Wednesday, June 14, 2006

DELL: AMD64 the fastest growing notebook market

Joseph Marengi, senior vice president of Dell Americas, said that "[t]he fastest-growing segment of the notebook market, which was the fastest growing segment of the PC market, was for machines with AMD processors."

Since its introduction last year, the Turion 64 controls 100% of the 64 bit notebook market. The recent introduction of Turion 64 X2 brings dual core 64 bit mobile computing to the masses. HP has a set of consumer and business mobile solutions based on Turion 64 X2.

I expect DELL to go AMD full force at the time of Conroe launch.

In other news, it was reported that AMD is moving away from dual core Rev F desktop chips with 2x1MB cache and will focus on the ones with 2x512KB. This is indeed a great move, as some readers have pointed out. AMD64 architecture is so efficient, it doesn't need big caches. In fact, even the next gen K8L has only 512KB L2 per core. With inefficient archirectures such as Intel's aging Pentium III based design, up to 4MB L2 is needed for achieving reasonable performance.

The 2x512KB Rev F still has a die size of 183mm^2, which is quite large. Using a geometry of 13.5 x 14, I found that a 300mm wafer can produce 336 dies, a 200mm wafer produces 144 dies. Each 1000wspm @ 300mm corresponds to 1 million dual core dies per quarter. FAB30 is 200mm at 30000wspm, it can produce 13 million dual core dies. FAB36 should be at 7000wspm by mid 3Q06. So AMD should be able to pump out 20 million dual cores per quarter by then. By 1Q07, FAB36 will be 50% converted to 65nm. At 65nm, the die size will be 95mm^2, which means a 300mm wafer can produce 690 dies. One 1000wspm at 300mm should produce 2 million dies per quarter. Thus, by 1Q07, AMD's capacity should more than double to produce close to 25 million dual cores per quarter. AMD will have capacity to supply both DELL and HP 100%. By July 2007, K8L time, FAB36 will be 100% 65nm, that FAB alone can crank out 30 million dual core dies per quarter. By 4Q08, AMD will be doing 45000wspm 300mm wafers.

BTW, I forgot Chartered FAB7 is delivering 1000wspm of 300mm wafers in July 06, 1 mil there. Based on demand, Chartered can crank a lot more chips for AMD.

Overall, AMD is poised to increase capacity to 100% of world's PC demand. I expect AMD to quickly cross 50% market share. Equally quick will be Intel's financial collapse, due to its higher cost and overhead.

78 Comments:

Anonymous Anonymous said...

Does anyone have a list of desktop/mobile applications that are currently designed for 64-bit? I have a 64-bit processor now but it seems taht all my software is still 32-bit (including the OS). Do 64-bit versions of things like Photoshop or MS Excel exist yet? Just wanting to know what to do with these extra 32-bits? Thanks!

9:59 PM, June 14, 2006  
Anonymous Anonymous said...

Turion X2 controls 100% of 64bit mobile plateform market.

where is that market? i don't really see any average customer (and most of enthusiasts) taking the advantage of 64bit.

Dell will go full force with AMD64 when Conroe launch.

that is a wishful thinking. not only Centrino Duo outperforms Turion, Merom outperforms Centrino Duo by at least 20%. that means Turion is at least 20% slower at the time when Conroe launches.

and FYI, Vista can run on both 64bit and 32bit. average consumers don't care which edition their vista runs at.

time to wake up! AMD fanboy!

9:59 PM, June 14, 2006  
Anonymous Anonymous said...

From 1MB down to 512KB? Dang! Didn't see that coming... I guess AMD are desperate enough on the supply (and demand) side of things. Things aren't looking bright for AMD at the moment.. I guess that rules out adding 2MB/4MB/8MB cache and ZRAM for AMD. The future is in 65nm (and upcoming 45nm), not 90nm anymore. AMD is trying to double production for a single fab using this method.. I wonder what would buyers say? What will benchies be like with half the cache?

BTW, do you know what the word "yields" are? Every wafer will not produce 100% good chips, usually only less than half are working silicon.. The rest goes to the thrash bin. Then the working pieces have to be spec graded (and speed binned) as well. Cherry picked ones will be Optys and FXs/X2s.. the rest will be single core A64/Sempys (those with only one proper working core). Yields, yields...

Now do you know why the more FABs, the bigger quantity can be produced? Intel has that..

10:21 PM, June 14, 2006  
Blogger Sharikou, Ph. D. said...

BTW, do you know what the word "yields" are? Every wafer will not produce 100% good chips
AMD's yield is over 90%. That's how it could take 22% market with one 200mm FAB.

10:22 PM, June 14, 2006  
Anonymous Anonymous said...

"AMD's yield is over 90%. That's how it could take 22% market with one 200mm FAB."

I think that fanboi is trying to compare Intel's yield with AMD's. Are you trying to bash AMD, the second poster?


APM is amazing. It can use 65nm data from East Fishkill(IBM) and use it at Fab 36.

"One of the big advantages of our Automated Precision Manufacturing system is that we can take 90-nm on 200-mm data from Fab30, 65-nm on 300-mm data from East Fishkill and see the commonalities and apply them to Fab36,” the AMD official said.

http://www.xbitlabs.com/news/cpu/display/20050408081329.html

10:56 PM, June 14, 2006  
Anonymous Anonymous said...

I'd like to see Apple 64-bit dual-core machines too. Big mistake for Apple's part actually, as the PowerPC G5 was already 64-bit and uses HyperTransport.

11:31 PM, June 14, 2006  
Anonymous Anonymous said...

Sharikou...

I have done some figures and am questioning the 20% performance benefit AMD will have with 64Bit over Conroe, that you stated a few days ago.

There is a review here...
http://www.pcstats.com/articleview.cfm?articleID=1959

They show benchmarks for 64Bit, and these are up againt old Netburst chips. I am not sure, but I am barely seeing 15% over the Netburst architecture.

Please look at this and explain.

I understand I may be missing something (setup/test themselves).

I am also going to post a link to the PDF I made to show the performance increases for the chips.

http://www.enumae.com/64-32Bit.pdf

Thanks.

11:40 PM, June 14, 2006  
Blogger Sharikou, Ph. D. said...

I have done some figures and am questioning the 20% performance benefit AMD will have with 64Bit over Conroe, that you stated a few days ago.

What I said was AMD64 runs about 10-40% faster under 64 bit mode. The comparison is made between the same AMD64 CPU under two modes.

11:47 PM, June 14, 2006  
Anonymous Anonymous said...

"AMD's yield is over 90%. That's how it could take 22% market with one 200mm FAB."
that's with 90nm part, but both AMD and Intel and progressing into 65nm. AMD has yet to demonstrate a 65nm working AM2 processor. that speaks a lot about their yield @ 65nm.

12:34 AM, June 15, 2006  
Anonymous Anonymous said...

"What I said was AMD64 runs about 10-40% faster under 64 bit mode. The comparison is made between the same AMD64 CPU under two modes. "
and can you please explain which real world applications will benefit from this 10~40% increase in performance by switching to 64bit?

12:39 AM, June 15, 2006  
Anonymous Anonymous said...

Hello
I have seen the test of Athlon64 3000 in one of the biggest polish music magazine.
Following software was used : Windows XP 32 and 64 bit and Cakewalk Sonar 4 (which offers support to 32 and 64 bit mode). Conclusion: Athlon 3000 works in 64 bit like processor PR4000 in 32 bit mode. So is is faster by 30%. From what I remember : used a lot of virtual instruments and few audio track.
Greetings

1:31 AM, June 15, 2006  
Anonymous Anonymous said...

Semprons have 512KB cache. Does this make the new Athlon X2s into Sempron X2s?

1:55 AM, June 15, 2006  
Anonymous Anonymous said...

