AMD tri core

[shargrath]

Now look at Figure 3.9, and match up 0.5cm2 and 283mm2 die size, you'll get a yield of....
Now, BM, your rebuttal please?

just looking at that its horrible im thinking that with the tri core there will be most likely some level 3 cache disabled in the cheaper tri cores to help salvage as much as they can with quads that have a defective cpu as well as level 3 cache.

 

[MU_Engineer]

Uh...what?
Why would Intel need a 3 core CPU in the first place? If one of the die fails on a dual core, it can be used as a Core Solo CPU. So, where exactly does Intel need a 3 core CPU.

They don't. The biggest die they make has two cores in it and they do exactly what you say, although a die destined for a Core 2 Quad would become a Celeron 400, not a Core Solo. The Core Solo was half of a Core Duo and is a completely different chip from the Core 2 Duos.

Another thing is that AMD would be in better shape trying to market a 3 core CPU, just to not let their quads with 1 bad die go into the garbage.

They wouldn't go in the garbage, they'd be made into Phenom X2s. This is exactly how AMD took a Manchester X2 with a bad core and sold them as single-core 3200+ and 3500+ units. AMD just doesn't want to have to turn off two of four cores when only one is bad, and they can likely sell the 3-core chip for more than a dual-core but less than a quad-core. They get more money for the same die, we get a chip that's faster than the duals but less-expensive than the quads.

The issue at hand is how to market a 3 core CPU.
What is the distinct advantage of 3 core CPUs over dual cores? And with quad core prices being low, what are the advantages of getting a 3 core over a quad core?

The advantage of a 3-core chip over a dual-core one is that you can potentially get 50% more performance at the same clock speeds. A 3-core chip will also clock higher since the thermal dissipation should be lower due to one fewer core being run. The only issue I see for this is that the quad-core prices being low would result in there not being much of a price difference between the 3 and 4-core chips and people not buying many of them.

 

[bixplus]

Speaking of wow, that was perhaps the most scathing criticism I have ever heard. .

Bahwaaaaaaa! Someone just made fun of AMD, so I (BM) must step up and defend her honor!

Go away Peewee! :lol:

... Intel says they can't make it feasible before 45nm and Nehalem. That in and of itself should gain AMD innovation points.

I'm sorry, which company has never had a quarter in the red? At the rate AMD is going, thry're going to innovate themselves out of the CPU business.

 

[TechnologyCoordinator]

Perhaps AMD should have made QFX an MCM but their first quad IS native at 65nm while Intel says they can't make it feasible before 45nm and Nehalem. That in and of itself should gain AMD innovation points.

But thank you for your comment.

I understand, but what's the advantage of a monolithic quad-core when Intel's supposedly flawed design:

a) came to market a full 10 months before AMD's
b) STILL performs better

Although, I must say I'm glad to see some quad core competition and hopefully price cuts.


AMD should abandon Quad-FX.

 

[Reynod]

AMD can do it so they will I spose... They have got to get the most out of their production process ... this seems to make good sense.

Clever idea ...

Hope it works.

I can imagine the performance might be quite good.

Baron posted something about this yesterday.

I'd imagine the L3 cache being chared across the three cores is still 2M or so ??


P.S. Jeez the anti-AMD fanboi stuff is really getting draining ... make a point with your posts or go back to licking the blue sticker on your box and drooling while your flupenthixol is still working well ...

 

[reconviperone1]

I understand, but what's the advantage of a monolithic quad-core when Intel's supposedly flawed design:

a) came to market a full 10 months before AMD's
b) STILL performs better

Although, I must say I'm glad to see some quad core competition and hopefully price cuts.


AMD should abandon Quad-FX.

I'm gonna agree and diagree with you. I think the last interation of quadfx sucked really bad, however, i think with 2 barcelona duals or quads with lower tdp and the new amd quadfx board, i think it would make a great enthusiast system. Of course you may have your wish, rumor is amd has cancelled the FASN8.

 

[MU_Engineer]

I'm gonna agree and diagree with you. I think the last interation of quadfx sucked really bad, however, i think with 2 barcelona duals or quads with lower tdp and the new amd quadfx board, i think it would make a great enthusiast system. Of course you may have your wish, rumor is amd has cancelled the FASN8.

