Mesa Now 2~5x Faster For SPECViewPerf Following OpenGL Optimizations

[bridgman]

... but I would like to argue that funding at least a 22 and a 14nm process definitely would have made a difference.

No argument there - I'm just saying that AMD would not have been able to fund that development.

EDIT: I have firmly believed that it was Dirk Meyers fault for AMD's failure to fabricate competitive products. I mean, I can remember reading his own words on the matter, though I can't find it right now... So without a link I can point to I can't really make that argument... But at that time it was plainly obvious to me that it was clearly his fault...

If you replace the word "fabricate" with "develop" or "design" then again no argument. Dirk had to operate within the constraints of the AMD/GF relationship.

EDIT: At the time of the spinoff, AMD was GF -only- client, they had to -buy- chartered to get their client portfolio. How could GF have funded new fabrication technology when they were definitely operating at an even worse loss than AMD would have? If AMD didn't fund it then who else could have? They were manufacturing AMD's products and they were the only ones who could due to SOI. (And AMD was their only client)

Investment funding came via Mubadala as part of Abu Dhabi's plans to diversify away from oil dependency. I haven't found a good year-by-year report of funding but GF capital spending in 2011 was over $5B and I suspect that nearly all of that came from Mubadala/ATIC investment.

That was the whole idea of spinning off the fabs - getting them lined up with enough financial backing to be able to keep investing in new equipment and processes. My impression is that GF ended up licensing more processes (ST, Samsung, IBM etc..) than they developed, but that still required big investments in equipment to fabricate on those processes.

[duby229]

Um... no. Hector Ruiz was Chairman and CEO until after the fabs had been spun off to Global Foundries, then he left AMD to become GF's chairman.

Dirk Meyer was president, focusing on product development and operations while Hector Ruiz focused on spinning off the fabs and getting GF established.

Decisions like "do we spend several billion dollars to develop new fab processes or do we spin off the fabs" would be board level, not individual executives.

Given that all the same process & fab engineers who would have developed 22nm FDSOI etc... were still working at those fabs just with a GF paycheck, what do you think caused GF's lack of success where you think AMD would have succeeded ?

Don't say "more money" because there would have been less investment available under AMD, not more - we were a $5B/yr company with over $5B of long term debt before making any fab investments.

One possibility might have been to forego acquiring ATI, which would have probably have left AMD with enough borrowing room to fund one more fab process cycle.

Bulldozer as implemented would still have been a problem with its narrow integer units and (AFAICS) less effective prefetching, and Intel was shipping competing products on 65 and 45nm fab processes at the time, so I don't think funding AMD fabs through one more major process step would necessarily have made things a lot better.

Well, I won't argue any more details, but I would like to argue that funding at least a 22 and a 14nm process definitely would have made a difference.

EDIT: I have firmly believed that it was Dirk Meyers fault for AMD's failure to fabricate competitive products. I mean, I can remember reading his own words on the matter, though I can't find it right now... So without a link I can point to I can't really make that argument... But at that time it was plainly obvious to me that it was clearly his fault...

EDIT: At the time of the spinoff, AMD was GF -only- client, they had to -buy- chartered to get their client portfolio. How could GF have funded new fabrication technology when they were definitely operating at an even worse loss than AMD would have? If AMD didn't fund it then who else could have? They were manufacturing AMD's products and they were the only ones who could due to SOI. (And AMD was their only client)

[bridgman]

Um... no. Hector Ruiz was Chairman and CEO until after the fabs had been spun off to Global Foundries, then he left AMD to become GF's chairman.

Dirk Meyer was president, focusing on product development and operations while Hector Ruiz focused on spinning off the fabs and getting GF established.

Decisions like "do we spend several billion dollars to develop new fab processes or do we spin off the fabs" would be board level, not individual executives.

Given that all the same process & fab engineers who would have developed 22nm FDSOI etc... were still working at those fabs just with a GF paycheck, what do you think caused GF's lack of success where you think AMD would have succeeded ?

Don't say "more money" because there would have been less investment available under AMD, not more - we were a $5B/yr company with over $5B of long term debt before making any fab investments.

One possibility might have been to forego acquiring ATI, which would have probably have left AMD with enough borrowing room to fund one more fab process cycle.