According to http://www.businessweek.com/investor/content/jun2006/pi20060614_844080.htm?chan=topStories_ssi_5 quotes "Goldman Sachs upgraded Intel (INTC) to outperform from in-line. It also lifted its recommendation for Advanced Micro Devices (AMD), Applied Materials (AMAT), and SanDisk (SNDK) to in-line." and "For AMD, he says losses are likely, but would cover shorts as the stock is at his $25 fair value.". REAL analysts are seeing otherwise... Looks like AMD is gonna have a rough time soon.

3:25 AM, June 15, 2006  
Anonymous Anonymous said...

64bit computing doesn't make sense for desktop computing. even for most of the servers it doesn't make sense right now!

i predict a slow transition to 64bit over the next 5 years.

4:08 AM, June 15, 2006  
Anonymous Anonymous said...

Dont't you people get it? AMD is going the right way, just look at those 64Bit benchmarks over at pcstats.com.

It's time to move forward with the technology, and not stand still, now imagine them Intel chips being 64Bit, then developers would make 64Bit apps and we could all benefit from the increase in computation power.

So it's about time Intel did this and stop holding the progress bar from moving forward.

4:12 AM, June 15, 2006  
Anonymous Anonymous said...

and can you please explain which real world applications will benefit from this 10~40% increase in performance by switching to 64bit?

under linux you get 10-50% speed increase between ia32 and x86_64 (amd64).

5:51 AM, June 15, 2006  
Anonymous Anonymous said...

So, it looks like AMD is getting ready to fight Intel in the big "Clear out the old P4's for free" strategy, by producing more/cheaper chips

My bet is that the lower end Conroe will be impossible to buy from Newegg etc. until New Years or Spring. The top tier vendors will get ALL the meger allocation (and special clients). That will leave the Intel fanbois with AMD or a $1000 conroe choice.(unless they want a slow P4 to heat the house with)

Question, Is this a ploy by Intel and Dell to win the antitrust suit? Here is the Scenario. Dell pretends to be pissed at Intel, negotiates special rates from AMD. When the case goes to court, they say "look judge, AMD uses the same vendor prefrential pricing". Just a thought.

6:14 AM, June 15, 2006  
Anonymous Anonymous said...

"It is no small feat for Intel to catch up with K8 in 32 bit performance. Let's wait and see some more tests and some 64 bit tests. 64 bit is quite relevant, don't you think? You don't want to buy a CPU that will run Windows Vista 20 slower."

1:41 PM, June 12, 2006

This is the post I was talking about.

Again am I missing something?

It seems as though AMD doesn't really have a 20% advantage in 64Bit.

Now Conroe, being Intel next generation, wouldn't it be fair to assume it would perform better than Netburst in 64Bit enviornment?

Thanks

8:00 AM, June 15, 2006  
Anonymous Anonymous said...

"AMD has yet to demonstrate a 65nm working AM2 processor. that speaks a lot about their yield @ 65nm."

I'm sorry! Did YOU say something???

http://s2.supload.com/image.php?get=IMGP1394.JPG

http://s2.supload.com/image.php?get=IMGP1395.JPG

Well I'll be damned, it's a 300mm 65nm production wafer AND a 65nm AM2 processor up and running! IMAGINE THAT! ;^) (thanks to Chris Morley @ HardOCP for the pics!) Smell the Intel fear buddy, they know it's almost here too,(Rev. G and come next year, K8L) why do you think Intel is moving so fast to get to 45nm and to get Conroe/Merom/Woodcrest out the door? Because they can't compete with AMD at 90nm with thier own 90nm process, that's why they had to move to 65nm and now rushing to 45nm! Thanks to the use of APM software AMD will be able to achieve near 100% yield right at start up at the FAB, they've already said 65nm chips will be here IN DECEMBER and Intel knows it will be right back in the same predicament. Any performance lead Conroe has will quickly be dissappearing when AMD moves to 65nm and eventually to K8L, even as it finally hits the shelves.

8:05 AM, June 15, 2006  
Anonymous Anonymous said...

OK. I'll ask it again since this IS the Journal of Pervasive 64-bit COmputing...

Does anyone have a list of desktop/mobile applications that are currently designed for 64-bit? I have a 64-bit processor now but it seems taht all my software is still 32-bit (including the OS). Do 64-bit versions of things like Photoshop or MS Excel exist yet? Just wanting to know what to do with these extra 32-bits? Thanks!

8:11 AM, June 15, 2006  
Anonymous Anonymous said...

"can you please explain which real world applications will benefit from this 10~40% increase in performance by switching to 64bit?"

Cryptography (the thing you use everytime when on-line shopping and banking, VoIP, and even - in near future - media broadcast and multicasting).

This is just ONE example...

8:22 AM, June 15, 2006  
Anonymous Anonymous said...

"i predict a slow transition to 64bit over the next 5 years."

Nop.. 5 months is more likely... or as soon as Vista comes out.

In the Unix world, ALL SunFire and IBM Power servers (oh and of course HP's Itanium) are already 64-bit, running 64-bit capable operating systems.

8:25 AM, June 15, 2006  
Anonymous Anonymous said...

Does anyone have a list of desktop/mobile applications that are currently designed for 64-bit?

there are lots of posibilities: first you have to buy a 64bit operating system like windows xp 64bit edition or windows server 2003 64bit edition (please make sure, that there are 64 bit drivers for your hardware before you buy!).

then you can install lots of desktop applications, like the enterprise databases from oracle, microsoft or ibm!

your 32bit legacy applications like office or photoshop can be run with only slight performance loss. isn't this cool? as soon as there exist any new 64bit office applications in 2 or 3years, you can replace this 32bit legacy crap immediatly! :)

8:37 AM, June 15, 2006  
Anonymous Anonymous said...

Native 64-bit Software
The move to native 64-bit software has, obviously when you think about it, occurred first where it's needed most. Digital content creation, CAD/CAM, software development and visualisation tools are just a few of the markets that have embraced 64-bit with open arms. Games development is now starting to take place in fully 64-bit environments, meaning the developer is spitting out 32-bit and 64-bit versions of a game at the end of the development process, with games prime targets for the 64-bit makeover and associated performance improvements.

Photoshop CS2 takes good advantage of a 64-bit Windows OS with support for taking up to 3.5GB of physical memory where available, popular applications like Ahead's Nero Burning ROM and NOD32 anti-virus software are 64-bit native, Sun offer a native 64-bit version of their Java 2 Runtime Environment (J2RE) and you can get 64-bit versions of Half Life 2 (from Steam), Far Cry and Chronicles of Riddick.

Mozilla offer 64-bit native versions of Firefox that work with all the extensions that the 32-bit copies do, you get a native 64-bit version of Internet Explorer with Windows XP 64-bit, O&O's defrag tool is available in a native 64-bit version, CloneDVD is 64-bit and Tiny's firewall is available as a native 64-bit download, too.

In short, there's a decent amount of native software available with more coming on-stream with every passing day.

8:37 AM, June 15, 2006  
Anonymous Anonymous said...

"Semprons have 512KB cache. Does this make the new Athlon X2s into Sempron X2s?"

Modern Semprons have either 256KB or 128KB total L2 cache.
The only Semprons that have 512KB L2 cache are the socket 462 ones based on the Barton core from the old Athlon XP.
In addition, the X2 cores have 1MB total L2 total because each core has its own 512KB L2 cache.

8:41 AM, June 15, 2006  
Blogger Sharikou, Ph. D. said...

There are plenty of 64 bit apps out there. Just google them out.

I am posting from Windows XP x64 right now. Actually, I don't know whether I am using 64 bit firefox or 64 bit IE. The OS itself is already much faster. It can run both 32 bit and 64 bit apps the same time. So this is transparently to me. But some dude with a Core Duo won't be so lucky, a few months from now, he will find he has to make sure he gets a 32 bit app.

