Dammit! This sucks if it is indeed true. I have been waiting patiently to upgrade my socket 939 3700+ cpu . I purposely avoided SB because of the socket changes and upgrade path. Hell even the option of getting a AM3+ board with a cheap X3 445 CPU for now and then popping in a BD when they finally come out may be in jeopardy if the whole platform turns to crap. Hopefully this is just a minor delay and while AMD might not hit a homerun with BD at least an extra base hit would be nice.
There are way too many contradicting reports to make any educated guess on Bulldozer's performance. It sort of sucks.....cause there's another delay and chances are, AMD is just gonna keep things quiet until it gets released. I'm just so curious to see what Bulldozer will perform like.
The thing with BD is they haven't release official clock speeds. And these 3.8 GHz number they are putting out without reference to how many cores are running could mean both Anandtech and the translated Google link is correct.
The 3.8GHz speed could mean only when 4 cores is running and the 2.5 GHz speed in the translated link could mean when all 8 cores is running.
Assuming the 3.8 GHz speed is for the 4 core, and assuming BD has a 10% improvement in IPC over X6 (some articles have suggested BD may even be slower then X6 clock for clock), it'll still be slower then the i7 2600K since in most benchmarks Intel is 30-50% faster then the X6.
Now for the translated article that says the 8-core BD at 2.5 GHz is slower then the X6. That is probably true as well, since the X6 has six cores at 3.3 GHz, that`s effectively 19.8 GHz of CPU power. 8-core BD at 2.5GHz would have an effective 20 GHz of CPU power... but because some of the resources of the BD modules are shared; early articles have said that 2 cores doesn`t double performance you get about 80-90% compared to a real extra core. So the 20 GHz of total CPU power would really only give you 16-18 GHz of performance making it slower then the X6.
I thought we had some NDA-breaking dude in the forum here who swore his BD samples were the bomb? (paraphrasing, but that was the gist of his posts)
That said...didn't Phenom have similar issues on release day clockspeeds in terms of not really besting the prior gen top-speed X2's?
I feel for the AMD engineers and GloFo engineers right now, I know that crisis-mode mad-dash work environment that they are all surely in the middle of right now. What a crappy way to spend your summer, late nights and 6-7 day work weeks as you rush around the hot lots and running from one the data analysis team meeting to the next
Its likely just a yield issue. Maybe they need 3.8GHz to be competitive with 2600K and they can get some 3.8GHz chips with B1 stepping but the yields are too low to justify volume release at this time.
But given that we are so close to the purported original release timeline before AMD decided to delay Bulldozer, that means they are either sitting on a fab full of borked unsellable B1 stepping zambezi's that were in the pipeline ready to support a solid launch here in a week OR they had not really intended this to be a hard-launch.
That is going to hurt to write-off all that inventory as scrap. Somebody in AMD made a difficult costly decision in the 11th hour.
I always find this argument amusing. AMD's every bit as guilty as Intel of platform updates and incompatibility, they've just been minimizing mechanical incompatibility of late. Here's a nice comparison of sockets targeted towards consumers/enthusiasts since the two companies diverged at socket 7.
Code:
Intel - AMD
Socket 8 - - 1995
Slot 1 - - 1997
- Super Socket 7 - 1998
Socket 370 - Slot A - 1999
Socket 423 - Socket A - 2000
Socket 478 - - 2000
- Socket 754 - 2003
- Socket 940 - 2003
LGA 775 - Socket 939 - 2004
- Socket AM2 - 2006
- Socket F - 2006
- Socket AM2+ - 2007
LGA 1366 - - 2008
LGA 1156 - Socket AM3 - 2009
LGA 1155 - ? - 2011Now, the favored argument against Intel primarily sprouts from the multiple iterations and non-forward compatibility of LGA 775 in particular. The original pentium 4 motherboards didn't work with some of the later 'high end' pentium 4's due to inadequate VRMs, nor did they work with the initial core 2 duo. And then initial core 2 duo motherboards didn't necessarily work with 45nm processors, so it could be claimed that LGA 775 was 3-4 sockets... at which point you've caught up with the number of sockets AMD released in the same time frame.