Bulldozer as implemented would still have been a problem with its narrow integer units and (AFAICS) less effective prefetching, and Intel was shipping competing products on 65 and 45nm fab processes at the time, so I don't think funding AMD fabs through one more major process step would necessarily have made things a lot better.

[duby229]

I think you mean Hector Ruiz, not Dirk Meyer.

The problem as I understand it was that ongoing fab development was getting exponentially more expensive, much more so than AMD could afford to fund on their own.

Even with a massive amount of new investment via Global Foundries the fab processes were not able to make it to 7nm.

No, I definitely mean Dirk Meyer. Absolutely for sure. Ruiz may have groomed Meyer to spin off manufacturing, but Meyer is the one who did it without -any- means of a 22nm or 14nm process. It -IS- his fault for sure.

It's not so much about losing money as much as it is about they never even -earned- it. It wasn't lost money, it was money they never earned because of Dirk Meyer.

Yeah and look at where that got them, they never even got to 22nm FDSOI, which they -could- have achieved. They never got to 14nm 3DFET SOI, which they also -could- have achieved. (People want to blame Bulldozer, but really it is AMD's failure to invest in fabrication technology that is to blame and that is Dirk Meyer's fault)

EDIT: I don't necessarily think spinning off the fabs was all bad, but spinning off the fabs without a 22nm or 14nm process -WAS- all bad. AMD is lucky they survived it.

[bridgman]

EDIT: AMD's profitability was broken by Dirk Meyer. He's the dude who fucked AMD's fabrication capability, he's the entire reason why BD was never scaled up.

The problem with Bulldozer is that AMD stopped investing in fabrication technology. That was Dirk Meyer's fault. He is squarely to blame, -not- Bulldozer. Bulldozer was truly fantastic.

EDIT: In the years that followed AMD's divestiture of its Fabs it lost -FAR- more money than they ever could have lost in their fab investments. It was -the- biggest mistake AMD ever made.

I think you mean Hector Ruiz, not Dirk Meyer.

The problem as I understand it was that ongoing fab development was getting exponentially more expensive, much more so than AMD could afford to fund on their own.

Even with a massive amount of new investment via Global Foundries the fab processes were not able to make it to 7nm.

[duby229]

Slightly more than highly miss understood. Remember Bulldozer ends up having customers suing them for false advertisement.
https://www.theverge.com/circuitbrea...ulldozer-chips

Its one thing to not have the fabrication capability its another thing to have customers thinking your product is a con job and leaving you with the inventory(the bill). Would not have mattered if Bulldozer was made on a more modern process the fact it was marketed wrong means it was a doomed product.

To be correct your product does not have to be bad to ruin your bank balance as a silicon maker. Miss advertised is very dangerous this is goof it.

The number of things silicon design companies have to get right over and over again not to end up in lack of hell is a lot.

No, I think -you- misunderstanding right now!

AMD screwed up when they called it 8 cores, the fact is, imo, that the Modules -was- the cores and that it only ever was 4 cores. It was 4 cores with 8 CMT threads. That's what the false advertising was all about. They advertised it as 8 cores/8 threads, it wasn't, it was 4 cores/8 CMT threads... Additional to that each CMT thread only had access to two integer units...

Now imagine a 7nm bulldozer with 16cores(modules)/32 CMT threads where each CMT thread has access to 3 integer units... Zen would probably still beat it in single threaded loads, but such a processor would annihilate Zen in multithreaded loads.

EDIT: Bulldozers -biggest- problem (far bigger than the false marketing scheme) was that AMD sold their fabs and stopped investing in their capability to fabricate new products. If AMD had invested in a 22nm SOI process then Bulldozer could have been scaled up further than it was. If AMD had invested into a 14nm 3DFET SOI process then Bulldozer could have wiped the floor with anything Intel could offer.

The problem with Bulldozer is that AMD stopped investing in fabrication technology. That was Dirk Meyer's fault. He is squarely to blame, -not- Bulldozer. Bulldozer was truly fantastic.

EDIT: In the years that followed AMD's divestiture of its Fabs it lost -FAR- more money than they ever could have lost in their fab investments. It was -the- biggest mistake AMD ever made.

