AMD's Fab 36 Grand Opening - 90nm and 300mm in Germany
by Anand Lal Shimpi on October 14, 2005 7:50 AM EST- Posted in
- CPUs
After 24 months of construction, today AMD's first 300mm semiconductor fab, Fab 36, celebrates its grand opening. Built adjacent to AMD's 200mm Fab 30 in Dresden, Germany, Fab 36's grand opening takes place 6 years after Fab 30's introduction. The name Fab 36 comes from its existence 36 years after the founding of AMD.
The announcement for Fab 36's site in Dresden came in November 2003, with first ground breaking at the end of that month. Approximately 12 months later, the building was ready for equipment, and another 12 months after that, Fab 36 began preparations for mass production.
Over 150 members of the press enter AMD's press conference at Fab 36 in Dresden
Unfortunately, at its grand opening, Fab 36 is still a 90nm-only fab; throughout the next year, AMD will begin the transition to 65nm production. The first CPUs built at Fab 36 will be shipping in the first quarter of 2006, with the first 65nm chips leaving Fab 36 by the end of 2006.
Sometime in 2007 AMD will have performed a "substantial amount" of the transition of Fab 36 to a 65nm semiconductor fab, bringing the grand total for the cost of Fab 36 to an astounding 2.5 billion US dollars. There is no word when Fab 36 will be completely converted to 65nm manufacturing.
By 2008, Fab 36 will be able to produce more than 2x the number of processors as Fab 30 (potentially 100M processors per year based on current die sizes).
The first 300mm wafer produced at Fab 36 actually took place back in March, but preparation for mass production continues up to and beyond today. Today's grand opening of Fab 36 marks a huge step in AMD history, as it will hopefully alleviate a number of the supply issues they have been plagued with in recent history.
The two individuals in the middle kicked off the day at Fab 36, AMD's CEO, Dr. Hector Ruiz is the second on the left, and VP/GM of AMD Dresden, Dr. Hans Deppe is to the right of Dr. Ruiz.
The day started off with a press conference kicking off the grand opening of the fab. Below we've got a picture of a clean room engineer (or someone dressed as one) posing with AMD's Dr. Hans Deppe, the Corporate VP and General Manager of AMD in Dresden.
While we don't have a 200mm wafer here to compare sizes with, imagine the wafer pictured above, but smaller.
After some brief words about the opening of Fab 36, a brief Q&A period started. One of the first questions was what will happen to Fab 30 in the future now that Fab 36 is ready to start mass production. Dr. Ruiz mentioned that Fab 30 could be used for more x86 microprocessor production, it could potentially become a chipset manufacturing fab (possibly indicating AMD's intentions to eventually return to chipset manufacturing), or it could be upgraded to future semiconductor technologies for use in the future (e.g. 65nm).
One other important question that was asked was "why Dresden" for Fab 36, to which the answer was obviously multi-faceted:
1) First and foremost, AMD has invested a lot in the people of Dresden, when it comes to training and expertise in semiconductor manufacturing. It was simply easier to leverage the existing human investment.
2) Obviously government subsidies played a large role in AMD's decision to bring Fab 36 to Dresden. Dr. Hans Deppe listed the total amount of government subsidies for Fab 36 as 500 million US dollars.
The final question of the press conference was about AMD's future fab plans, which Dr. Ruiz answered with a timeframe of 2008 for the start of production on the next major fab plant.
Facebook
Twitter
RSS
70 Comments
View All Comments
JarredWalton - Friday, October 14, 2005 - link
"Ummm...If no one really knows, how do you know that they are going to "put up one hell of a fight"? :)"It's the general attitude of Intel right now. They are not trying to tell everyone that NetBurst is great, and have freely admitted on occasion to the problems with the design. To me, Prescott looked like a bad design from the beginning. More pipelines (when Northwood already was too long) combined with higher cache latencies? I never did like the design much.
