• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

Discussion [WSJ]AMD Is in Advanced Talks to Buy Xilinx

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

piokos

Senior member
Nov 2, 2018
554
202
76
Francois thinks the shareholders of Xilinx should turn down the deal.

If you look at the top 10 institutional investors of AMD and Xilinx, both lists are very much alike (to be honest: Intel's as well).
In other words: it's not "AMD shareholders" vs "Xilinx shareholders". It's just "shareholders". If they think this deal makes sense for both companies, they'll push it on both sides.

This is all built around rumors at this point.
AMD fans went berserk and praise this as a great decision by AMD (or even the saint Su in particular).
For some very weird reason hardly anyone here considers that this could be inspired by the shareholders and AMD directors could be against it. :D

Maybe shareholders want AMD to be more diversified (like Intel is), but current AMD team would prefer to focus on what they know?

Or what if this is more about securing Xilinx partnership as a move against Intel?
Let's for a moment imagine FPGA becomes very welcome in datacenters and embedded (inc. IoT). Maybe in PCs as well.
Unlike Intel, AMD doesn't make them. They have to find a supplier. Xilinx and Intel are the top 2 manufacturers. What if Intel's offer is much better?
AMD would have to choose them - potentially sharing big chunk of profits with the main CPU competitor. :)
If they become connected to Xilinx (a holding would probably be enough - they don't have to merge), they can choose Xilinx as the supplier even if it's more expensive. :)
 

piokos

Senior member
Nov 2, 2018
554
202
76
Francois thinks the shareholders of Xilinx should turn down the deal.
There's also one more decent reason why shareholders may want such a deal to happen. But I needed a coffee to notice it. :) Cashflow!

AMD doesn't pay a dividend, which is a huge pain for the investment funds.
This means the only way for institutional investors to get a positive cashflow from the investment is to sell the stock - something that doesn't look interesting as long as AMD has potential to grow (bubble or not).

However!
If AMD offers $30B of additional shares to buy Xilinx, it's almost like if they paid a one-time dividend of 30% - at least from the perspective of common shareholders (Vanguard, BlackRock, State Street and so on).
 
  • Like
Reactions: lightmanek

Ajay

Diamond Member
Jan 8, 2001
7,417
2,602
136
1) Is this confirmed?
2) What does AMD bring to the table that Xilinx needs?

I have a tough time seeing this as a worthwhile deal.
 

piokos

Senior member
Nov 2, 2018
554
202
76
2) What does AMD bring to the table that Xilinx needs?
Large sales potential for datacenters, supercomputers and everything else.

Since both companies outsource manufacturing to TSMC, they would likely save on fabrication.
 

beginner99

Diamond Member
Jun 2, 2009
4,625
1,012
136
The FPGA market is the product area that Xilinx has targeted, but they have a broad & deep expertise in chiplet tech. They are ~2 years ahead of even AMD in high performance, high bandwidth, chiplet based computing. This is an essential skillset for the future.
If true that would actually explain it, somewhat. Source for this information?
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,114
1,584
136
it could also work quite well on the DC AI side, AMD seem to want to take a general compute path, ( standard SIMD with AI specific formats) while Nvidia is happy to go down the specific accelerator. xilinx based AI core's as a chiplet on a package/socket solution with any combination of GPU / CPU chiplets would fit right into AMD's vision.

Would be interesting if on the xilinx edge compute side if AMD's IP would be a worthwhile inclusion on its SOC's? for example Tesla's self driving Chip has a fp 16/32 GPU thats about 1/2 the size of the NPU.
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,114
1,584
136
IMO it would be better if they acquire small talent company
this a full scale corporation
merging together will distract the good engineers to politics
thats a risk but it doesn't always happen,

This would be a pretty atypical merger, most big mergers are medium company into massive company and the medium just get squashed in. This is more medium to medium, smart leadership with little to no bean counter "synergises" ie firing people. Run both as separate lines to begin with , work on "collaborations" and slowly over a few years integrate more and more.

Thats the way I would do it :) .

The company I work for was ~27k people was bought by a company of 400k we slowly over 7 years merged into the borg ( ive been there for all of it) , its been a rather painless process.
 

piokos

Senior member
Nov 2, 2018
554
202
76
30 billion dollars in AMD stock?
Ekhm... I though you were thinking about a public offering and buying Xilinx with money. That made sense.

Paying with stock is convenient when a listed company buys a private company.
Xilinx is a large, listed company. They would have to take it off the stock exchange first.
Whereas just normal stock takeover can be done over many months and AMD only needs 50%+1 votes to take control.
Save on fab may be a bit optimistic, but, with their combined volume and diversity of portfolio, they will have a bit more pull with TSMC for sure.
Better position in negotiations is what I meant.

But they could also share the reserved volume. That's always a bit more flexibility in production.
 

maddie

Diamond Member
Jul 18, 2010
3,333
2,212
136
If true that would actually explain it, somewhat. Source for this information?
This is all I saw with a quick search. Years ago, around the time of Fury graphics cards, I remember looking at XILINX's work as a window to the future as they were far advanced from AMD in very high bandwidth applications. Their info at the time, mentioned a 1-2 ns delay between chiplets when using a interposer. Note the dates.