9:15 AM, June 15, 2006  
Anonymous Anonymous said...

"Do 64-bit versions of things like Photoshop or MS Excel exist yet? "

If they don't yet, I'll bet a pretty penny that they are working on it and I'd bet a million of them that they will support it.

9:20 AM, June 15, 2006  
Anonymous Anonymous said...

"Digital content creation, CAD/CAM, software development and visualisation tools are just a few of the markets that have embraced 64-bit with open arms."

Sorry man but your wrong.

I work with Autocad all day every day, it "IS NOT" a 64Bit program. I also do beta testing for Autodesk and again there is nothing in the forseable future for 64Bit.

"The 64-bit processors do not provide any benefit to precision. The main benefit of 64-bit is the addressable memory, which does not equate to precision. You do not really gain any speed at this point either."

That quote was taken from...
http://autodesk.blogs.com/between_the_lines/2004/01/the_64bit_cad_m.html

9:48 AM, June 15, 2006  
Blogger RawSushi said...

64bit computing doesn't make sense for desktop computing. even for most of the servers it doesn't make sense right now!

i predict a slow transition to 64bit over the next 5 years.


Go around and read up on recent IT news and you will find that virtualization and consolidation is the next big thing on the horizon to counter growing system management costs. Virtualization only makes sense when you have big enough systems to carve into smallers systems. This means that increasingly, servers with 4 or more sockets and 8GB or more RAM will be more and more in demand. Virtualization will drive 64-bit computing adoption much faster than your 5 year estimation. AMD is sitting on a really nice sweet spot here.

I don't even understand why Intel wants to dabble in virtualization. There isn't much point in carving up a small system into tiny parts.

9:57 AM, June 15, 2006  
Anonymous Anonymous said...

"But some dude with a Core Duo won't be so lucky, a few months from now, he will find he has to make sure he gets a 32 bit app."

What the hell are you talking about?

I have Vista Beta 2 64Bit, running on an Intel 651. I have heard nor read such things about having to run only 32Bit apps on it. Its a 64Bit opperating system, but also runs 32Bit apps.

So what makes you think it cant run a 64Bit app?

Only here on your blog to I read such biassed statements.

Point me to some of those facts you like to talk about.

10:00 AM, June 15, 2006  
Blogger Sharikou, Ph. D. said...

64bit computing doesn't make sense for desktop computing. even for most of the servers it doesn't make sense right now!

AMD64 on 64 bit means higher performance, about 10-40%. Even with less than 4GB ram, AMD64 64bit programs are substantially faster than 32 bit. Search this blog, you will find links to various benchmarks on Windows and Linux. Part of the erason is under 64 bit, AMD64 has double number of registers. Also, today, the .NET engine has to emulate 64 bit with 32 bit, with true 64 bit, the emulation becomes native. The desktop I am using is just a Athlon 64 2800+ Socket 754 with 1GB ram. Windows x64 boots much much faster than Windows XP 32 and everything runs much faster.

10:04 AM, June 15, 2006  
Blogger Sharikou, Ph. D. said...

This means that increasingly, servers with 4 or more sockets and 8GB or more RAM will be more and more in demand.

The argument on RAM is an Intel deception trying to divert people's attention. 64 bit is more about performance. People have been doing 64 bit for a very long time already. From Nintendo, to PS2, to Xbox360, why do they have to be 64 bit? They don't even have 512MB of ram. It's speed.

So far, AMD's true 64 bit runs 10-40% faster under 64 bit mode. Intel's EM64T is often 10-20% slower in 64 bit mode.

10:10 AM, June 15, 2006  
Anonymous Anonymous said...

This means that increasingly, servers with 4 or more sockets and 8GB or more RAM will be more and more in demand. Virtualization will drive 64-bit computing adoption much faster than your 5 year estimation.

this is true from a hardware view. probably 80% of the hardware sold right now is capable of 64bits. it's completely illusive from a software standpoint. real 64bit windows support is comming next year with vista. companies are usually extremly conservative with their hardware and software upgrade strategy, so many will wait up to 2 to 3 years with a transition to vista. the software vendors know this behaviour - so no immediate transision of apps.

10:30 AM, June 15, 2006  
Blogger Sharikou, Ph. D. said...

Point me to some of those facts you like to talk about.

So, you thought Intel's Core Duo support 64 bit? Intel is dumping a lot of 32 bit crap to clueless customers.

10:37 AM, June 15, 2006  
Anonymous Anonymous said...

"Well I'll be damned, it's a 300mm 65nm production wafer AND a 65nm AM2 processor up and running!"
OMG!! I just crapped in my pants!!

wow.. looks like AMD is getting SOMETHING from their 300mm wafer plant. something AMD fanboys are very proud of huh...

"Conroe/Merom/Woodcrest out the door? Because they can't compete with AMD at 90nm with thier own 90nm process, that's why they had to move to 65nm and now rushing to 45nm! Thanks to the use of APM software AMD will be able to achieve near 100% yield right at start up at the FAB, they've already said 65nm chips will be here IN DECEMBER and Intel knows it will be right back in the same predicament. Any performance lead Conroe has will quickly be dissappearing when AMD moves to 65nm and eventually to K8L, even as it finally hits the shelves."
OMG I JUST CRAPPED IN MY PANTS AGAIN!!!
humm originally AMD said 65nm will hit the shelve BEFORE Conroe's launch. oh wait.. i think they changed it, for undisclosed reasons, that all the new AM2 are 90nm. well i'll be damned. if AMD is getting near 100% yield on 65nm using APM, then why 65nm AM2 are delayed? THINK!! with logic.

as i said long before when i was fooled by sharikou.. transition to 65nm with the same architecture doesn't give you an automatic increase of 20% in performance. example? take a look at PD line up. does 900 series prove a less failure than 800 series? not really, and i'm sure its the same with AM2.

for all you AMD fanboys.. PRAY! pray that K8L really has what it takes to outperform a Conroe, or shall we say, Kentsfield/Clovertown?

10:38 AM, June 15, 2006  
Anonymous Anonymous said...

AMD64 on 64 bit means higher performance, about 10-40%. Even with less than 4GB ram, AMD64 64bit programs are substantially faster than 32 bit.

why is this so? because they added 32bit to each register? not really!
in 64bit mode the processor can use twice as much registers, which can't be used in 32bit mode. the performance gain is created not by 64bit in each register, but by more registers...

10:43 AM, June 15, 2006  
Anonymous Anonymous said...

"AMD64 on 64 bit means higher performance, about 10-40%. Even with less than 4GB ram, AMD64 64bit programs are substantially faster than 32 bit. Search this blog, you will find links to various benchmarks on Windows and Linux. Part of the erason is under 64 bit, AMD64 has double number of registers. Also, today, the .NET engine has to emulate 64 bit with 32 bit, with true 64 bit, the emulation becomes native. The desktop I am using is just a Athlon 64 2800+ Socket 754 with 1GB ram. Windows x64 boots much much faster than Windows XP 32 and everything runs much faster. "

i believe the shift to 64bit will definately be done in the next half year or so. however, it looks like Intel hits the right time of processor migration.

AMD64 was out in 2003, when 64bit apps were so scarce. even though AMD advertised AMD64, most companies and normal users don't see any incentive of transiting to 64bit. after 3 years, we still don't see any large incentive for an immediate transition to 64bit.

10:53 AM, June 15, 2006  
Anonymous Anonymous said...

"Now Conroe, being Intel next generation, wouldn't it be fair to assume it would perform better than Netburst in 64Bit enviornment?"

Intel's IA64 is a copy of AMD64...

Lot's who's cloning who now?? :)

To the poster with the Core Duo... (NOT Core 2 Duo that's NOT EVEN OUT YET), it's 32-bits buddy.

The Apple deal makes sense to me know... Who on earth can offload a bunch of obsolete 32-bit CPU's on the mass market? APPLE!!