Either way, you can still something 95% similar in buying a dual 1207+ enthusiast-workstation motherboard and two Opteron 23xx CPUs. Sure, you will have to use ECC registered DDR2-667 instead of unbuffered DDR2-1067 and there will be a locked multiplier, but you can still overclock, have 8 10h cores, and have a bunch of SATA and PCIe ports. The Opteron 23xx units are very inexpensive for DP server chips and the motherboards and RAM aren't ridiculous, either.

 

[BaronMatrix]

Well in case you're aware, Phenom and Barcelona shares the exact same architecture. The only differences are the binning, and the validation of HT3 capable.

With that being said, let's take a look at a slide from AMD's technology analyst day.

http://www.iian.ibeam.com/events/th...070726084721294707/default_large/Slide158.JPG

The defect density is at 0.5cm2 range.

Then, the following article is from ICEC, detailing yield management.

http://smithsonianchips.si.edu/ice/cd/CEICM/SECTION3.pdf

Now look at Figure 3.9, and match up 0.5cm2 and 283mm2 die size, you'll get a yield of....

I'm not saying this won't change overtime. However, given the date of presentation (7/26), it is unlikely AMD would have some major breakthrough...

Now, BM, your rebuttal please?

My rebuttal is that data does not include any specific CPU makers, is from 1996 and talks more about DRAM. Find some actual AMD data for 2007.

 

[BaronMatrix]

Bahwaaaaaaa! Someone just made fun of AMD, so I (BM) must step up and defend her honor!

Go away Peewee! :lol:


I'm sorry, which company has never had a quarter in the red? At the rate AMD is going, thry're going to innovate themselves out of the CPU business.

Which company is being investigated for illegal behavior? Which company was successfully sued for said behavior and which company got $1B for it?

 

[crow_smiling]

Either way, you can still something 95% similar in buying a dual 1207+ enthusiast-workstation motherboard and two Opteron 23xx CPUs. Sure, you will have to use ECC registered DDR2-667 instead of unbuffered DDR2-1067 and there will be a locked multiplier, but you can still overclock, have 8 10h cores, and have a bunch of SATA and PCIe ports. The Opteron 23xx units are very inexpensive for DP server chips and the motherboards and RAM aren't ridiculous, either.

Sure you can do this with AMD and Intel workstation parts but it’s still not as easy and cheap as using dedicated 2P boards that use unbuffered DDR2. Over-clocking workstation/server boards is a lot more hit and miss than with desktop boards. The CPUs are cheap for both platforms provided you can overclock them by a good margin otherwise not so.

Intel has a new chipset for 2P supposedly with regular RAM support due Q4 (San Clemente) but with the Penryn Xeon quads all seemingly running at 400 FSB there isn’t so much headroom for easy over-clocking. They are cheap and start at 2GHz ($209) allegedly. Not sure about SLI support though for the chipset!
We’ll have to wait and see how Barcelona and its derivates compare at 3GHz and above.

 

[bixplus]

Which company is being investigated for illegal behavior? Which company was successfully sued for said behavior and which company got $1B for it?

All of which have nothing whatsoever to do with your so called "innovation points", and/or making an educated business decision regarding when it is/isn't feasable to manufacture a certain product. Point is, Barcelona is large and expensive to make by comparison...so Intel seems to be making the right decision here.

But, back on topic...it does make business sense to sell a 3 core product in order to attempt to recover revenue lost when one core is bad, so good for them. On the flip side, all this AMD native quad innovation talk becomes a bit muddied when you have to announce something like this less than a week past the supposed launch date. AMD should be riding high on the wave of this new product for a couple months after launch if it were all that. And, as stated above, this is purely a business decision, not an attempt to create a new market where your competitor cannot tread.

 

[yomamafor1]

My rebuttal is that data does not include any specific CPU makers, is from 1996 and talks more about DRAM. Find some actual AMD data for 2007.

I guess you did not take a look at the figure very well... perhaps you don't want to accept it?

My point here is that, yield is a function of defect density and die size. The function is clearly written, and the figure clearly illustrated the function based on defect density.