Conroe appears to be taking everything good from both NetBurst and Pentium M/Yonah and enhancing it. I expect Yonah to basically equal Athlon X2 Toledo clock for clock, and in several instances beat it. I also expect Conroe to end up about 25% faster than Yonah clock for clock. If both those guesses are correct, Conroe at 2.0 GHz would basically match the X2 4800+ (and I admit I might be wrong).
Once Yonah is released, though, we'll have an even better estimate of Conroe performance. Conroe goes to 4-wide issue, which should at least add 10 to 20% more performance if done properly. Double the L2 cache from 2MB shared to 4MB shared, and you might see another 10% performance boost.
So what does AMD have planned for the near future? The only thing I see coming is DDR2 support. Hopefully, there's more to the socket S1/M2/F chips than that, but if so AMD isn't talking. The higher bandwidth of DDR2 might add 10% to the performance of the Athlon chips, but unless I hear something more I expect Conroe to outperform M2 chips clock for clock. If AMD can command a clock speed advantage, we might end up with a tie.
I'm not being anti-AMD or pro-Intel; this is simply the way I see things right now. I really hope AMD can prove my guesses wrong (from their side). If Intel can match or exceed my guesses, more power to them.
Viditor - Friday, October 14, 2005 - link
Fair enough...but didn't we only learn about that sometime near the launch? For the years prior to launch, Intel was saying that it was going to scale beyond 5GHz...(I believe they even mentioned a 10GHz possibility)
Clock for clock (IMHO) isn't really a valuable comparison. If you compared (for example) the Athlon to the P4 clock for clock, you'd wonder why Intel was even in the CPU business...
I realize that the Yonah architecture is a much closer comparison, but we are going to have to see exactly what kind of "clocks" are available when it's released...and don't forget that Yonah will be operating without 64bit (which helps it tremendously in performance but may hurt it quite a bit in sales).
As to Intel's Nextgen architecture, I agree that it sounds good on paper, but there are any number of unanswered questions about it that we won't know for quite awhile! For instance, how well will it be able to handle cache coherency on a shared cache? How efficient will the added issues be? etc...
Admittedly, we know even less about AMDs future...here are some of the things Anand mentioned
1. Hypertransport 3 then 4
2. On-chip co-processors (North and South bridges as well as PCIe?)
3. FBDimms...this will be huge in the Opterons!
But I agree it would be nice to have more info...it's quite possible that AMD are keeping quiet because they want to hand Intel some surprises, but that means that any changes they are planning won't require external development (or they'd have to announce sooner rather than later).
I know you're not...and I appreciate the debate! If you want, we can change sides and I'll take Intel and you can take AMD next time. :)
Cheers!
JarredWalton - Saturday, October 15, 2005 - link
HT 3 and 4 don't excite me much. For SMP setups, that's useful. I'm fine with dual-core, though (I don't even tax that), and since HT is really just used for the CPU to NB connection, it rarely gets anywhere near maxed out. It won't be bad to have more, but a faster HT speed isn't going to be the panacea that a faster cache or higher clock speed would be.I really hope AMD is just keeping quiet, but if so their partners are also in the dark. That leads me to believe that it's more likely that there just isn't that much in the way of major changes coming. FDB is something that's useful for a certain market, but the home user isn't that market. Sort of like selling registered memory systems to the home user... hmmm, where have I seen that before? ;)
Yonah's lack of 64-bit might prove to be a problem long term, which is why Conroe (P8? Yeah, let's just call it P8 for now!) will be important. However, I've tried XP-64 with an Athlon 64, and frankly I think it's crap. Worse driver support, worse installation, worse software compatibility, and to top it all off performance was no better in any of the applications I tried.
Until we get fully 64-bit OS, drivers, and applications, 64-bit support does nothing. Vista will be best case scenario, I think, and really we need to have systems shipping with at least 4GB of RAM before I'm concerned. 2GB is still very high-end for the desktop, and only one typical application (Battlefield 2) shows real benefits. I suppose Photoshop users that like to open a bunch of large images at the same time will also benefit from 2GB of RAM. (I'll have to try that one of these days....)
