic's DE10-nano FPGA

hothardware.com

 

[Valantar]

From whence they were collaborating with the industry to build Fury Series, the name of Xilinx was right up there.

They were multiple partners. They went over the reticle limit.

You mean interposers? "HBM connector" isn't a term I've ever seen used.

 

[mtcn77]

You mean interposers? "HBM connector" isn't a term I've ever seen used.

I didn't want to dig deep. Getting roasted right now, to be frank.
I think they helped to design it and UMC manufactured it.

 

[jeremyshaw]

Digging around in my old box of parts... it would be much more interesting if AMD reclaimed their IP with a Lattice Semi acquisition

Though that would have a 100% chance of being blocked by China in a severe bout of sour grapes.

 

[Dristun]

AMD can easily afford to buy a company with a market cap 1/4th of its own, given that most of the price is likely to be paid in stock, as per WSJ report. Their liabilities have been low for almost a decade, the difference now is that the stock is extremely expensive and their revenue is sharply up, so of course they're looking to capitalize on this.
Don't think of corporate M&A in terms of your wallet or worth, they operate on an entirely different plane.

 

[MAXLD]

Can anyone explain what is AMD looking for with Xilinx acquisition? I am lacking deeper understanding. What benefit would AMD gain?

For PC enthusiasts, apparently not that much directly. But for professional scenarios, there's a gazillion of opportunities.

"Another trend in the use of FPGAs is hardware acceleration, where one can use the FPGA to accelerate certain parts of an algorithm and share part of the computation between the FPGA and a generic processor.[2] The search engine Bing is noted for adopting FPGA acceleration for its search algorithm in 2014.[40] As of 2018, FPGAs are seeing increased use as AI accelerators including Microsoft's so-termed "Project Catapult"[18] and for accelerating artificial neural networks for machine learning applications. " [wikipedia]

Anything related to "Machine Learning" and "AI" is a very attractive business nowadays. And AMD being now in such a growing and attractive proposition in servers, datacenters, and supercomputers, makes this basically another professional field where there's tech benefits and money to be made.

A combination with Xilinx would give AMD Chief Executive Officer Lisa Su more of the pieces needed to break Intel Corp.’s stranglehold on the profitable market for data-center computer components. It would follow moves by rival Nvidia Corp., which bought Mellanox Technologies Ltd. and aims to use its pending acquisition of Arm Ltd. to grab more of that business.

Acquiring Xilinx, which makes programmable chips for wireless networks, would also help AMD expand into a new market just as telecommunications carriers spend billions to build fifth-generation, or 5G, networks.

What Bloomberg Intelligence Says:
"AMD’s interest in Xilinx is both a shield and a sword against Intel’s diverse data-center computing assets, Nvidia’s GPUs, Mellanox’s assets and its potential purchase of U.K.-based Arm. Xilinx’s FPGAs would expand AMD’s computing variety, widening its CPU and GPU capability into AI and networking. The deal may also be financially accretive."
- Anand Srinivasan and Marina Girgis, analysts

Bloomberg

Plus, of the two major companies of this type of product, one is already owned by Intel.
"In 2016, long-time industry rivals Xilinx and Altera (now an Intel subsidiary) were the FPGA market leaders.[52] At that time, they controlled nearly 90 percent of the market. "

As long as it makes AMD a stronger company in the professional application sphere, which is where the big business tech opportunities are, it will benefit us in the end, since it makes the company more robust.

 

[bug]

Here's hoping they don't repeat ATI, bankrupting themselves in the process. They may not survive that again.

 

[InVasMani]

Here's hoping they don't repeat ATI, bankrupting themselves in the process. They may not survive that again.

