What Should AMD Do Next? (Real)

Yeti · Jul 13, 2002, 09:27 AM

Quite frankly, the Thoroughbred was a tad dissapointing for me. It seems that while AMD's Athlon design is insanely powerful, it has some weaknesses. So which of the above would put AMD into a strong lead over Intel?

I did not include the Opteron because that is still quite a bit away, i would rather push AMD to make some changes in their barton core. And i am not looking for the opinions of those who preach "AMD is dead" from an altar with an intel logo on it

Zardon · Jul 13, 2002, 09:31 AM

I think they should lower their processes per clock cycle intensely like Intel and ramp up the MHZ to like 5 gig. it looks good for grandmothers when they are out buying a computer.

Yeti · Jul 13, 2002, 09:33 AM

Posted by Zardon, in the thread i screwed up (sry...):

Quote:

I think they should lower their processes per clock cycle intensely like Intel and ramp up the MHZ to like 5 gig. it looks good for grandmothers when they are out buying a computer.

I would have to disagree... the problem is not in the IPC. Intel uses a lot more real estate for its CPU's than AMD... if you've read the recent THG article... there are approximately 2.5-3 times as many transistors per area on the AMD than on the Intel. This lets AMD make cheaper processors but also lets Intel ramp theirs up higher(better heat dissipation and stability)..

Frodo301 · Jul 13, 2002, 09:47 AM

I think jacking it up to 200 FSB is a good choice. On the other hand a larger core to disperse the heat would be nice also.

DallasStar · Jul 13, 2002, 10:28 AM

People always underestimate the difference more cache can make on a processor.... Just look at what happened with the p4 with northwood... 256 more cache and whoa! you now have the fastest processor on the market. If the same happened to barton, just think of the possibilities

crazy overclock · Jul 13, 2002, 08:07 PM

Quote:

Originally posted by dallasstar
People always underestimate the difference more cache can make on a processor.... Just look at what happened with the p4 with northwood... 256 more cache and whoa! you now have the fastest processor on the market. If the same happened to barton, just think of the possibilities

That's true, but who's gonna buy a barton when hammer will be likely to be launched this fall

JavaFox · Jul 13, 2002, 11:30 PM

What should AMD do? From a business standpoint, making the core bigger is about the stupidest thing anyone has ever said. Smaller cores are cheaper to produce. Did you know that AMD can sell any of their Athlons for sub-$100 prices and still turn a profit? Considering AMD is having trouble keeping up with earning estimates (like every other tech company), why would they increase the core size, slow production, and raise costs?

If AMD was indeed planning on continuing with the Athlon as their only means of making money, they should go with an even smaller fabrication process and then triple the cache. See, the thing about Athlon is, they're so small that they approach a point where they almost have to make the core bigger than it needs to be, just so they can connect it electrically! This would be like a computer manufacter making a PC so small, that he actually had to make it a little bigger, just so the power plug would fit! That said, AMD could make a chip so small, and then beef up the size a little with a considerable cache.

"What AMD will do next?" is a totally different question, however.

They will release Hammer, and they will add cache to Barton. I also suspect that Barton will also be optimized in other ways. I find it difficult to believe double the cache is going to justify 600 more performance rating points.

And, basically, that's it for now.

JavaFox · Jul 13, 2002, 11:30 PM

Quote:

Originally posted by crazy overclock
That's true, but who's gonna buy a barton when hammer will be likely to be launched this fall

You probably won't see them til early 2K3, but the answer to your question is: people not interested in upgrading their motherboard.

Silentst2000 · Jul 14, 2002, 05:26 AM

Quote:

Originally posted by JavaFox
What should AMD do? From a business standpoint, making the core bigger is about the stupidest thing anyone has ever said. Smaller cores are cheaper to produce. Did you know that AMD can sell any of their Athlons for sub-$100 prices and still turn a profit? Considering AMD is having trouble keeping up with earning estimates (like every other tech company), why would they increase the core size, slow production, and raise costs?

If AMD was indeed planning on continuing with the Athlon as their only means of making money, they should go with an even smaller fabrication process and then triple the cache. See, the thing about Athlon is, they're so small that they approach a point where they almost have to make the core bigger than it needs to be, just so they can connect it electrically! This would be like a computer manufacter making a PC so small, that he actually had to make it a little bigger, just so the power plug would fit! That said, AMD could make a chip so small, and then beef up the size a little with a considerable cache.

"What AMD will do next?" is a totally different question, however.

They will release Hammer, and they will add cache to Barton. I also suspect that Barton will also be optimized in other ways. I find it difficult to believe double the cache is going to justify 600 more performance rating points.

And, basically, that's it for now.

well by not only increasing the cache on the barton they could also be increasing the data path width from 64 bit to maybe 128 or 256 (like intel) that could give a huge boost imo.

JavaFox · Jul 14, 2002, 05:33 AM

Quote:

Originally posted by Silentst2000
well by not only increasing the cache on the barton they could also be increasing the data path width from 64 bit to maybe 128 or 256 (like intel) that could give a huge boost imo.

What do you base this assumption on? The fact that the Athlon is faster than most P3s, or that fact that the Athlon is faster than most P4s? They both have larger cache pipelines.

JavaFox · Jul 14, 2002, 05:36 AM

Oh my frickin' Lord, I cannot believe three people voted to "space transistors out for better heat dissipation."