As a result, based on AMD's published defect density (0.5cm^2), and Barcelona's die size (283mm), it can be concluded that AMD's yield will hover around 30~35%.

There is a reason AMD wants to come up with a "tri-core", and definitely not because of market demand.

 

[TechnologyCoordinator]

I'm gonna agree and diagree with you. I think the last interation of quadfx sucked really bad, however, i think with 2 barcelona duals or quads with lower tdp and the new amd quadfx board, i think it would make a great enthusiast system. Of course you may have your wish, rumor is amd has cancelled the FASN8.

I have a hard time seeing the need for eight cores. If you do need eight cores, then you're probably talking data center, in which case, you want ECC.

I guess the way I see it is if you REALLY want multiple sockets then go to the server side.

Now, if they came out with games that would use all eight cores, then that might be different!

 

[yomamafor1]

Which company is being investigated for illegal behavior? Which company was successfully sued for said behavior and which company got $1B for it?

humm.... care to provide proof on .... what you say... FUD? "which company got 1B for it?"

 

[NMDante]

Another thing is that AMD would be in better shape trying to market a 3 core CPU, just to not let their quads with 1 bad die go into the garbage.

They wouldn't go in the garbage, they'd be made into Phenom X2s. This is exactly how AMD took a Manchester X2 with a bad core and sold them as single-core 3200+ and 3500+ units. AMD just doesn't want to have to turn off two of four cores when only one is bad, and they can likely sell the 3-core chip for more than a dual-core but less than a quad-core. They get more money for the same die, we get a chip that's faster than the duals but less-expensive than the quads.

You're right. I forgot that they were making a dual core Phenom, but would the dual core Phenom come from a single defect die quad? I can see it if it had 2 bad die, but a single. Also, dual core Phenom
isn't a quad with 2 disabled die, is it?

The issue at hand is how to market a 3 core CPU.
What is the distinct advantage of 3 core CPUs over dual cores? And with quad core prices being low, what are the advantages of getting a 3 core over a quad core?

The advantage of a 3-core chip over a dual-core one is that you can potentially get 50% more performance at the same clock speeds. A 3-core chip will also clock higher since the thermal dissipation should be lower due to one fewer core being run. The only issue I see for this is that the quad-core prices being low would result in there not being much of a price difference between the 3 and 4-core chips and people not buying many of them.

Maybe...but we haven't seen any real evidence of performance/thermal gains.
I'm sure there are advantages, but if AMD is going to push this 3 core CPU, they had better start planning on how they intend to sell them over their dual core parts, but not their quad core parts.

 

[bixplus]

humm.... care to provide proof on .... what you say... FUD? "which company got 1B for it?"

Yeah, making up data to suit his needs is pretty typical of BM. If fact, there was no fine at all...just that Intel had to "educate" its staff and "inform" its staff of the rules. I'm guessing this is why Intel didn't even contest the charge since there was no real fine involved. I'm betting that if they would have contested this, it would have ended in a stale-mate at worst. Though, since Intel chose not to fight this in Japan, they made it so much easier for AMD to sue them in the U.S., EU and Korea.

http://opensource.sys-con.com/read/48583.htm

 

[reconviperone1]

I have a hard time seeing the need for eight cores. If you do need eight cores, then you're probably talking data center, in which case, you want ECC.

I guess the way I see it is if you REALLY want multiple sockets then go to the server side.

Now, if they came out with games that would use all eight cores, then that might be different!

I understand what your saying, but just years ago people wer saying to get a fast single core instead of a dual core, build the machine and the games will come. The four graphics card thing i will never understand, but I'm not a uber gamer.

 

[BaronMatrix]

Yeah, making up data to suit his needs is pretty typical of BM. If fact, there was no fine at all...just that Intel had to "educate" its staff and "inform" its staff of the rules. I'm guessing this is why Intel didn't even contest the charge since there was no real fine involved. I'm betting that if they would have contested this, it would have ended in a stale-mate at worst. Though, since Intel chose not to fight this in Japan, they made it so much easier for AMD to sue them in the U.S., EU and Korea.

http://opensource.sys-con.com/read/48583.htm

Well, back in the 386 days, Intel decided to hold back the 386 design from AMD which violated a license agreement. AMD sued and won $1B. After reverse engineering the 386 and releasing their own.

http://money.cnn.com/magazines/fortune/fortune_archive/2006/08/21/8383598/index.htm

In 1993, McCoy began handling AMD's litigation against Intel (then over AMD's rights to clone Intel's 386 and 486 chips), and two years later he joined AMD as its general counsel. That's when he began hearing firsthand about Intel's practices, he claims.