10:59 AM, June 15, 2006  
Anonymous Anonymous said...

I really don't care about 64-bit at this moment. And, I plan to use my notebook only for the next 2-3 years, and contrary to what you may think, I am sure than 32-bit apps will be more than 64-bit ones for the next 5-6 years, at least in the mobile space.

What is more important to me is performance. Today, I find that the Core Duo gives better performance than similarly prices Turions in alsmost every application, and that is what matters to me. I will think of 64-bit as the primary factor when I buy my next notebook, maybe after 3 years.

I am not saying that 64-bit is nice. I would any day choose the Merom over Yonah. But I will not sacrifice performance in 32-bit apps, which are definitely more in number and importance than 64-bit ones, simply for gloating power that "my lappy can do 64-bit".... And AMD may own the server and DT space at the moment, but really, in the mobile space, the Turions suck.

11:01 AM, June 15, 2006  
Anonymous Anonymous said...

"i believe the shift to 64bit will definately be done in the next half year or so. however, it looks like Intel hits the right time of processor migration."

I would be really surpised if even Vista comes out within 6 months.

I guess we can hope for at least a few 64-bit apps once Vista hits the scene, which looks like will be delayed again and again till 2025.

11:03 AM, June 15, 2006  
Anonymous Anonymous said...

"Intel's IA64 is a copy of AMD64..."

Do you even know what IA64 is?

11:08 AM, June 15, 2006  
Anonymous Anonymous said...

let's look at this objectively. pretend you're an average consumer, what do you look for in a laptop?

Price! then probably specifications.

if Intel is dumping Yonah like crazy, as you said, then Centrino Duos should be the most appealing laptop platform for an average consumer. So far, Intel has basically held its ground in mobile segment for years. there is almost no way for average users to suddenly switch from Centrino to Turion, just because Turion can run 64bit.

11:23 AM, June 15, 2006  
Anonymous Anonymous said...

"why is this so? because they added 32bit to each register? not really!
in 64bit mode the processor can use twice as much registers, which can't be used in 32bit mode. the performance gain is created not by 64bit in each register, but by more registers... "

64-bit CAN give a boost to precision: 2^64-1 (the maximum 64-bit value) is a considerably large number. In other words:

18446744073709551615 owns 4294967295

There are some exclusives for AMD processors too. For example, here's a memory copy technique that gives a good performance boost in 64-bit: load a whole cache line into 64-bit ALU registers then store into memory from there, this works great because A64/Opteron loads are executed in-order and cannot bypass stores but they can execute 2x 64-bit loads per clock cycle and the 64-bit ALU are capable of 2x 64-bit stores every cycle. Intel processors do that out-of-order so they cannot reap rewards as much.

There's a possibility K8L will fuse concurrent 64-bit ALU loads or stores into 128-bit ones, in some form of macro-op fusion, who knows? It just makes sense since AMD64-optimized code makes use of 64-bit ALU stores.

11:37 AM, June 15, 2006  
Blogger netrama said...

FYI ..
the Intel EMT64 doc is a cut and paste of the AMD64 doc and ..even some of the spelling mistakes are repeated.

11:39 AM, June 15, 2006  
Anonymous Anonymous said...

"So, you thought Intel's Core Duo support 64 bit? Intel is dumping a lot of 32 bit crap to clueless customers."

I was thinking of the Core 2 Duo (Conroe - Merom), not Core Duo (Yonah)...

These test show it working on 64Bit...

http://www.xtremesystems.org/forums/showthread.php?t=101344

The inquirer...

"Intel Merom is 64-bit, and with 4MB L2 cache"

11:47 AM, June 15, 2006  
Anonymous Anonymous said...

To the guy who posted this...

"I'm sorry! Did YOU say something???"

Those images sure look alot like these here...

http://www.anandtech.com/tradeshows/showdoc.aspx?i=2710

12:02 PM, June 15, 2006  
Anonymous Anonymous said...

"damn 64 bit blows " (Conroe performance)

That wasn't from Sharikou ;)

12:14 PM, June 15, 2006  
Anonymous Anonymous said...

Hey Sharikou, in what field do you hold a ph.d in?

12:55 PM, June 15, 2006  
Anonymous Anonymous said...

"I find that the Core Duo gives better performance than similarly prices Turions in alsmost every application"

You must be kidding. How did you "find" that? You benchmarked & compared the prices? Did you use the same harddrive and memory? What chipsets are you using? Do you know that notebook performance bottleneck for most end users is never the processor, but all these other factors?

Turion64 has better performance/price than Dothan, period. A Dothan-based that performs like my T64, and weight as light as mine and battery as long as mine, cost at least $1.5K, but mine is just over $1K.

Yonah is just two Dothans sharing one L2; I have no experience with T64 X2 but won't be surprised if it exceeds Yonah's performance/price ratio. BTW, Yonah's battery life sucks:
http://www.tomshardware.com/2006/01/16/will_core_duo_notebooks_trade_battery_life_for_quicker_response/page21.html"

Remember that Turion64 X2 has the same TDP as Turion64-ML (35W) and thus should have similar battery life, which is far better than Core Duo.

Lesson: AMD releases stuff when their goods are good enough; Intel releases stuff to boost its market and fool its customers.

1:03 PM, June 15, 2006  
Anonymous Anonymous said...

"OMG I JUST CRAPPED IN MY PANTS AGAIN!!!"

Sounds like a personal problem buddy...

"humm originally AMD said 65nm will hit the shelve BEFORE Conroe's launch. oh wait.. i think they changed it, for undisclosed reasons, that all the new AM2 are 90nm."

Drop the crackpipe man, I don't remember reading ANYWHERE about AMD saying 65nm will hit the shelves BEFORE Conroes's launch, my point is I estimate by the time you see it on shelves(OEM will gobble them all up at launch and afterwards for some time as they ramp production as Intel plans at best 30% market saturation of Core2 at year end AFAIK) AMD 65nm will start shipping. I guarantee that you are going to see Conroe but expect to pay out the butt for it early on as it will be in small volume until the end of the year.

"Well i'll be damned. if AMD is getting near 100% yield on 65nm using APM, then why 65nm AM2 are delayed? THINK!! with logic."

Well I guess you're just damned then, take some of your own advice pal, as you obviously have no clue about AMD production schedule of AMD's 65nm chips, particularly FAB36. The plant opened @ 90nm in Nov. 2005, began shipping 90nm product shortly thereafter and started 65nm production in March this year.

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2564

To get a plant up and running, transitioned to 65nm and now at already mature yields is nothing short of impressive in light of Intel's FAB efforts.

http://eetimes.com/news/semi/showArticle.jhtml;jsessionid=J1QN0ZO3GEH4OQSNDBCSKH0CJUMEKJVN?articleID=160502234

Even with all of the massive resources at Intel's disposal, they still take far longer on average to get a production FAB up to the levels of what AMD did with FAB36. Intel perfects a process in thier development FAB and then take many months to near a year to get a process down at thier production FAB, then it takes however long to ramp the process to what they usually see as acceptiable yields (50%-60%), 3-6 months I am estimating from what I have read. Intel has 2 65nm FABs pumping Core2 at around those yields and AMD is pumping out near 100% with one FAB. The difference between AMD and Intel FAB technology and processes is pretty clear.

"as i said long before when i was fooled by sharikou.. transition to 65nm with the same architecture doesn't give you an automatic increase of 20% in performance. example? take a look at PD line up. does 900 series prove a less failure than 800 series? not really, and i'm sure its the same with AM2."

Intel and AMD's production processes while not exactly night and day, they are very distinct, period. See above. Just because Intel can't do it well doesn't mean AMD can't, wait and see.

"for all you AMD fanboys.. PRAY! pray that K8L really has what it takes to outperform a Conroe, or shall we say, Kentsfield/Clovertown?"