[oiaohm]

All I'm saying is that Bulldozer was highly misunderstood. It was by far the most scalable x86 architecture ever conceived. The problem with it was only that they never manufactured it on a more modern process and they never scaled it up to its potential. It was a fantastic architecture.

Slightly more than highly miss understood. Remember Bulldozer ends up having customers suing them for false advertisement.

https://www.theverge.com/circuitbreaker/2019/8/28/20837336/amd-12-million-false-advertising-class-action-lawsuit-bulldozer-chips

Its one thing to not have the fabrication capability its another thing to have customers thinking your product is a con job and leaving you with the inventory(the bill). Would not have mattered if Bulldozer was made on a more modern process the fact it was marketed wrong means it was a doomed product.

To be correct your product does not have to be bad to ruin your bank balance as a silicon maker. Miss advertised is very dangerous this is goof it.

The number of things silicon design companies have to get right over and over again not to end up in lack of hell is a lot.

[duby229]

Except AMD position has not been the best. With the different failure of product lines AMD got themselves critically in debt and it taken 10 years to dig their way out. Just because a company is now a huge company with decent amount of free cash to speed does not mean they were that way 12 months ago or 2 years ago. As AMD debt problem reducing more money is going into driver development in general.

Its really simple sandy8925 to miss how much the product flops of the past dug AMD into a financially troubled location. Comes hard when you are choosing between more developers and meeting debt repayment requirements and keeping share holders happy with their dividend.


That not would have they have been over their Windows drivers due to lack of developers working in that area as well.

This is a nightmare for a silicon company like AMD. Designing and producing silicon is expensive as hell with no returns policy. You goof it like with the AMD Bulldozer CPUs from 2011 the effects on the companies operations hurt the next made products as well the Debt of the mistake means you don't have the money to invest to properly make out the next products.

There have been a nice chain of product goofs with AMD causing nice high debt levels. This harms the amount of developer resources you have.

Horrible as it sounds like it or not AMD has not been in a location just to throw money at developers and say make me drivers. Instead have had to be attempting to exactly right number of developers to make the quality drivers to have as much money left over to clear debt. Please note AMD failing to clear debt will mean if they make another Bulldozer like product so double stack the debt they will be dead.

Silicon production yes you are talking billions of dollars game but its really gambling with billions of dollars and if you get it wrong its really wrong.

AMD still needs a few more successful products to be clear of the AMD Bulldozer CPU caused debt. We really do hope AMD history stops repeating. This is history just getting out of debt so able to start good investment in development then the next product be goof and puts them straight back in.

All I'm saying is that Bulldozer was highly misunderstood. It was by far the most scalable x86 architecture ever conceived. The problem with it was only that they never manufactured it on a more modern process and they never scaled it up to its potential. It was a fantastic architecture.

EDIT: AMD's profitability was broken by Dirk Meyer. He's the dude who fucked AMD's fabrication capability, he's the entire reason why BD was never scaled up.

https://en.wikipedia.org/wiki/Dirk_Meyer

[Yoshi]

someday AMD will have a GUI panel control

You mean something like this?
GitHub - BoukeHaarsma23/WattmanGTK: A Wattman-like GTK3+ GUI

It's not official AMD but it works just fine

[bridgman]

Yeaaaahhhh, that's a bullshit excuse. AMD isn't some three-person band working out of a garage, they're a huge company with hundreds of millions of dollars. They can afford to hire enough developers to write proper quality drivers for all of their hardware and have it ready ahead of release.

Its really simple sandy8925 to miss how much the product flops of the past dug AMD into a financially troubled location. Comes hard when you are choosing between more developers and meeting debt repayment requirements and keeping share holders happy with their dividend.

I'm not disagreeing with you other than to say "it was much worse than that". The choice was between hiring more developers and being able to pay the ones we had. I don't believe AMD has ever paid a dividend.

Guest remember that AMD lost money almost every quarter from late 2006 through 2017, more-or-less broke even until 2019, and has only really gotten back to normal operation in the last year.

If review and benchmark sites were told that drivers/CPU software wasn't ready at launch, and will only be released a few months later, or that they have to build unstable software from scratch themselves, in order to benchmark/review the hardware, they would NOT be providing favourable reviews for AMD. AMD would have been thrashed publicly.

I don't understand this comment. Are you suggesting that we are somehow lying to the reviewers and that they don't do any testing themselves ?