 

[TechnologyCoordinator]

The four graphics card thing i will never understand, but I'm not a uber gamer.

LOL

Funny story:

So I bought a socket 939 board. I specifically got an SLI board. My plan was to buy one video card and upgrade to a second one later. So I bought a GeForce 7800gt. Here I am a year later and there's no way in hell I'd put another 7800gt in my system, I would just go out and get a single low-end 8800 card that would BLOW AWAY any 7800 SLI configuration.

So basically I'm just going to get decent single card systems from now on. It made sense in concept, "Oh, I can upgrade later!", but in practice, it just didn't work. But it is a nice choice for enthusiasts that love to have FPS rates higher than their monitor's refresh rates (if your monitor is only 70 hertz you're only going to see 70fps).

 

[bixplus]

Well, back in the 386 days, Intel decided to hold back the 386 design from AMD which violated a license agreement. AMD sued and won $1B. After reverse engineering the 386 and releasing their own.

http://money.cnn.com/magazines/fortune/fortune_archive/2006/08/21/8383598/index.htm

In 1993, McCoy began handling AMD's litigation against Intel (then over AMD's rights to clone Intel's 386 and 486 chips), and two years later he joined AMD as its general counsel. That's when he began hearing firsthand about Intel's practices, he claims.

Ok, now you're boring me...at least make this discussion interesting.

First, the link you provided says nothing about any settlement over the 386/486 license agreement. Besides, Intel made a calculated decision not to share and knew it would have to pay...and I don't have the figures in front of me, but I hardly believe it was 1 billion. Additionally, this was actually good for AMD because it could no longer use Intel as a crutch...of course they use IBM et al now, but not Intel so much anymore.

Secondly, the whole anti-trust issue has nothing to do whatsoever with 386/486 tussle either, so why bring it up?

For the sake of keeping this thread on topic, I'm checking out of this discission with you BM...you're going to draw your own conclusions anyway...regardless of the facts.

 

[DiverDown1]

 

[TechnologyCoordinator]

http://i160.photobucket.com/albums/t199/SirHeck1/flame.jpg

Funny, but not really accurate.

We are behaving like saints compared to the glory days. I still think the discussion is going well with a few spats here and there.

The important thing is that debate of information is driving the continuation of the thread, not name calling.

 

[DiverDown1]

The glory days...Good times, good times.
good to see your still around and giving sound advise.

 

[yomamafor1]

Well, back in the 386 days, Intel decided to hold back the 386 design from AMD which violated a license agreement. AMD sued and won $1B. After reverse engineering the 386 and releasing their own.

http://money.cnn.com/magazines/fortune/fortune_archive/2006/08/21/8383598/index.htm

In 1993, McCoy began handling AMD's litigation against Intel (then over AMD's rights to clone Intel's 386 and 486 chips), and two years later he joined AMD as its general counsel. That's when he began hearing firsthand about Intel's practices, he claims.

So it was back in 1993, and yet there was no "1 billion" mentioned anywhere on the article you just quoted, on the google, or on wikipedia.

Again, want to provide a proof to your...what you say... FUD?

And what's your thought on AMD's "a little over 30% yield" data I just provided? If you want credibility, put up some data, not some FUD.

EDIT: ok i see your point
http://www.perkinscoie.com/Experience/ExperienceDetail.aspx?exid=2014

so 10 million USD in BM's world is 1 billion. I see now. That logic makes complete sense. How can I be so dull?

 

[yomamafor1]

Funny, but not really accurate.

We are behaving like saints compared to the glory days. I still think the discussion is going well with a few spats here and there.

The important thing is that debate of information is driving the continuation of the thread, not name calling.

LOL..... you mean... Fuggar, MMM, and 9-nm?