AMD doesn't need a prayer, they have a well thought out plan and strategy for all fronts, server, mobile and desktop and a brilliant team of engineers the likes of Intel could only dream of, namely quite a number of ex-DEC geniuses who are masters of server CPU architectures, like the Alpha and even Itanium. Intel's only saving grace is the Isreali team that tweaked the hell out of a Pentium III to what we see now as Core 2, they have a ways though to go to develop a true server architecture that will measure up to AMD64, it's intergrated memory controller and ccHT 3.0. Intel is just now finally issuing some semblance of a plan and strategy to react to the K8 to defeat it's performance but as it seems, once again, AMD is anticipating thier move and have a response ready and waiting when the time is right. Whoever is best prepared usually ends up on top in the end, AMD has expected Intel's response with Core 2 and already have it's more than capable answer. Go read up on the K8L's L3 z-ram cache, double FPU units, tweaked OOO execute units, ccHT 3.0 and so forth. Don't tell it won't put a hurting on the Conroe because everything it basically has, AMD has DOUBLED on the K8L's design.

1:26 PM, June 15, 2006  
Anonymous Anonymous said...

in the engineering field
64bit is already here.
from e-m simulators like ansoft HFSS to harmonic balance simulators like Goldengate. More and more tools are available which take advantage of 64bits. If you're using 32bits you start with a disadvantage.

1:43 PM, June 15, 2006  
Anonymous Anonymous said...

"I guarantee that you are going to see Conroe but expect to pay out the butt for it early on as it will be in small volume until the end of the year."

What kind of quaanities will we see from AMD then (2006)?

In 2007?

This is quoted from your link to Anand...

"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"

What does "substantial amount" mean?

Could somebody show me where "AMD" says, (not some blog, or forum), they are near 100% yields?

Thanks

2:28 PM, June 15, 2006  
Anonymous Anonymous said...

Yonah is just two Dothans sharing one L2; I have no experience with T64 X2 but won't be surprised if it exceeds Yonah's performance/price ratio. BTW, Yonah's battery life sucks:
http://www.tomshardware.com/2006/01/16/will_core_duo_notebooks_trade_battery_life_for_quicker_response/page21.html"


you better do your homework, befor you spout out something that is allready known and it is not the fault of the processor.

http://www.tgdaily.com/2006/01/28/toms_hardware_uncovers_power_drain_issue/

and yonah is not just 2 dothans stuck together...

i hate it, when so much fud is posted.

2:41 PM, June 15, 2006  
Anonymous Anonymous said...

Intel's IA64 is a copy of AMD64

no it's not.

3:21 PM, June 15, 2006  
Anonymous Anonymous said...

Yonah is just two Dothans sharing one L2; I have no experience with T64 X2 but won't be surprised if it exceeds Yonah's performance/price ratio. BTW, Yonah's battery life sucks:

have you really read (and understood) the article you reference? in the conclusion thg states, that the larger power consumption of the core duo system is due to the chipset/graphic card... a really excellent article to prove your arguments! *lol*

3:34 PM, June 15, 2006  
Anonymous Anonymous said...

What chump wouldn't want to future proof their investment in expensive laptop by going 64 bit? At $1278 on HP site for Turion 64 X2 based laptop, you'd have to be a howling idiot to stick with 32 bit if you plan on keeping it over next 4 years or so.

5:38 PM, June 15, 2006  
Anonymous Anonymous said...

"Drop the crackpipe man, I don't remember reading ANYWHERE about AMD saying 65nm will hit the shelves BEFORE Conroes's launch"
AM2 was orignally designed at 65nm only, but somehow, somewhere, 90nm AM2 was released. humm.. i wonder what was that all about.

"Intel and AMD's production processes while not exactly night and day, they are very distinct, period. See above. Just because Intel can't do it well doesn't mean AMD can't, wait and see."
Intel moved up their launch date for Conroe. AMD moved back their launch day for AM2 65nm. I guess this pretty much says it.

"Intel's only saving grace is the Isreali team that tweaked the hell out of a Pentium III to what we see now as Core 2,"
so that means Intel's Pentium III actually outperforms AMD's k8 architecture. i don't know if i should feel sorry for intel or for k8.

"Intel is just now finally issuing some semblance of a plan and strategy to react to the K8 to defeat it's performance but as it seems, once again, AMD is anticipating thier move and have a response ready and waiting when the time is right."
lolz. Conroe was actually developed way back in August of 2005. Conroe's release was completely out of AMD's expectation, because they thought k8 can basically whooped any Intel line up. well.. i guess now, its time for them to start scrambling.

"Don't tell it won't put a hurting on the Conroe because everything it basically has, AMD has DOUBLED on the K8L's design."
If AMD can really create something to outperform Conroe, then good for them. i don't want to see Conroe to take all the spotlight, or it will really hurt us as consumers.

6:51 PM, June 15, 2006  
Anonymous Anonymous said...

"What chump wouldn't want to future proof their investment in expensive laptop by going 64 bit? At $1278 on HP site for Turion 64 X2 based laptop, you'd have to be a howling idiot to stick with 32 bit if you plan on keeping it over next 4 years or so. "
humm i doubt the transition to 64 bit is that quick. Not only Vista is 32bit fully compatible, i believe most programs will be 32bit compatible also. its not like everyone has a ransom of 1300USD for AMD.

6:57 PM, June 15, 2006  
Anonymous Anonymous said...

My understanding about using 64-bit OS is that you have to have all the drivers 64-bit as well. All drivers must be 64-bit native. How long before my wireless card, my printer, scanner, fax/modem and bluetooth devices fully support 64-bit. I'd like to make the switch but won't do it until I can take all my devices with me and not upgrade them or have to replace them.

8:45 PM, June 15, 2006  
Anonymous Anonymous said...

"Every wafer will not produce 100% good chips
AMD's yield is over 90%. That's how it could take 22% market with one 200mm FAB."

90%? Are you kidding?! Do you know anything about chip production to make that kind of statement?

NOT according to Charles Demerjian of The Inquirer.. See http://www.theinquirer.net/?article=32452 quotes "Let us throw another wooden shoe into the factory cogs. The Rev. F cores are approximately 20 per cent larger than their earlier Rev. E brethren. When considering such things as lower yields due to larger die sizes (greater chance of defects per centimetre squared), not to mention more dies being lost around the edges due to chips being square on a round wafer, AMD is looking at between 30 and 40 per cent fewer good dies per wafer, depending on the maturity of the yields."

Got fragged by another AMD fanboy?

2:25 AM, June 16, 2006  
Anonymous Anonymous said...

I think that some people are really not very informed:

-First came the MMX processor then software that use MMX.
-Then came the SSE processor then software that use SSE.
-Now came the x86-64 processor then I expect software that uses x86-64.

Now do you want me to develop 64bit programs with a 32bit processor?! How do I do that?!?!?!?

How do I run 64bit applications (and will) on a 32 bit processor?

4:14 AM, June 16, 2006  
Anonymous Anonymous said...

How long before my wireless card, my printer, scanner, fax/modem and bluetooth devices fully support 64-bit.

it's very likely, that you will never get 64bit drivers for all of your hardware.

5:26 AM, June 16, 2006  
Anonymous Anonymous said...

"AM2 was orignally designed at 65nm only, but somehow, somewhere, 90nm AM2 was released. humm.. i wonder what was that all about."

Obviously your reading and comprehension skills are lacking, AMD explicitly said FAB36 was doing 90nm FIRST then transitioning to 65nm later, yes that means socket AM2. Try reading SLOWER, perhaps you will absorb more information/facts and pull less out of your ass sir.

"Intel moved up their launch date for Conroe. AMD moved back their launch day for AM2 65nm. I guess this pretty much says it."