As for the counter-argument that there's some degree of compatibility between AM2, AM2+, and AM3... Sure, in the same way that there's a degree of compatibility between all those variations of LGA 775. I can find a number of early AM2 motherboards that are not compatible with AM2+ processors (sure it's likely because the manufacturer was too lazy to update the BIOS, but they always could have implemented something in a fashion that resulted in actual incompatibility.) Same goes for AM2+ boards supporting AM3 processors - there are some that don't, again, primarily for the same sort of reasons why older LGA 775 boards might not work with newer processors. And indications thus far are that AM3+ will be the same thing once more, with only newer AM3 boards being compatible, though as usual, without the new features enabled by the socket update.
Note that I have zero expectations of ivy bridge actually working in my P67 motherboard, but I wouldn't be surprised if P67 motherboards released/revisioned later this year and hence still well before ivy bridge will support it.
Phynaz
Lifer
- Mar 13, 2006
- 10,140
- 819
- 126
I want to make sure I'm clear, I'm not saying the CPU sucks, I'm saying the situation sucks.
Nemesis 1
Lifer
- Dec 30, 2006
- 11,366
- 2
- 0
Would they really have built so many broken ones?
My guess is that they were devoting most of the fab capacity to Llano anyway, and they recognized there was a problem back when that NDA-breaking guy came on here. He may have been lying, but there were plenty of other reports of ES's going out on the internets.
So somebody had to make a tough decision, but probably not that costly of a decision, and probably more like around 9:30.
Phynaz
Lifer
- Mar 13, 2006
- 10,140
- 819
- 126
I feel for the AMD engineers and GloFo engineers right now, I know that crisis-mode mad-dash work environment that they are all surely in the middle of right now. What a crappy way to spend your summer, late nights and 6-7 day work weeks as you rush around the hot lots and running from one the data analysis team meeting to the next
Been there, had large Incident Management project suffer from scope creep. 7x15 leading up to and through the Christmas holidays. Wives were unhappy.
That's the part I don't get. If it's a yield issue shouldn't it have been found a while ago? I would think they would have known about it three months ago and would have tempered expectations.
The only thing I can think of is nobody wanted to slip the project time-line until there was absolutely no way it could be met. But even that point has to be sooner than a couple of weeks before launch.
I was just thinking, I'd like to see the date the slide was created, and then something just popped into my head...AMD said 60 - 90 days until BD. That probably doesn't bode well, in that they aren't willing to give a firm date. Really when I think about it, here they are at the end of a four year program, and they are stating that wide of a range? That tells me they are dealing with a lot of unknowns, and most likely don't know of a solution to whatever issues they may be having.
I wouldn't want to be the program manager going to the executives saying essentially "I don't have visibility 60 days out".
I'm sure they will get it taken care of, but it sure looks pretty scary right now.
Nemesis 1
Lifer
- Dec 30, 2006
- 11,366
- 2
- 0
Well its been talked about alot here . AMD went to a new process@32nm + gate first plus a new arch. If AMD would have gotten it on the B0 or B1 stepping that would have really been something to brag about at AMD & GF. If they get it befor B4 I will still be impressed. The last thing will likely ever here is gate alignment on gate first . As intel said its a problem , that annealing process is hell on metal gates.
Last edited:
Uhm...because they were intending to launch in H1 2011?
Unless that was going to be a paper-launch, they had to have had the fab "fully loaded" with bulldozer in anticipation of stocking up every reseller around the globe for the most anticipated AMD product launch since Phenom.
It takes 90 days to make a 32nm product wafer.
Whatever BD product they thought they were going to be shipping to resellers this month had to have been in the fab for the past 3 months.
That's a lot of inventoried/stockpiled product to purge. Again this is all assuming AMD intended to not do a paper-launch.
If the whole "H1 2011 launch" thing was really planned to be a paper-launch then yeah, 3 months ago they would not have needed to start making hordes of B1 stepping bulldozers.
Ouch! Nothing hurts like smashing up family time right around the holidays
Could be they found it a long time ago, or rather they identified the symptom, but they may be spinning their wheels (and reticles) attempting to nail the solution to the bug without killing the yields.
I just read this BSN article, has some interesting tidbits in it. (assuming Theo is telling the whole truth here)
http://www.brightsideofnews.com/news/2011/5/30/amd-delays-bulldozer-citing-performance-issues.aspx
Last edited:
Phynaz
Lifer
- Mar 13, 2006
- 10,140
- 819
- 126
I think people underestimate the number of chips a fab can produce, and GF has two of them (although they call them "modules")
From what I've been able to surmise, GF Fab 1 Module 2 (The old Fab 38) is dedicated to 32nm production and has a capacity of 25,000 wafer starts per month.