That depends what they pay for Xilinx. That said Lisa Su is a much better CEO as well so I would defiantly anticipate a more seamless merger than what happened with the ATI merger. I also think FPGA's are more far reaching perhaps in terms of IP than what AMD got out of ATI though GPU's are fairly diverse as well and what they do they do very well at those specialized tasks. I really think Xilinx has a lot of IP that will integrate great alongside AMD's own IP portfolio in tandem. I defiantly feel FPGA's aren't going away in any hurry they are programmable so why would they and if anything where things are headed we'll see a lot more use of them a significant amount perhaps when you take into account AI, deep learning, and data analytics that could have specialized programming.

I think that MAXLD's post was pretty much spot on. Then take into account AMD's chiplets apply that with infinity fabric to FPGA along with a nice helping of CPU/APU/GPU tech and you've got a good set of tools for self learning robotics AMD will have it's terminators before you know it, but seriously there are lots of good applications for that robots that can perform neurosurgery with far less error than any human. There tons of fields where that combined IP portfolio from a merger would benefit a lot. I think there will be a whole trickle down effect if the merger happens on the FPGA front in sweeping amount of area's of technology. Really with the IP they already have it could help usher in a fairly new era of computing in a much more cohesive way cognitive computing. Imagine a singular chip with infinity fabric with FPGA/CPU/GPU chiplets and several cores on each of those and throw in some 3D stacking as well.

The future looks bright and maybe we'll see a bit of quantum computing as well. I could see a optical path potentially interconnecting them all quickly as well. I wonder if AMD will do some Wifi/Bluetooth chiplets too that could even be interesting as a interconnect between different regions of those 3 chips. It might not make sense in all cases, but in some wouldn't be bad and more practical than a multi socket solution for example in like a robotic with different chips in different regions of the robot controlling and analyzing things.

Something that crossed my mind not too far back is something a bit like Vulkan on software and hardware side that numerous players could modify and adapt profile use case schemes for it. A open ended software/hardware Vulkan inspired FPGA combination. Something that's part fixed and yet re-programmable to a degree on the hardware and software side of things perhaps in both area's to a degree. Perhaps FPGA's could be useful for variable rate shading methods especially given it's a emerging field of usage. See that's the beauty to the FPGA being programmable new and emerging API's or revisions to existing ones and ways of doing things could be adapted to current hardware to some degree or another more easily aka partly cognitive programmability.

 

[R-T-B]

MiSTer DIY: Accurately Simulating Retro Game Consoles With FPGAs

Thanks to a tight community of talented developers, the MiSTer project brings cycle-accurate emulation and low latency to Terasic's DE10-nano FPGA

hothardware.com

Yeah. That's cool, but I thought he meant widescale commercial products. That's more a hobbyist toy.

Still, I'll take the correction.

 

[TheoneandonlyMrK]

Can anyone explain what is AMD looking for with Xilinx acquisition? I am lacking deeper understanding. What benefit would AMD gain?

When intel release server chips (or heaven forbid consumer)with an FPGA attached via emib to leverage on the EDGE via their one API and dominate CPU performance in some workloads.
AMD might have an answer.

Plus like all/ most chip developers/designers, the use of FPGA to simulate designs is prolific.

 

[Valantar]

Yeah. That's cool, but I thought he meant widescale commercial products. That's more a hobbyist toy.

Still, I'll take the correction.

That's the thing: if you want to avoid emulation with all its drawbacks, you need either original silicon or an FPGA, and only the latter will do if you also want modern I/O. Of course FPGAs are expensive, which turns this into something (relatively wealthy) enthusiasts (who are the most likely to mind the drawbacks of emulation anyhow). For the rest of us, there are the retail retro mini consoles or RetroPie.

 

[Redwoodz]

Nvidia has pretty much forced AMD into these markets, as we know any success by them will be not be shared with others.... i.e. proprietary tech.