What? Again, drop the crackpipe buddy, it's screwing with your brain! AMD originally had the AM2 date for 6/6/06 at 90nm and they INSTEAD pushed it forward as they are NOW out on the market. Where is Conroe I wonder? Oh yea, we're still waiting. They have long said since they gave a solid date for 65nm that it would be here in DECEMBER, how many times must I repeat this for you to understand???

"so that means Intel's Pentium III actually outperforms AMD's k8 architecture. i don't know if i should feel sorry for intel or for k8."

I could care less about the K8's performance at this point, it's served it's purpose for AMD and at the end of it's life. The K8 for all intents and purposes was a overhauled K7.(as are most next gen chips, see the evolution of the Pentium series and now the Core series) The K8 is 3 years old now and due for a overhaul, AMD already knows this and has it's successor ready. Like most sensible people, I knew Conroe would surpass the K8, Intel would be commiting suicide if they didn't produce a chip that could do so.(see the Willamette, socket 423 P4 bested by Intel's own P3 for example) Honestly, how much performance is the real question it seems though. 20%-40% is quite outlandish and overexaggerated IMHO seeing as they are close in design, save the difference in cache. (Intel's marketing in full effect along with thier shady business practices, Mr. Sharikou goes into plenty of detail here for that)

"lolz. Conroe was actually developed way back in August of 2005. Conroe's release was completely out of AMD's expectation, because they thought k8 can basically whooped any Intel line up. well.. i guess now, its time for them to start scrambling."

Keep on dreaming and try doing some research into the matter, you're better at spreading FUD than Intel's marketing team.

The Isreali team has been working on Conroe basically far longer than that, what do you think Yonah and Banias (Pentium M) evolved into??? Just like it's predecessors, Conroe is nothing more than a mere evolutionary design, NOT A REVOLUTIONARY ONE as you Intel cheerleaders like to tout it. Yea, it beats the K8, a 3 year old processor, I say to Intel it's about time they get it right, took them long enough. However they won't be able to enjoy that performance advantage long enough IMHO. As I have stated before, AMD already has Conroe's answer coming, it's just not coming out as fast as some of us impatient people would like.

"If AMD can really create something to outperform Conroe, then good for them. i don't want to see Conroe to take all the spotlight, or it will really hurt us as consumers."

Hey, I'm all about competition and cheap prices, but I ain't down for anti-trust violations or monoplistic acts by a massive corporation like Intel.(feel free to read up on www.faceintel.com, or the AMD vs Intel anti-trust suit) Once AMD gets into a 50%/50% market share with Intel, Intel starts acting like a fully legally, morally competitive company and AMD gets some respect as a equal alternative in the industry with clients as Intel enjoys, I'll be happy along with most everyone else. We'll all enjoy competitive prices, advancing technology due to the competition (like Nvidia and ATI) and then maybe these stupid "processor wars" will finally come to an end.

8:40 AM, June 16, 2006  
Blogger Richard P said...

>>>
90%? Are you kidding?! Do you know anything about chip production to make that kind of statement?

NOT according to Charles Demerjian of The Inquirer.. See http://www.theinquirer.net/?article=32452 quotes "Let us throw another wooden shoe into the factory cogs. The Rev. F cores are approximately 20 per cent larger than their earlier Rev. E brethren. When considering such things as lower yields due to larger die sizes (greater chance of defects per centimetre squared), not to mention more dies being lost around the edges due to chips being square on a round wafer, AMD is looking at between 30 and 40 per cent fewer good dies per wafer, depending on the maturity of the yields."
>>>

Wow. Clearly stupidity reigns supreme with some of you. Mature yields are, to AMD, 90% or higher. Well, 65nm is seeing mature yields. The Inquirer quote you have here, is not talking about the yield percentage. He's talking about the AM2 die being bigger, and therefore reducing the number of physical dies available on a wafer. Yield percentage is the number of functioning dies as a total of the physically complete dies. There are always die around the edges that have parts cut off because of the circular nature of the wafers. These die don't count in the yield percentage. When Charlie mentions lower yields, he means fewer total processors per wafer, not necessarily more "defective" CPU's per wafer.

Clearly, you are the one that does not understand CPU manufacturing. Intel historically has notably lower yields, but they have so many fabs that it doesn't hurt them. AMD has to squeeze out every chip. Because they are doing so well, thanks to APM, and thanks to the extra space they dredged up by moving some processes out of the main clean room/s, AMD is able to actually produce more from FAB 30 then it was rated to produce.

Do us all a favor and get a clue before posting something like this again.

6:59 PM, June 16, 2006  
Anonymous Anonymous said...

Richard P said..."Clearly, you are the one that does not understand CPU manufacturing. Intel historically has notably lower yields, but they have so many fabs that it doesn't hurt them. AMD has to squeeze out every chip. Because they are doing so well, thanks to APM, and thanks to the extra space they dredged up by moving some processes out of the main clean room/s, AMD is able to actually produce more from FAB 30 then it was rated to produce."

The enlighten us as the ACTUAL yield percentage of these AMD wafers? 90% sounds UNREALISTIC...

12:27 AM, June 17, 2006  
Anonymous Anonymous said...

"Wow. Clearly stupidity reigns supreme with some of you. Mature yields are, to AMD, 90% or higher.

Says who ? For example I can say that mature yeilds are around 60%.

I can also say that X2s have yeilds around 30-40% while Intel gets in the mid 70s % with Cedar Mill dies.

And finally I can also say that 8% of AMDs production is dual core while for Intel that is 60%.Yes , you see it well , 60%.

Can you contradict me ? I thought so...

"Well, 65nm is seeing mature yields. "

Probably that's why 1st 65nm CPUs from AMD will be all running below 2.4GHz and the FX64 will still be using the 90nm process which is pushed to the limit.

Let me give you a clue : when a smaller process sees mature yeilds it is at least capable to beat the previous generation.

1:58 AM, June 17, 2006  
Blogger Altamir Gomes said...

About the Cinebench 64-bit bench @ xtremesystems forums, processors based off the "old" A64 architecture gain over 30% from going 64-bit in CB, while Conroe earns only a 15% speedup.

6:32 AM, June 17, 2006  
Blogger Richard P said...

Man, some of you don't read press releases very well do you. There is plenty of information about AMD's yields. They never state their exact yields, but during a conference call it was stated that their yields were over 90%. This yield rate has also been tied to the statement of what AMD defines, today, as mature yields. Especially since Mature Yields has also been specifally stated, in documents from AMD, to mean same current yields from the previous process; i.e. When AMD says that 65nm is hitting "mature yields" they are directly stating that they are meeting the same yields as the previous (90nm) process, which was stated, by the VP I believe, to be above 90%.

If you are still too much of an Intel Fanboy (or an idiot Intel employee who thinks they know their enemy so well), simply due the math. Look at the fact that Fab 30 is producing wafers at 147% of capacity, then look at how many CPU's sold for the quarter (AMD has been sold out in 4Q05 and 1Q06, nearly for 2Q06) and you'll find that AMD has to be yielding functional processors at a very high percentage.

It's clear to me that the two of you who (or is it one) that responded to my comment is/are illinformed about this market. Being a stock holder, I've followed it quite closely. Haven't dumped any stock, either, because I see what AMD is planning to bring out in two quarters, and I can see that AMD will still rule the price/performance/power ratios. Conroe will beat AMD for a couple of quarters, at best, but there's already a lot of speculation that yields aren't great, since Intel would switch as fast as possible if they were. Netburst is an iventory drag on them, so they want to convert over as fast as possible, yet they've stated 20% by year's end.

As far as the second poster's last statement, it is you who is clueless. Again, AMD's whole definition of mature yields is same-same new process yielding same speeds and functionality percentage, or better, as old process. It may not be your definition, or Intel's, but it is AMD's, and they have stated 65nm is mature. Their 65nm process is also, by the way, superior to Intel's, so don't be shocked if 65nm X2's hit around 3.2-3.4GHz out of the block, maybe higher, though AMD may or may not sell them at that speed initially.