Lack of 64-bit capability will hurt it from the marketing perspective, but in real world usage I don't think it will matter much. That goes double for the mobile sector, where 1GB (2x512MB) configurations are high-end, and 2x1GB builds are the maximum amount of RAM you can have anyway. 64-bits with 2GB of RAM is such a niche market that I don't bother worrying about it. It might get you 10% performance boost at best, and only compared to generic code that doesn't use MMX/SSE for 64-bit int calculations. Bleh.
Okay, enough babble. We'll sort out the real victor in mid-2006. AMD (IMO) really needs more than just DDR2 and a few tweaks (in 2006) to convince me they'll maintain a performance lead. K9 (*woof!*), where are you? WHAT are you? Please don't say you're just quad-core, because I'm barely sold on dual-core for many people, and quad-core will be of true use to a fraction of the market. (I've done software development, and I just don't see multi-threaded code that really benefits from SMP coming soon enough to make multi-core the best path forward - outside of certain workstation/server tasks, of course). At least a theoretical 33% increase in dispatch rates (P8) has a clear benefit to most applications.
jiulemoigt - Saturday, October 15, 2005 - link
well I read that and I can tell how long you have been using PCs
because the reason so many people jumped on the AMD 64 before there
was a cunsumer{non workstation} 64bit OS was because the mem controler
is on die, which reduces latecny and and allows the cpu to go directly
to the ram and get what it needs with out calling to the north bridge
and then to the ram. when they first came out with NF3 the buss was
slower and it showed when via came out with a 1gb bus to nvidia 800mb
bus if they can make that bus faster then they can data too and from
the cpu faster, which is very significant, even on a home users machine.
Of course a faster or large cache on the cpu will be faster by magnitudes
than any off die mem unless it is the first time the cpu needs the data
like in rendering where the data is almost always new, and the cache can't
help unless it is prefetching the data, which is why half of most modern
cpu are prefetch logic {though they all have different names to do the
same thing}. Which most of the above beyound most home users to even
understand which develpers will continue to have to fight for decent
systems because the bosses home machine does not use 10% of his systems
resources.
As to AMD leaving their partners in the dark most changes can be down on die
without effecting the pin out it only makes sense when you sell both the north bridge and the cpu to keep changing the pin out to force users to upgrade their mobos.
this quote is what made me log in and post, most computer profits are made on volume not home users and those who welcome vista deserve what they get, with vista your data is no longer yours, I'm not even refering to files you get from other people but stuff you create on your own software. On top of that I have a machine running a full 64 bit OS and it is fast, only it does not play my games... for games and only games do I run a windows box despite being in charge of windows 2000 pro, windows 2000 server, windows 2003 server, the sql sever app and all the idiots who perfer to use outlook still at work. My work will pay for me to have copy of any the windows for home use and I'm still runing 2000 over xp because it is faster on the same hardware, windows xp codepath is done windows 64 is based on server 2003 which is built on 2000 server with a few more service added and some annoing wizzards for domain control for those who don't understand what an A record is. Also any 32bit windows shipping with more that 3gb of ramm is going to have trouble four gb is the limit but windows apps in a 32 bit enviroment can not touch that last gb. Though I'm considering using server 2003 64 bit for my gaming machine since it does not have some of the crap{outlook express, burning toy and a few others} built in and handles threads better {not refering to hyper threads}. If i'm harsh it is because I get sick of people posting out of their ass.
for anyone thinking I"m anti-MS I'm part of MSDN, and upset that an OS which was going in a direction of usiblity has made changes to a business OS that hurts the bisness aspects of it, the problem is their business OS was outperforming their home OS so they put the guys in charge of the Home OS in charge of both and let them break the solid bussiness one.
PrinceGaz - Sunday, October 16, 2005 - link
Yes, I liked Windows 2000 Pro, it was lean and mean and got the job done with the minimum of fuss or overheads.But I have to admit XP has its bonuses like last night when "Webhancer" was sneakily installed onto my box with some "free" software, and removing it as recommended (by Add/Remove Programs) resulted in a totally dead net connection. That's spyware which interferes with critical Windows files in action, it can't uninstall itself properly and leaves you screwed. Fortunately booting up XP in Safe Mode and using a System Restore to earlier that evening got everything back to normal, I couldn't have done that with Win2K, not without third-party software which I'd have already need to have installed anyway.