That's greater than AMD's entire capacity in the Athlon heyday when they had 15,000 wafer capacity and were building out to 30,000, if my memory isn't failing me.
Unless yields are terrible, and / or Llano is physically huge, I don't see it sucking up that much capacity.
I'm with IDC on this one, I believe GF is having yield issues. I mean after all, who besides Intel hasn't had issues below 45nm?
Last edited:
This is looking more and more plausible:
http://www.bit-tech.net/news/hardware/2011/04/05/amd-and-global-foundries-good-32nm-chips/1
"In a press statement, AMD said that from now on its 'price for 32nm products will be based on good die,' implying that AMD was previously seeing a few too many duff dies per 32nm wafer for its liking, and therefore only wanted to pay for working chips"
O_O they'll let them pay per good die? That seems kinda off. I bet the price is a lot higher, its not like the fab company has total control over whether or not a CPU comes out okay or not, is it?
It is industry standard practice for foundries whenever their yields are low on a given customer product. There's no specific yield threshold where it comes into play, but generally if yields are below 20% or so then the contracts will tend to be negotiated to be based on a NUB (net units built) basis rather than a wafer count basis (where you pay a flat rate per wafer regardless of the yields).
So there is nothing unusual about the contract as reported, it just doesn't bode well for the health of GloFo's 32nm process is all.
So as I understand it, at TSMC, AMD had no problem with 40nm but Nvidia couldn't figure it out. Who gets the blame for the bad yields there? The GPU designers or TSMC?
If AMD is the one that has to come out with a new stepping here, doesn't that mean that it's their fault and not GloFlo's?
OK, so yield issues, but does that mean that there is something wrong with Bulldozer or does it mean that there aren't enough good die to go around between Llano and BD?
Llano is launching in 2 weeks(probably) and they've already shipped it to OEMs and it's pretty obvious that reviewers have them, so GloFlo can't be doing SO bad, right?
... I feel like I am once again way out of my comfort zone in terms of how much I know about what goes on with these things. What does "respin" mean, what's a "stepping"? I know what they mean from a consumer POV, but technically, what do they actually change in the chip? Also, "metal layer"?
mmmm.... Wikipedia: http://en.wikipedia.org/wiki/Semiconductor_device_fabrication Now I have something to read over the summer
If AMD is the one that has to come out with a new stepping here, doesn't that mean that it's their fault and not GloFlo's?
OK, so yield issues, but does that mean that there is something wrong with Bulldozer or does it mean that there aren't enough good die to go around between Llano and BD?
Llano is launching in 2 weeks(probably) and they've already shipped it to OEMs and it's pretty obvious that reviewers have them, so GloFlo can't be doing SO bad, right?
... I feel like I am once again way out of my comfort zone in terms of how much I know about what goes on with these things. What does "respin" mean, what's a "stepping"? I know what they mean from a consumer POV, but technically, what do they actually change in the chip? Also, "metal layer"?
mmmm.... Wikipedia: http://en.wikipedia.org/wiki/Semiconductor_device_fabrication Now I have something to read over the summer
It's definitely a possibility. One important point to remember being that 32nm is the first node where anyone else is making use of HKMG for production.
Nemesis 1
Lifer
- Dec 30, 2006
- 11,366
- 2
- 0
Ya llano is the straw that could cause AMD considerable problems if yields are good there . Thats on 32nm also . So if AMD only has to pay for good dies. It will show up in the QT reports for both AMD and GF . You can bet alot of people will be crunching those numbers if yields are good on llano . Than GF may be giving under the table ass, to AMD and thats not legal its share holder fraud. GF better lose alot of money this qt. or this bubble will burst.
Vesku
Diamond Member
- Aug 25, 2005
- 3,743
- 28
- 86
With 40nm at TSMC, AMD ran their 4770 batch earlier than nVidia did their 40nm test gpu. They were also more pessimistic about the promised yield improvements. So they made a design choice to go with a medium sized power efficient die. They're hunch was correct and they reaped the benefits.