Please get informed, so we can stop informing you.

5:51 PM, June 17, 2006  
Anonymous Anonymous said...

To Richard P...

In reading your last post I have a few questions maybe you can answer.

As I understand chip manufacturing not all chips are perfect (obviously), and there imperfections inturn, make them lower quality chips, say 512KB of the 1MB cache is bad, thus making a lower quality chip?

Looking at AMD pricing for FX62, and most of their top chips it would seem that there yield for high performance chips would not be great.

What percent of the mature yields are FX or there top chips?

Next, could you point me to a release date for K8L? I ask because I have searched but am not having any luck in that area.

Does AMD overclock as easily as Intel Netburst? I am curious because I hav an OC'ed Intel 651, and running some benchmarks it is doing quite well(vs an FX60), not aginst multithreaded tasks though.

Granted its a 65nm procesor, but if AMD are able to achieve 3.2 or 3.4, even at 65nm, they are going to be pushing their TDP, right?

Thanks, and I await your response.

PS: GOOOOOOOO BUCS!!!!!!!!!!!

9:38 PM, June 17, 2006  
Anonymous Anonymous said...

"They never state their exact yields, but during a conference call it was stated that their yields were over 90%. "

Why do you lie ?
AMD stated 2 things :

-SRAM yields are mature
-65nm development complete

Where does it says that yields are 90% or that 65nm has mature yeilds ? I hope you can understand the difference between SRAM and logic.

90% for complex chips like MPUs is impossible.

Cell for example which is similar in size with the X2 has on IBM/Sony process yields around 5-15%.

" Look at the fact that Fab 30 is producing wafers at 147% of capacity, then look at how many CPU's sold for the quarter (AMD has been sold out in 4Q05 and 1Q06, nearly for 2Q06) and you'll find that AMD has to be yielding functional processors at a very high percentage."

First of all Fab 30 produced at 147% because umh , they increased capacity by 47% ? There was an article on The Inq which explained this very well.

As for sold out , here's one for you : AMD has a $300 million inventory.How are you sold out when you have inventory ?

"Their 65nm process is also, by the way, superior to Intel's, so don't be shocked if 65nm X2's hit around 3.2-3.4GHz out of the block, maybe higher, though AMD may or may not sell them at that speed initially."

Hahaha...what a joke.
100nA/um@1V
Intel (P1264) 710/1210 NFET/PFET

200nA/um@1V
AMD/IBM/TOSHIBA DC 700/1137 and AC 735/1259

Scaling Intel (P1264) to 200nA/um@1V we have ~740/1280

So , we have Intel's plain bulk SS vs. AMDs SS-PD-SOI with DSL and other enhancements at 65nm and still Intel manages to eek out a victory.

Problem is too many here live in an alternate reality , make up BS at night and state it as a fact next day.

1:12 AM, June 18, 2006  
Blogger Ajay S. said...

wow, I put off posting on this thread when it had 60 comments because I tought the thread had died down.

There seems to be lot of BS here, many guys seem to have confused Yields with Fab Capacity,

AMD is using APM, which they say is allowing them to use the FAB beyond its planned capacity. So FAB 30 can do 30,000 WSPM instead of the planned 20,000 WSPM.

http://www.theinquirer.net/?article=30832

APM which includes about 450 patents, is supposed to be so good, that even Chartered opted to use it in Fab7 and is now able to ramp up quickly with better yields than earlier

http://www.theinquirer.net/?article=30831

Yields is the average number of functional chips produced per wafer, and chips with larger die areas have lower yields. AMD has never claimed to have superior yields than industry average, and says its at about 50%

http://www.fabtech.org/content/view/1390

Coming back to Conroe vs Athlon debate, I think the question will be answered when some site does "Full Review" (and not a Intel specified one) after Conroe launch, and with the new AMD prices. Cost vs Performance should be a deciding criteria.

One should also note that I can replace the 90nm AM2 chip with a 65nm one when its out

10:28 AM, June 18, 2006  
Blogger Richard P said...

>>
Why do you lie ?
AMD stated 2 things :

-SRAM yields are mature
-65nm development complete
>>

First, AMD is a processor company, as Spansion's been sold off (and they do FLASH, BTW).

Second, one rep mentioned that 65nm was now mature.

>>
90% for complex chips like MPUs is impossible.

Cell for example which is similar in size with the X2 has on IBM/Sony process yields around 5-15%.
>>

This is where you make a total fool of yourself. AMD, in a telecon, from the VP, no less, stated directly that 90nm yields were near 100%. Again, they never state exact yields.

I'm not aware of the yields of the IBM Cell, but they have been rumored to be horrible. That has no bearing on the X2, nor the K8L. IBM has always had a great process, but hasn't been very good at processor design, from a yield stand point.

The whole point of AMD's APM is to get yields over 90% and to allow the rapid transition to new Fabs and new processes.

>>
As for sold out , here's one for you : AMD has a $300 million inventory.How are you sold out when you have inventory ?
>>

Once again, your show your lack of knowledge. CPU's aren't manufactured in a day, or a week. It takes quite a few weeks to go from a wafer to validated/binned core. AMD has inventory because inventory includes parts and materials used to make the CPU's; i.e. wafers, metals, and chemicals. They are all considered inventory. I've been in manufacturing since '95, and every place counts not only the finished product, but the parts used to make those products. AMD had sales, on the books, for all CPU's they could produce during 4Q05 and 1Q06. This is why pricing didn't change much during that time frame. When demand meets or exceeds supply, the vendor can charge top dollar for their product. Econ 101 there, pal.

>>
Hahaha...what a joke.
100nA/um@1V
Intel (P1264) 710/1210 NFET/PFET

200nA/um@1V
AMD/IBM/TOSHIBA DC 700/1137 and AC 735/1259

Scaling Intel (P1264) to 200nA/um@1V we have ~740/1280

So , we have Intel's plain bulk SS vs. AMDs SS-PD-SOI with DSL and other enhancements at 65nm and still Intel manages to eek out a victory.
>>

I'm not a CPU engineer (Test Engineer for Avionics, actually), but comparing transistor design/performance, with process performance seems to be the wrong way to evaluate the process. I'll be the first to admit that Intel's 65nm transistor design is superior to AMD's 90nm transistors. But we don't know anything about AMD's 65nm transistors, do we. Besides, I'm not sure what Intel does, but AMD has 3 styles of transistors on the current K8 process. One is high speed/high leakage, another is moderate speed/moderate leakage, and a low speed/low leakage one. If I remember Dirk correctly, AMD uses the high speed for 90% of the transistors, and the rest for the remaining 10% where speed is less critical. This allows AMD to customize areas of the core depending on what is more important. So the comparison seems even worse under this light (NFET/PFET comps, that is).

12:33 PM, June 18, 2006  
Blogger Richard P said...

>>
As I understand chip manufacturing not all chips are perfect (obviously), and there imperfections inturn, make them lower quality chips, say 512KB of the 1MB cache is bad, thus making a lower quality chip?

Looking at AMD pricing for FX62, and most of their top chips it would seem that there yield for high performance chips would not be great.

What percent of the mature yields are FX or there top chips?
>>

FX pricing is only partly based on yields. The fact that most Athlons can hit FX speeds when overclocked shows just that. The price for FX's is more driven by the speed bin yield then the functional yield (what I've been talking about in terms of "mature yields" is functional yields not speed bin yields). Meaning that AMD may be getting 90-95% of the physically viable CPU's to function at a speed that is sellable, but maybe only 5-10% can be sold as an FX. Then demand will set how many of those will be actuall sold as FX's.