I've often been tempted to try Windows 2003 Server, but I've always been concerned that a "server" version would not be suitable for home use especially when gaming, much like Windows 2000 Server. If the home and server products are almost merged now like you suggest, maybe it is better to treat Windows 2003 as the last safe version of Windows before MS take away control of your computer with Vista. It certainly seems on the MS site that Windows 2003 and XP are treated very similarly, so maybe the "server" tag with 2003 is just to clarify that it supports multi-processor computers.
JarredWalton - Saturday, October 15, 2005 - link
The NVIDIA nF3 150 was a 600 MHz HT connection, running 8-bits in one direction. Of course that was slower than 800x16-bits, as 600x8-bits would be less than half as fast. I've done plenty of testing with current AMD systems, and the difference between 1000x16 and 600x16 is not that great.Overall, though, maybe you ought to do a bit more research. http://www.anandtech.com/mb/showdoc.aspx?i=1883&am...">nForce3 150 wasn't that slow, which only serves to support what I said. Yes, a few benchmarks can put the slower HT performance of the chipset in a bad light, but if they had even had 600x16 for the upstream connection, that would have basically fixed the bottleneck.
Regarding 64-bit, you completely missed the point of what I said. I repeat, it doesn't matter much unless you're running over 2GB of RAM. I haven't seen such a desktop system, though I'm sure a few people have 4x1GB just because they can. 64-bit OSes in the home and business will really only be useful when we're using more RAM than at present.
Viditor - Friday, October 14, 2005 - link
I thought of a good analogy to explain my point...1. Firstly remember that no matter how many Fabs you have, there are still only a certain number of chips sold each year.
2. I believe that the current estimate for total x86 chips used globally is ~240 million/year. AMD has the capacity to produce ~50-60 million of those with just Fab30 on 200mm at 90nm. If you add Fab36 (even at 90nm) on 300mm, that's a capacity of much more than 50% of the total x86 cpus used in the world (and closer to 75% at 65nm)! What good would it do to have more Fabs at this point?
On to my anology...if you drive to work and back every day in a traffic jam, and you have a choice between a V12 car or a V6, which is the better car to drive (considering gas prices, etc...)?
JarredWalton - Friday, October 14, 2005 - link
Which is better to drive? Pentium M, I guess. But not really for gamers - who are the equivalent of 4x4 offroad types or race car drivers, depending on perspective.The fab capacity has other advantages besides CPUs produced. What costs more to make, Smithfield or Toledo? Both are similar in size, I believe, and while Toledo has a clear performance advantage, Intel can produce twice as many chips per wafer, allowing things like a $250 820 instead of a $380 3800+. (Yes, I know the 3800+ actually comes out ahead of the 840 in most tests. Cost is the bigger concern for businesses, though.)
-----------
Here's another take on the Intel vs. AMD "war":
P3 vs. Athlon = tie, more or less. Athlon is a good design.
P4 vs. Athlon XP = P4 wins in performance, but costs more. (Don't try to sell me on a 3200+ actually beating or even matching a 3.2C, because it didn't.) Athlon XP is a tweaked Athlon core, while P4 is totally new.
P4 Prescott vs. Athlon 64 = A64 wins. Prescott is once again a pretty major revamp of NetBurst. Calling it the same architecture as Northwood is like calling the Athlon 64 the same as the original Athlon. There are many similarities, but the change to the pipelines and other internals is quite drastic. Unfortunately, many were marketing driven changes as opposed to engineering ("clock speed sells!").
PM vs. Athlon 64 = tie. Athlon 64 is often faster, but uses more power. Dothan on a desktop platform puts up some very good numbers in many tests, and only raw FP/SSE performance hold it back. PM is based off the earlier P6 core, but with many major changes. It can easily be considered a new architecture.
Yonah vs. Athlon 64/X2 = ?. I'm going to guess Yonah comes out ahead overall, but not by a huge margin. With A64 running at up to 2.8 GHz, it will likely lead performance overall, but not clock-for-clock.
Conroe vs. M2 = ?. As stated earlier, I'll bet on Conroe. This is a pretty major overhaul of the PM architecture. M2 is a tweaked K8, which was a tweaked K7.
So, keeping score, Intel has had P6, NetBurst, PM, and an upcoming Conroe architecture. AMD has had K7 and K8. Dual-core, Intel has NetBurst and PM, as well as Conroe. AMD has K8. Intel has done a lot of running around, but much of it to no avail. Or rather, half of their work was in areas that really weren't that wise - and the engineers actually knew this! Now they're forgetting about the marketing war and simply focusing on true performance.
Even with their mistakes, Intel has still driven a lot of technological advancements, and I for one expect the next architecture to shed some shackles and allow the Intel engineers to really strut their stuff. I've known some Intel guys that are really sharp, and there's little reason other than management/marketing for many of the past errors.
Viditor - Saturday, October 15, 2005 - link
Actually, games is the only area where Pentium M is competitive...(please note that I said "competitive", not superior) everything else they are significantly weaker at.
This isn't a question of Fab capacity, it's design.
Yes, it's more efficient to slap 2 chips together than to manufacture them on a single die...but you pay for it in performance (as you've noted)
Yes, 300mm can produce twice as many chips/wafer as 200mm (die size being equal)...but a 300mm wafer costs 50% more than a 200mm wafer. And being able to tweak yield efficiency on your line at a milliseconds notice will give you an overall advantage so that Smithfield may actually cost the same or more to manufacture (remember that even though they sell them cheaper, doesn't mean they cost less to manufacture...).
Then we shall have to agree to disagree...later Northwoods were a definite Intel win, earlier Northwoods were an Athlon win...
Hmmm...I think you need to do some reading on this.
1. Turion and PM use the same power (don't confuse power use and TDP, they aren't the same thing)
2. A64 and Turion win every bench except games (most by a substantial margin) against the PM. And the PM is unable to run 64bit (meaning that it's not exactly an apples to apples comparison)
I'll bet a lot that you're very wrong about that one...! The comparison will be the Yonah (32bit only) vs the dual core Turion64...due out near the same time.
JarredWalton - Saturday, October 15, 2005 - link
I'll give you Athlon XP over Willamette, and I'll give you XP in price/performance over P4. In terms of raw performance, though, once Intel shifted to Northwood I never really felt AMD had an overall performance lead. There were certain tasks where they did well - they maintained a lead in compiling performance basically since the launch of the Athlon XP - but looking at a wide variety of uses (gaming, video, audio, office, etc.) I have to give Northwood the win. Don't get me wrong: I went from a P3/Celeron 1.1A to a Barton core and never actually owned a Willamette or Northwood of my own. But if money wasn't a concern, I would have stuck with Intel chips up until the Athlon 64 launch.Regarding Turion vs. Dothan (which is the fairest match), I think Dothan still comes out ahead in power. I haven't been able to get a Turion chip, though. :( I do know that a fully loaded 2.0 GHz Dothan desktop only draws about 55W idle and 120W load (and most of that is for the GPU on load). A 90nm SOI Sempron is 110W idle and 190W load in a similar configuration. Even taking into account chipsets and performance differences, there's a pretty substantial power draw advantage for the Dothan. It a 2.0 GHz Turion really draws significantly less power than a 1.8 GHz Sempron 64 Palermo (what I tested)... maybe it's a tie. I really doubt that the Sempron is using more than 35W-40W on its own, though, despite the 65W rating.
That's all still ignoring the fact that Dothan is really a better design for mobile computing, though. Turion is to Athlon 64 what P3-M was to P3. In other words, it was a minor tweak, some binning for lower voltage, and presto! You have a mobile computing processor. Any way you look at it, though, Dothan offers some amazing performance per Watt; end of story. Price, on the other hand.... :(