Wikipedia: ( note this is on 28nm)

In 2011, Xilinx began shipping sample quantities of the Virtex-7 2000T FPGA, which combines four smaller FPGAs into a single package by placing them on a special silicon interconnection pad (called an interposer) to deliver 6.8 billion transistors in a single large chip. The interposer provides 10,000 data pathways between the individual FPGAs – roughly 10 to 100 times more than would usually be available on a board – to create a single FPGA.[30][31][32] In 2012, using the same 3D technology, Xilinx introduced initial shipments of their Virtex-7 H580T FPGA, a heterogeneous device, so called because it comprises two FPGA dies and one 8-channel 28Gbit/s transceiver die in the same package.[33]

As Xilinx introduced new high capacity 3D FPGAs, including Virtex-7 2000T and Virtex-7 H580T products, these devices began to outpace the capacity of Xilinx’s design software, which led the company to completely redesign its tool set. The result was the introduction of the Vivado Design Suite, which reduces the time needed for programmable logic and I/O design, and speeds systems integration and implementation compared to the previous software.


Award nominee from 2013: https://www.3dincites.com/2013/06/xilinx-virtex-7-h580t/

Testimonial
As the world’s first 3D heterogeneous all programmable FPGA, the Virtex-7 H580T FPGA has clearly advanced the state-of-the-art in 3D IC technology. Like other Xilinx 3D IC devices, the Virtex-7 H580T FPGA addresses the challenges of interconnecting multiple FPGAs utilizing Xilinx’s Stacked Silicon Interconnect (SSI) technology, enabling high-bandwidth connectivity between multiple die and providing a 100x improvement in inter-die bandwidth per watt compared to multi-chip approaches. It also imposes much lower latency and consumes dramatically lower power than multi-FPGA or multi-chip module approaches, while enabling the integration of transceivers and on-chip resources within a single package. SSI technology leverages proven microbump technology combined with coarse pitch through-silicon vias (TSVs) on a passive (no transistors) 65nm silicon interposer to deliver high reliability interconnect without performance degradation on one FPGA device. This breakthrough technology provides the next level of advanced system integration for applications requiring high logic density and tremendous computational performance.



Their custom software tools alone must be unique.
 
  • Like
Reactions: Tlh97 and Vattila

gorobei

Diamond Member
Jan 7, 2007
3,089
243
106
xdf 2018 presentation by their vp of hardware Liam Madden. mark papermaster spoke there as well. their ACAP chip is a heterogeneous chip with 2 different arm cores + fpga + custom ai engines. their interposer tech is the basis of what everyone else is using. they also integrated dsp for 5g radio. the more relevant stuff to amd is in the back half of the presentation, he goes over what they did and why.
 

Saylick

Senior member
Sep 10, 2012
861
513
136
xdf 2018 presentation by their vp of hardware Liam Madden. mark papermaster spoke there as well. their ACAP chip is a heterogeneous chip with 2 different arm cores + fpga + custom ai engines. their interposer tech is the basis of what everyone else is using. they also integrated dsp for 5g radio. the more relevant stuff to amd is in the back half of the presentation, he goes over what they did and why.
It almost seems too obvious that you could swap out the Arm cores with Zen and the AI vector engine with CDNA...
 

slashy16

Member
Mar 24, 2017
151
58
71
40
Do you have to post negative things about every piece of news about AMD ? Are you an out of work Intel employee or something ?
Im amazed every time I visit this forum that cancer hasn’t blessed us by taking you away.

You ABSOLUTELY can not say anything like this ANYWHERE on the forums.
If it ever happens again, you will be gone for a significant amount of time.

AT Mod Usandthem



You will not say anything like this here, ever again.

Goodbye.


esquared
Anandtech Forum Director
 
Last edited by a moderator:
  • Wow
Reactions: spursindonesia

gorobei

Diamond Member
Jan 7, 2007
3,089
243
106
It almost seems too obvious that you could swap out the Arm cores with Zen and the AI vector engine with CDNA...
there may not be any need to replace the arm cores. Will Smith of Tested fame on his podcast indicated that the arm cores on the ML and HPC servers are more for coordinating data traffic to the vector/gpu parts. so the cpu just has to be good enough to feed the array.

from a manufacturing perspective, this may be about heterogenous computing and amd's custom designs. with chiplets and interposers amd will be all about lego block mix match of cpu/gpu/ai vector array/etc. incorporating any customer's ip or asic into an interposer is fine for large volume contracts but probably wouldnt be worth it for small runs. by making a standardized fpga chiplet that fits into amd's volume product packaging they can offer 2 or 3 skus with programmable logic for the customer to tweak after purchasing. they can produce in volume and the smaller customers dont have to deal with incorporating their asic into the packaging. amd acquiring xylinx means they can simplify designs and sku while getting economy of scale.
 

piokos

Senior member
Nov 2, 2018
554
202
76
Is english your second language?
Otherwise, what you just said should get you a ban, IMO.
Maybe it should.

Which doesn't change the fact that Markfw is a highly polarizing individual, who takes part in (or provokes) many of these situations.
He's extremely active and with very strong opinions, despite not having that much knowledge on the matter (I think the recent ARM topic really confirmed that).
He's also a long lasting member (among other things...) - so, I'd imagine, likely in good terms with the rest of the staff.

And more and more people will get bans because they don't agree with him (even in a much more polite manner).
Is this really the way forum staff wants to run this place? As a one man show? I find this amusing.


Discussing moderator actions in a thread is not allowed. Furthermore, implying the additional moderators
are in some type of "quid pro quo" is questing our integrity, also not allowed.

You've been told this at least four times now, but there is only one place to discuss moderators decisions/actions,
and that is to create a thread in Moderator Discussions.

AT Mod Usandthem
 
Last edited by a moderator:

Zucker2k

Golden Member
Feb 15, 2006
1,234
594
136
And more and more people will get bans because they don't agree with him (even in a much more polite manner).
This. If only he's as tolerant of opposing views as others are of his, but, oh well.

In any case, no member should wish bad things on another (including bans, mind you) simply because they expressed an undesired opinion about hardware.
 

ASK THE COMMUNITY