 

[R0H1T]

Forced or not remember the last acquisition by AMD, called ATI back in the day, actually helped them stay relevant & exist today. It wasn't ATI that forced AMD to make *dozers, nor was it ATI that forced them to take the deal with Intel over their OEM bribing circa 2008 (hint ~ it was the worst financial crisis before the current one) & neither did ATI force them to take off the pedal when they were at or near the peak, in CPU market share, just around the time Conroe launched. Yes at the time they overpaid for ATI but without Radeon there'd be no AMD dGPU's, no console deals & probably no AMD today ~ so anyone thinking ATI wasn't a good deal overall, think about what you're saying?

I'm pretty sure Xilinx will do well under AMD, they just need to structure the deal with minimal debt on the books.

 

[Vayra86]

Forced or not remember the last acquisition by AMD, called ATI back in the day, actually helped them stay relevant & exist today. It wasn't ATI that forced AMD to make *dozers, nor was it ATI that forced them to take the deal with Intel over their OEM bribing circa 2008 (hint ~ it was the worst financial crisis before the current one) & neither did ATI force them to take off the pedal when they were at or near the peak, in CPU market share, just around the time Conroe launched. Yes at the time they overpaid for ATI but without Radeon there'd be no AMD dGPU's, no console deals & probably no AMD today ~ so anyone thinking ATI wasn't a good deal overall, think about what you're saying?

I'm pretty sure Xilinx will do well under AMD, they just need to structure the deal with minimal debt on the books.

Its really hard to tell isn't it. Might be true, but the amount of time that went over this... man. One could also defend ATI took away focus within AMD and spread them too thin...

Either way, this seems like a sensible step for them.

 

[Yazzia]

Yeah. That's cool, but I thought he meant widescale commercial products. That's more a hobbyist toy.

Still, I'll take the correction.

All the analogue clone systems use FPGAs, among other systems. They really are the hot thing in retro consoles right now.

 

[$ReaPeR$]

check xilinx's web page and you will understand why AMD wants it, well actually, needs to acquire it.

Xilinx - Adaptable. Intelligent | together we advance_

Xilinx (now a part of AMD) is the inventor of the FPGA, programmable SoCs, and now, the ACAP & delivers the most dynamic processing technology in the industry.

www.xilinx.com

 

[mtcn77]

check xilinx's web page and you will understand why AMD wants it, well actually, needs to acquire it.

Xilinx - Adaptable. Intelligent | together we advance_

Xilinx (now a part of AMD) is the inventor of the FPGA, programmable SoCs, and now, the ACAP & delivers the most dynamic processing technology in the industry.

www.xilinx.com

It is so interesting that they can purchase Xilinx. They are up to snuff.

Xilinx vs Intel (Altera) FPGA performance comparison

Xilinx vs Intel (Altera) FPGA performance comparison - Download as a PDF or view online for free

www.slideshare.net

 

[InVasMani]

I think a big issue with the ATI merger was AMD's management at the time the former CEO ran the company rather poorly. You can defiantly argue AMD probably overpaid perhaps for ATI and that's fine. You could even argue they missed a oppertunity to get Nvidia to which they did the very same CEO that ran AMD into the ground is the one that wouldn't cede control of the company over to Jensen Huang in a merger deal so a Nvidia fell thru as a result. That is actually telling of who was managing AMD at the time of the ATI merger Lisa Su is a much better company CEO. I trust her leadership if a Xilinx deal does happen if anyone guide AMD in the right direction Lisa Su certainly would be one of the best possible candidates to do so the world over. FPGA's have a very broad reach and that's perfect for AMD to branch out further and carve out markets they can thrive within.

 

[XiGMAKiD]

They have to be extremely careful if they're serious about it as currently they're on a roll.

A piece of FPGA in console for developer to tinker with? A *cat core with FPGA on Pi-like board? Sounds good to me.

 

[Valantar]

A piece of FPGA in console for developer to tinker with?

Never, ever going to happen.

A *cat core with FPGA on Pi-like board? Sounds good to me.

You actually want a CPU architecture from ~2015 (Puma+) that was really weak even for its time? Even a single OG Zen core with HT would be better than that.