Seems that with 32nm at Glofo as noted by Nemesis1 AMD had made the decision to keep 32nm as gate first before spinning off Glofo. They thought they could power through the yield issues based on having a lot of engineering experience with gate first. When you are operating at non-Intel levels of fab funding you tend to want to delay full overhauls of process as long as possible. Rumor plus the reworking of the AMD+Glofo agreements, cited by Idontcare, seems to suggest that the engineers have had a tougher time making it work out than AMD originally expected. Combine that with a desire to be as close to Sandybridge performance as possible or beyond just to be able to sell some desktop processors and you have this long wait situation.
Hope this extra wait allows AMD to pull a PH II C3 stepping or Radeon 58xx feather out of the engineering cap.
That all said, judging by the clearing inventory sales on low end notebooks, OEMs are pretty confident about Llano.
Seems that with 32nm at Glofo as noted by Nemesis1 AMD had made the decision to keep 32nm as gate first before spinning off Glofo. They thought they could power through the yield issues based on having a lot of engineering experience with gate first. When you are operating at non-Intel levels of fab funding you tend to want to delay full overhauls of process as long as possible. Rumor plus the reworking of the AMD+Glofo agreements, cited by Idontcare, seems to suggest that the engineers have had a tougher time making it work out than AMD originally expected. Combine that with a desire to be as close to Sandybridge performance as possible or beyond just to be able to sell some desktop processors and you have this long wait situation.
Hope this extra wait allows AMD to pull a PH II C3 stepping or Radeon 58xx feather out of the engineering cap.
That all said, judging by the clearing inventory sales on low end notebooks, OEMs are pretty confident about Llano.
piesquared
Golden Member
- Oct 16, 2006
- 1,651
- 473
- 136
It's not that hard to understand.
-Brazos blew away AMD's expectations on demand and outrstipped supply.
-AMD has said about a billion times that the Fusion ecosystem is where the company is headed.
-Need to prioritize production.
-With the success of Brazos, only a fool wouldn't anticipate huge demand for Llano.
-They expect a 10% Brazos, 20% Bulldozer, 70% Llano product mix.
-Llano launches in 2 weeks.
-They currently have a respectable presence in desktop.
-The highest margin products are in server.
-Server Bulldozer schedule remains Q3.
-Priority to grow mobile and server market share.
-Zambezi is based on the same Orochi die as in Valencia.
-Llano much more appealing to OEM's.
-Focus on OpenCL.
-Common sense.
-Brazos blew away AMD's expectations on demand and outrstipped supply.
-AMD has said about a billion times that the Fusion ecosystem is where the company is headed.
-Need to prioritize production.
-With the success of Brazos, only a fool wouldn't anticipate huge demand for Llano.
-They expect a 10% Brazos, 20% Bulldozer, 70% Llano product mix.
-Llano launches in 2 weeks.
-They currently have a respectable presence in desktop.
-The highest margin products are in server.
-Server Bulldozer schedule remains Q3.
-Priority to grow mobile and server market share.
-Zambezi is based on the same Orochi die as in Valencia.
-Llano much more appealing to OEM's.
-Focus on OpenCL.
-Common sense.
Well that sure is one theory.....
Any particular reason to completely ignore the performance and yield rumours?
ItsAlive
Golden Member
- Oct 7, 2005
- 1,147
- 9
- 81
This article seems to be a reasonable clarification, but I'm unsure if it is reputable or not.
http://www.planet3dnow.de/cgi-bin/newspub/viewnews.cgi?id=1306941618
http://www.planet3dnow.de/cgi-bin/newspub/viewnews.cgi?id=1306941618
piesquared
Golden Member
- Oct 16, 2006
- 1,651
- 473
- 136
Well that sure is one theory.....
Any particular reason to completely ignore the performance and yield rumours?
Sure, how about because they're rumours.
And, it makes zero sense to hamstring the future of the company to satisfy a few demanding enthusiasts. Curiously, the most vocal of those seem to be new users popping up on various forums around the net having fits and that they're going to head down to the store and buy a sandy bridge CPU. Well so what, go! Go fast! If you need the fastest gaming chip money can buy, why not do just that. It's not going to make AMD show their hand I can guarantee you that.
Anyway, as their tagline is "the future is Fusion", being a shareholder i'd be livid if they chose to sacrifice the future of the company and most lucrative markets to appease the demands of some self entitled enthusiasts, and a bunch of strategically placed FUD.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter