• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Could Feature Next Generation HBM Memory on Volcanic Islands 2.0 Graphics Cards – Allegedly Laun

http://www.synapse-da.com/Corporate/Clients

ABzFRth.jpg

Interesting.

The 350mm^2 looks like it would be the 380X as its the same size as the 280X

The 500mm^2 is a little confusing, unless they have shrunk the die again without actually shrinking it (like they did going from Tahiti to Hawaii) its not big enough to have 4000 cores, Hawaii with 2816 is 430mm^2. 500mm^2 would give it about 3300 cores.

Having said all of that you would have though they would do the same shrink to the 380X that they did with Hawaii, it would then be about 250mm^2.
 
ok after reading more into it, it would seem that this company designs chips, which if I'm reading this correctly it means that AMD haven't even designed their own chips they farm that out, as well as everything else. :p


Rather disappointing to find out really if I'm honest.
 
THe information to take from it is HPM... TSMC don't have a process at 28nm called HPM, global foundries do.

Secondly, AMD is supposed to be introducing high density libraries around the 2014-2015 time frame... meaning they could get anything up to a 20-30% die size reduction.

Lastly they claim a 500mm^2 + core tape out in the slide. Being that it's unlikely anyone else they work with on their list of clients makes large gpu's in such sizes and with such high clocks. You might find some custom client making something for HPC or, some other use but outside of gaming it's rare to want high power usage, it's usually better to scale things and run multiple cores at lower clocks. SO it's probably helped on taping out products for AMD at Glofo.
 
ok after reading more into it, it would seem that this company designs chips, which if I'm reading this correctly it means that AMD haven't even designed their own chips they farm that out, as well as everything else. :p


Rather disappointing to find out really if I'm honest.

That isn't really what they do, but well done.

Company A designs a chip, company B implements the design, company C produces the design, company D tests the design.

A/B/C/D can all be the same company, or any combination of companies, or a couple the same, a couple different.

Sometimes you have a project that needs extra work, rather than hiring 200 people who you only need for 3 months extra work, you hire a company with 200 guys working for them who can help you for 3 months instead.

They basically don't work on design, at all, just helping customers implement a design, IE take a cpu design and help any stage in getting it to a finished set of masks which TSMC/any other fab can take and produce. They can also work with TSMC/whoever, in ramping up yields, testing and getting it to be a finished product.

If you think Intel haven't hired temporary people or outsourced anything, you're made, all huge companies do so on some things.

With some companies they get more help, with others it will be a very small part they get help with, with others the company is capable of doing A through D themselves but due to time constraints they simply outsource some of the work.

It's quicker to contract Synapse to get 50 people for 3 months than to interview 500 people, to find 50 people you need, of which almost none will want to leave their current jobs for a 3 month contract.

But you're always negative about AMD regardless of if you understand what is being talked about. If AMD had never designed their own chip and outsource anything, so what?
 
Last edited:
But you're always negative about AMD regardless of if you understand what is being talked about. If AMD had never designed their own chip and outsource anything, so what?


Of course all you see is me ****ging off AMD.

You completely missed the stick out tongue smiley or the "Rather disappointing to find out really if I'm honest."



You go right ahead and believe what you want too.

Do I prefer Nvidia to AMD? Yes without a doubt.

Do I recommend AMD products, to friends and family, Yes a lot of the time.

Does this always seem to come back and bite me in the backside? yes it certainly has this weekend.

Does any of this have any bearing on my stick out tongue comment in the previous post? No not at all.

Does this mean that maybe Nvidia don't all ways design their own chips? More than probably.

Does that disappoint me as well? definitely.

Does this info disappoint any one else? well I would say yes definitely but it would be best for people to answer for them selves.


oh yes and thanks for calling me stupid
 
Last edited:
A long time ago there was a story about AMD layoffs (maybe this? http://www.theregister.co.uk/2011/11/03/amd_restructuring_job_cuts/) - I recall reading some comments I didn't understand at the time about part of the workforce being let go being chip designers. People were upset at AMD moving to "automated design"? Anyone remember this and can correct my memory?

Yeah, I vaguely remember something about the people not liking the automated design thing. Can't actually add anything to what you've said, but at least you weren't imagining it!
 
A long time ago there was a story about AMD layoffs (maybe this? http://www.theregister.co.uk/2011/11/03/amd_restructuring_job_cuts/) - I recall reading some comments I didn't understand at the time about part of the workforce being let go being chip designers. People were upset at AMD moving to "automated design"? Anyone remember this and can correct my memory?

They down sized as they had a work force twice the size of Nvidia, Down sizing does not = sacking everyone in a department.