Both AMD and Intel have frequently down-clocked CPU's so that they can sell CPU's at a lower price point, and get differentiation in their CPU portfolio. As per AMD, 90nm is near 100%. Whether that's 95 or 98 or 99%, I do not know, as they do no reveal specific yields. Quite frankly, even mentioning a range is a little unusual, but when you are producing to the max and booking all of your CPU's for sale, it becomes much harder to hide what that percentage is.

>>
Next, could you point me to a release date for K8L?
>>

K8L is scheduled for 1H07. The month I have heard most often used is April. A lot of this will depend on how far along AMD is in the process (tape out to hitting design milestones). AMD has been, for once, very, very tight lipped about it's progress. That has led to rumors at all ends of the spectrum (way behind...way early..not coming out until '08...etc...).

>>
Does AMD overclock as easily as Intel Netburst? I am curious because I hav an OC'ed Intel 651, and running some benchmarks it is doing quite well(vs an FX60), not aginst multithreaded tasks though.

Granted its a 65nm procesor, but if AMD are able to achieve 3.2 or 3.4, even at 65nm, they are going to be pushing their TDP, right?
>>

Raw GHz wise...no. Performance level wise, yes. An Athlon 64 running at 3.0GHz will equal or surpass a 4.0GHz P4 in most apps, though not all. So it really depends on what you mean. Seen a lot of Athlons overclocked into the 3.2-3.4GHz territory on water cooling, and consider that the FX-62 is a 2.8GHz part, which beats all stock Netburst CPU's in pretty much everything but a very small few benches.

As far as TDP goes, AMD's 65nm process is introducing new technology to further reduce gate leakage, while improving transistor performance. I expect to see thier TDP drop roughly 15-20W, on average, for the same speed grade. Remember that AMD always states absolute maximum power, while Intel's TDP is a average maximum, and thusly not directly comparable. Further, when you look at Intel CPU power consumption, immediately add roughly 23 watts for the northbridge. AMD has the northbridge functionality built into the CPU (hypertransport and the integrated memory controller). So AMD's TDP effective included the northbridge.

Overall, I think AMD will be able to produce a 3.4GHz cpu with a TDP in the 65W range. My understanding is that AMD's 65nm process is offering a much greater reduction in power usage than the transition to 90nm did from 130nm. Proof is in the pudding, however. We'll have to wait and see how the CPU's do when they come out (Dec. '06?)

>>
PS: GOOOOOOOO BUCS!!!!!!!!!!!
>>

Just 6 weeks 'till training camp!!

12:56 PM, June 18, 2006  
Anonymous Anonymous said...

Capacity = amount of chips you can produce

Yield = #good die per wafer/total die per wafer

When AMD states they are running F30 at 147% above originally planned capacity that means they are stating 47% more wafers per week (the yields of which could be 1%, 5%, 50%, 90%, 95%....)

Yields are product, wafer size and technology node dependent. The poster who said SRAM yield does not equal actual product yield is correct. An IC manufacturer generally works on SRAM test chips until process is "debugged" and then starts converting over development work to actual product chips. Also of note is 200mm yield is generally better than 300mm yield on a given technology node (like 90nm). Don't assume yield just carries over across the wafer size transistion.

2:14 PM, June 18, 2006  
Anonymous Anonymous said...

The poster on top of me.

So you are telling me that Intel's 90nm process is better than AMD's even with leakage problem right? Then how did AMD manage to kill Intel's ass in 90nm?

9:48 PM, June 18, 2006  
Anonymous Anonymous said...

Richard P said... "believe to be 90%"? You swallowed so easily AMD's marketing BS? I though you had worked in semiconductor fab industry before.. but from what I've read, you just "hear them"... This is why I stated "90% is UNREALISTIC" figure! Make up your mind first before discounting others...

11:04 PM, June 18, 2006  
Anonymous Anonymous said...

"First, AMD is a processor company, as Spansion's been sold off (and they do FLASH, BTW).

Second, one rep mentioned that 65nm was now mature. "

Bat clue : Spansion has nothing to do with test SRAMs.If you had a clue you'd know SRAMs are produced 1st on a new process to test its characteristics.Once everything is ok , they try to produce MPUs.

As for the 65nm , reps can say what they want, it is called " PR spin " .Same AMD reps said they won't be far behind on 65nm , yet they are 1 year late. ( this difference has never happened before )

"
This is where you make a total fool of yourself. AMD, in a telecon, from the VP, no less, stated directly that 90nm yields were near 100%. Again, they never state exact yields. "

Get a clue "AMD's first foundry partner, Chartered Semiconductor has already been able to produce fully working die with yields equal to or even better than those achieved at AMD, according to a March 15th report from Bear Stearns.The reports authors said that yields were at 50 percent at Chartered's Fab7 but sometimes even better than those achieved within AMD's own fab!"

http://www.fabtech.org/content/view/1390

That is 90nm tech.Looks like the fool is one who believes the 90% yeilds story.

"The whole point of AMD's APM is to get yields over 90% .."

That's what you'd like it to be to suit your point.Don't put your dreams as reality.90% yields on complex devices like MPUs is impossible

" meets or exceeds supply, the vendor can charge top dollar for their product. Econ 101 there, pal. "

That doesn't back up your claim of running full. Econ 102.

"I'm not a CPU engineer (Test Engineer for Avionics, actually), but comparing transistor design/performance, with process performance seems to be the wrong way to evaluate the process. "

Huh ? It is the only way.Stupid way to do it is by comparing designs on different processes and claim process X is better than process Y , but that suits AMD better doesn't it ?

" I'll be the first to admit that Intel's 65nm transistor design is superior to AMD's 90nm transistors. But we don't know anything about AMD's 65nm transistors, do we. "

I was comparing 65nm transistors for both , as for the latter part , maybe you don't know , but do not assume others don't

" Besides, I'm not sure what Intel does, but AMD has 3 styles of transistors on the current K8 process. One is high speed/high leakage, another is moderate speed/moderate leakage, and a low speed/low leakage one. So the comparison seems even worse under this light (NFET/PFET comps, that is). "

Irrelevant and intel uses that too , just that they don't brag about it as AMD. Besides , Intel has 2 65nm processes : high performance one called P1264 and low-power , extremly low leakage one called P1265.

Comparing P1265 with AMD's process is kinda hard , simply because AMD doesn't have anything in that league.

Here's a clue for you : P1265 , 1.2V and 0.1nA/um Intel gets 380/640 which is truly remarkable , in other words this process has 1000x lower leakage and still puts respectable numbers.

"So you are telling me that Intel's 90nm process is better than AMD's even with leakage problem right? Then how did AMD manage to kill Intel's ass in 90nm? "

They aren't even in the same league.
Intel's plain bulk SS vs. AMDs S-PD-SOI with DSL on 90nm

100nA/um@1V
Intel: 1.15 / 0.67 mA/um
AMD: 0.96 / 0.48 mA/um

Intel (P1262) advantage 20% NFET and 40% PFET....it isn't even funny.

The fact that Prescott is hotter than K8s is due to design implementation, or more exactly , Prescott has 50% more logic transistors than K8. ( cache is irrelevant for power )

12:54 AM, June 19, 2006  
Anonymous Anonymous said...

"I can also say that X2s have yeilds around 30-40% while Intel gets in the mid 70s % with Cedar Mill dies."

Well Intel is making COPY -> PASTE of single core chips into Double core packing.
That’s very easy to say that they have better yields, and can sell them cheaper too.

Also the TRUE dual core processor from Intel (Core Duo) is selling 250$ the cheapest (1.66Ghz).

I think if it’s that "expensive", they have yield problems too?

I heard Intel is also thinking in putting Core Duo on Desktop because they are lacking Core 2 Duo (Conroe) processors (bad yields?).

P4 will be released again (rev. D).

6:01 AM, June 19, 2006  

Post a Comment

<< Home