AMD don't do everything in house, they don't design a given architecture in house in its entirety, or if they do they don't always use everything designed in house, like most good companies they will farm out the same goal to multiple vendors, as well as do it in house, who ever comes up with the best solution will be the one thats used.

i would be very surprised if Intel and Nvidia don't do the same.

Bru, is just putting 2 and 2 together and thinking he has come up with an answer "AMD don't do anything, they have others do it for them" he likes that answer so much he's latched on to it.
Of course he can't see through his joy how wrong he is.
 
Last edited:
Bru, is just putting 2 and 2 together and thinking he has come up with an answer "AMD don't do anything, they have others do it for them" he likes that answer is so much he's latched on to it.
Of course he can't through his joy of that how wrong he is.

erm before reading this thread and finding out what goes on at the synapse company, it never occurred to me that companies like AMD/Nvidia/Intel would farm out designs. Now maybe that in itself makes me stupid/naive, but the fact remains that I am disappointed to find this out. Now tell us 'honestly' when you first found out that AMD don't do these tings in house, you weren't disappointed?

And you have never posted in a thread a comment which could be taken humorously with a smiley face at the end.
 
On a completely side note has anyone actually found the page that slide is on, on the website?
 
erm before reading this thread and finding out what goes on at the synapse company, it never occurred to me that companies like AMD/Nvidia/Intel would farm out designs. Now maybe that in itself makes me stupid/naive, but the fact remains that I am disappointed to find this out. Now tell us 'honestly' when you first found out that AMD don't do these tings in house, you weren't disappointed?

And you have never posted in a thread a comment which could be taken humorously with a smiley face at the end.

As I and drunkenmaster explained; Just because synapse have AMD in a partner does not mean they do everything for AMD, if you think that then you are naive.

This does not disappoint me at all, it stands to reason that they do this, so will Intel and Nvidia, to think they don't is just as naive.

What they are doing is hunting around through multiple specialised companies for the best solutions, sometimes those solutions are better than their own.

I want them to use the best solutions, why would i be disappointed in that?
 
Last edited:
Now I see what has happened here, when I originally said...
ok after reading more into it, it would seem that this company designs chips, which if I'm reading this correctly it means that AMD haven't even designed their own chips they farm that out,as well as everything else. :p

I was referring to these two chips listed in the image that Orangey posted up, not everything that AMD have ever made, as that really would be bizarre.
the 'as well as everything else' was a quip about the way AMD like to get others to do the work for them.

Now if you thought I meant everything AMD has ever designed then reread what I said and if I was meaning everything then why didn't I use ever instead of even.

Anyway I am still disappointed about it, and Humbug, before you it stood to reason that they did this there was a time when you didn't know, where you disappointed to find out ?
 
They still don't farm out the design and Synapse still don't do actual design or IP, for the most part they translate a design to production.

AMD do the design, they are fully capable of taking a design all the way through to finished product. Sometimes they'll outsource part of that process, sometimes they won't most likely.

One of the big deals with laying out a chip very specifically, and there were threads on comments from an ex AMD guy who said they were giving away 5-10% performance/power by doing it automatically. Is that there comes a stage where, when you are manually laying out small blocks of either hundreds or several thousand transistors, and the entire chip is 100million transistors, that is fine and doable. When it goes 200mil, still doable, 400mil... starting to be a lot of work, 800mil... wow 1.6billion, yowzer. 3.2billion uh oh.

Ultimately there comes a time where you need to go from a manual to a automatic process because anything else is simply not viable. 10% performance is great, but if you go from 10 guys taking a year to 200 guys, your costs just went up exponentially, and effectively these would continue to rise with every doubling of transistor count.

This is where HDL libraries come in, automatic process isn't perfected, it takes time, hence the high density libraries for the tools that carry out the work. They have stated that a Piledriver(I believe it was) would have been 30% smaller and less power(if that is both reduced by 30%, or pick and chose, 30% smaller same power, 30% less power same size, or go 15/15, it was unclear, either way that is either great or epic). I tried to look it up, it's perfectly possible that Synapse helped them with the libraries, but I get the impression that part is still done by AMD as I can't see any references to the HDL and it being outsourced. Though to an engineer, going from 10-50 engineers doing it manually to letting a computer doing it automatically would basically be the same thing if they get fired.


It's likely Synapse employs a bunch of ex AMD, ex Intel, ex IBM, ex TSMC, ex UMC and ex Apple engineers to do the work anyway. So it could literally be the same group of people doing the same work, just under a different company name and no I wouldn't be disappointed.

There is a reason AMD spun off the foundry, why Intel are now taking on more foundry customers than ever before, why IBM have sold off their foundries, why Semi Chartered sold their foundries(taiwanese or sinagpore group of smaller fabs that GloFo bought shortly after being formed). A single company, even Intel, can't continue to do everything themselves.

Because taping out is pretty much process specific(many parts are, the design rules etc) then one company doing all the work on 28nm at TSMC and everyone using them makes much more sense than 50 companies making chips to all do the same work. You would have taping out specialised workers... who tape out say 3 products a year... rather than hire those specialised workers for the 3 months of a year that you need.

I forget the company name, something like Applied Materials, who make all of TSMC's fab equipment, and they make all of Intel's equipment. Share the development costs or double the work and make everything more expensive. Like object oriented programming, use the best, most efficient solution to everything. Architecture, do yourself, anything else, if someone else can do it quicker, or better, or both, why would you throw away money doing it yourself?

This is more of an issue since the spin off of GloFo, GloFo needed to become more industry standard, and as such more people specialise in their processes and gain experience of it. They taped out 33 products, AMD didn't, they have more experience taping out products than AMD at 28nm because they do it more often. To not utilise that experience is simply daft.

I find it odd that anyone thinks subcontracting out work, when talking about the scale of billion dollar companies.... is disappointing. 10 guys employed by AMD do a job, or 10 guys who work for Synapse, who are employed by AMD, do the job.... I see literally no difference at all.

Look at any huge company, call every department a different name, no different. When you get to huge scale companies, smart decisions keep you afloat, bad decisions see you go bankrupt. AMD are making smart decisions now, part of that is outsourcing where required.
 
Last edited:
They still don't farm out the design and Synapse still don't do actual design or IP, for the most part they translate a design to production.

It's like an architect/designer, designing a house but then passing it on to another team to draw up more detailed blueprints or work out some very specific things like which materials to use. It could be anything, one architecture will design and spec the entire thing, make the blueprints, work out every material, every mm, layout every wire. Another architect focuses on design and passes the smaller scale detail to others, another guy does the design, the blueprints but passes off a small part of it to a specialist.

AMD do the design, they are fully capable of taking a design all the way through to finished product. Sometimes they'll outsource part of that process, sometimes they won't most likely.

The architecture is almost certainly done exclusively in house, though you have various teams in a company working on different projects and sometimes they'll have a better idea than another team for getting something to work. There isn't much reason purposefully setting the same goal to multiple different teams to come up with the best thing, that would be insanely expensive and negate the point of hiring some of the best people in the business if you have another team without them trying to do the same thing.

One of the big deals with laying out a chip very specifically, and there were threads on comments from an ex AMD guy who said they were giving away 5-10% performance/power by doing it automatically. Is that there comes a stage where, when you are manually laying out small blocks of either hundreds or several thousand transistors, and the entire chip is 100million transistors, that is fine and doable. When it goes 200mil, still doable, 400mil... starting to be a lot of work, 800mil... wow 1.6billion, yowzer. 3.2billion uh oh.

Ultimately there comes a time where you need to go from a manual to a automatic process because anything else is simply not viable. 10% performance is great, but if you go from 10 guys taking a year to 200 guys, your costs just went up exponentially, and effectively these would continue to rise with every doubling of transistor count.

This is where HDL libraries come in, automatic process isn't perfected, it takes time, hence the high density libraries for the tools that carry out the work. They have stated that a Piledriver(I believe it was) would have been 30% smaller and less power(if that is both reduced by 30%, or pick and chose, 30% smaller same power, 30% less power same size, or go 15/15, it was unclear, either way that is either great or epic).


At the end of the day anyway, it's likely Synapse employs a bunch of ex AMD, ex Intel, ex IBM and ex Apple engineers to do the work anyway. So it could literally be the same group of people doing the same work, just under a different company name and no I wouldn't be disappointed.

There is a reason AMD spun off the foundry, why Intel are now taking on more foundry customers than ever before, why IBM have sold off their foundries, why Semi Chartered sold their foundries(taiwanese or sinagpore group of smaller fabs that GloFo bought shortly after being formed). A single company, even Intel, can't continue to do everything themselves.

Because taping out is pretty much process specific(many parts are, the design rules etc) then one company doing all the work on 28nm at TSMC and everyone using them makes much more sense than 50 companies making chips to all do the same work. You would have taping out specialised workers... who tape out say 3 products a year... rather than hire those specialised workers for the 3 months of a year that you need.

I forget the company name, something like Applied Materials, who make all of TSMC's fab equipment, and they make all of Intel's equipment. Share the development costs or double the work and make everything more expensive. Like object oriented programming, use the best, most efficient solution to everything. Architecture, do yourself, anything else, if someone else can do it quicker, or better, or both, why would you throw away money doing it yourself?

This is more of an issue since the spin off of GloFo, GloFo needed to become more industry standard, and as such more people specialise in their processes and gain experience of it.

THey taped out 33 products, AMD didn't, they have more experience taping out products than AMD at 28nm because they do it more often. To not utilise that experience is simply daft.

all thoze wordzz :eek:
 
They still don't farm out the design and Synapse still don't do actual design or IP, for the most part they translate a design to production.

AMD do the design, they are fully capable of taking a design all the way through to finished product. Sometimes they'll outsource part of that process, sometimes they won't most likely.

One of the big deals with laying out a chip very specifically, and there were threads on comments from an ex AMD guy who said they were giving away 5-10% performance/power by doing it automatically. Is that there comes a stage where, when you are manually laying out small blocks of either hundreds or several thousand transistors, and the entire chip is 100million transistors, that is fine and doable. When it goes 200mil, still doable, 400mil... starting to be a lot of work, 800mil... wow 1.6billion, yowzer. 3.2billion uh oh.

Ultimately there comes a time where you need to go from a manual to a automatic process because anything else is simply not viable. 10% performance is great, but if you go from 10 guys taking a year to 200 guys, your costs just went up exponentially, and effectively these would continue to rise with every doubling of transistor count.

This is where HDL libraries come in, automatic process isn't perfected, it takes time, hence the high density libraries for the tools that carry out the work. They have stated that a Piledriver(I believe it was) would have been 30% smaller and less power(if that is both reduced by 30%, or pick and chose, 30% smaller same power, 30% less power same size, or go 15/15, it was unclear, either way that is either great or epic). I tried to look it up, it's perfectly possible that Synapse helped them with the libraries, but I get the impression that part is still done by AMD as I can't see any references to the HDL and it being outsourced. Though to an engineer, going from 10-50 engineers doing it manually to letting a computer doing it automatically would basically be the same thing if they get fired.


It's likely Synapse employs a bunch of ex AMD, ex Intel, ex IBM, ex TSMC, ex UMC and ex Apple engineers to do the work anyway. So it could literally be the same group of people doing the same work, just under a different company name and no I wouldn't be disappointed.

There is a reason AMD spun off the foundry, why Intel are now taking on more foundry customers than ever before, why IBM have sold off their foundries, why Semi Chartered sold their foundries(taiwanese or sinagpore group of smaller fabs that GloFo bought shortly after being formed). A single company, even Intel, can't continue to do everything themselves.

Because taping out is pretty much process specific(many parts are, the design rules etc) then one company doing all the work on 28nm at TSMC and everyone using them makes much more sense than 50 companies making chips to all do the same work. You would have taping out specialised workers... who tape out say 3 products a year... rather than hire those specialised workers for the 3 months of a year that you need.

I forget the company name, something like Applied Materials, who make all of TSMC's fab equipment, and they make all of Intel's equipment. Share the development costs or double the work and make everything more expensive. Like object oriented programming, use the best, most efficient solution to everything. Architecture, do yourself, anything else, if someone else can do it quicker, or better, or both, why would you throw away money doing it yourself?

This is more of an issue since the spin off of GloFo, GloFo needed to become more industry standard, and as such more people specialise in their processes and gain experience of it. They taped out 33 products, AMD didn't, they have more experience taping out products than AMD at 28nm because they do it more often. To not utilise that experience is simply daft.

I find it odd that anyone thinks subcontracting out work, when talking about the scale of billion dollar companies.... is disappointing. 10 guys employed by AMD do a job, or 10 guys who work for Synapse, who are employed by AMD, do the job.... I see literally no difference at all.

Look at any huge company, call every department a different name, no different. When you get to huge scale companies, smart decisions keep you afloat, bad decisions see you go bankrupt. AMD are making smart decisions now, part of that is outsourcing where required.

+1

Totally agree

To put it in another context, did Irvine Sellar build the Shard? No, he bought in investors and architects to aid the end goal. Did the investors / architects build it? No they appointed a main contractor Mace. Did Mace build it? No they employed many subcontractors and specialist trades. Does any of this detract from the end product? Absolutely not. If it wasn't for the many individuals and organisations clubbing together it would never of been built or been any kind of success whatsoever.

The point is this semiconductor business has got to be one of the hardest things on the planet to design, build and get functioning correctly. Its completely naieve to think AMD or nVidia do the whole thing from start to finish themselves. Theres surely outsourcing at some points in the process. Does this detract from the overall functionality or end user enjoyment of the product? No. In fact it probably aids it, because otherwise we wouldn't be getting any gpus.. Or they would be years late to the table and horrendously expensive.

ZOMGZ nVidia / AMD outsource part of the process to GloFo / TSMC lets all be dissapointed in the product... Its either that or refer to the last line above.

If they could be doing it cheaper themselves they would. Just remember who stumped up a lot of the initial costs of 20nm, Apple. AMD / nVidia should hopefuly ride off the back of this a little bit and cheaper gpu's for all of us. Does that make me dissapointed? No.
 
Back
Top Bottom