One, the P4 does not have "spaced out transistors" -- it has MORE transistors, and that's what makes it larger. They are not spaced out, and anyone that thinks that only need compare the transistor count with that of the Athlon to see how dumb that assertion is.

Secondly, WHY should AMD do this? To decrease production? To increase costs? 'Cause a larger die will do both of those things.

Yeti · Jul 14, 2002, 07:16 AM

http://www6.tomshardware.com/cpu/02q...ghbred-07.html

Quote:

Oh my frickin' Lord, I cannot believe three people voted to "space transistors out for better heat dissipation."

Hmmm
Read the above page, you will find out that the Northwood P4 has roughly the same amount of transistors on a much larger area.... the ration i would say is roughly 4:3 in comparison with the athlon.

Sure this raises production costs... but it will also let the CPU scale better and gives more surface area for heat dissipation. So more cost... but better performance... which do you care more about?

I'm serious.

JavaFox · Jul 14, 2002, 08:29 AM

Let's see... so you propose that we take a Thoroughbred, and space its transistors out?

Well, AMD already made that. It's called the Palomino.

The P4 Northwood, at 55 million transistors and a die-size of 146 mm^2, has some 380,000 transistors per square millimeter. The Palomino, in comparison, has barely 300,000.

So there you have it, Yeti. There's a chip with "spread out transistors."

Spread out transistors don't make the P4 scale well. Its architecture lets it hit high clock speeds. The reason the Thoroughbred can't scale well is not heat -- if that were true, we'd see good o/cs out of watercooling folk-- it's youth. It's not mature enough yet.

Yeti · Jul 14, 2002, 09:40 AM

Quote:

Let's see... so you propose that we take a Thoroughbred, and space its transistors out?

Well, AMD already made that. It's called the Palomino.

The P4 Northwood, at 55 million transistors and a die-size of 146 mm^2, has some 380,000 transistors per square millimeter. The Palomino, in comparison, has barely 300,000.

So there you have it, Yeti. There's a chip with "spread out transistors."

Spread out transistors don't make the P4 scale well. Its architecture lets it hit high clock speeds. The reason the Thoroughbred can't scale well is not heat -- if that were true, we'd see good o/cs out of watercooling folk-- it's youth. It's not mature enough yet.

No... i think you missed my point. There's a very interesting reason regarding why we can't microminiaturize parts. Its a matter of quality. If NatSemi, NEC, Intel, AMD or any other chipmaker were to start using .13 technology and machines (processes capable of producing .13 parts) to build .18 parts the yields would be extremely good the parts would overclock well. Of course the Pentium IV was designed to scale well... i'm not contradicting that. I am saying, however, that a contributing factor to the .13 PIV's scalability is the fact that the transistors on it aren't as dense as those on the AMD chip... as you will see from the link i posted before.

It is perfectly reasonable to think that if AMD were to use higher quality machinery to produce the more-than-slightly successful Palomino core, or an enlarged variant of the Thoroughbred core... it would provide for a better processor in general. The switch back to a larger core would of course mean having to increase the voltage again. But as it has been throughly explored by various hardware sites, the Palomino core had a much easier time dissipating heat than the Thoroughbred (due to its larger surface area).

It would seem then that if a switch were to be made back to the larger size, this would provide for more stability (and scalability), while keeping heat output within reason and helping heat dissipation in general.

See what i mean?

Jul 13, 2002, 09:27 AM	#1
Yeti DriverHeaven Senior Member Join Date: May 2002 Posts: 662 Rep Power: 0	What Should AMD Do Next? (Real) Quite frankly, the Thoroughbred was a tad dissapointing for me. It seems that while AMD's Athlon design is insanely powerful, it has some weaknesses. So which of the above would put AMD into a strong lead over Intel? I did not include the Opteron because that is still quite a bit away, i would rather push AMD to make some changes in their barton core. And i am not looking for the opinions of those who preach "AMD is dead" from an altar with an intel logo on it __________________ --Oh lord not again

Jul 13, 2002, 09:47 AM	#4
Frodo301 Outraged Join Date: May 2002 Location: The mountains Posts: 584 Rep Power: 0	I think jacking it up to 200 FSB is a good choice. On the other hand a larger core to disperse the heat would be nice also. __________________ The greatest thing you'll every learn is to love, and be loved in return.

Jul 13, 2002, 10:28 AM	#5
DallasStar Junior Join Date: May 2002 Location: Wave Existence Posts: 2,065 Rep Power: 0	People always underestimate the difference more cache can make on a processor.... Just look at what happened with the p4 with northwood... 256 more cache and whoa! you now have the fastest processor on the market. If the same happened to barton, just think of the possibilities __________________ "A picture of my existence... would show a useless wooden stake covered in snow... stuck loosely at a slant in the ground in a ploughed field on the edge of a vast open plain on a dark winter night." --- Franz Kafka

Jul 13, 2002, 09:31 AM	#2
Zardon DriverHeaven Founder Join Date: May 2002 Posts: 32,480 Rep Power: 179	I think they should lower their processes per clock cycle intensely like Intel and ramp up the MHZ to like 5 gig. it looks good for grandmothers when they are out buying a computer.

Jul 13, 2002, 11:30 PM	#7
JavaFox E Pluribus Unum Join Date: May 2002 Posts: 3,203 Rep Power: 0	What should AMD do? From a business standpoint, making the core bigger is about the stupidest thing anyone has ever said. Smaller cores are cheaper to produce. Did you know that AMD can sell any of their Athlons for sub-$100 prices and still turn a profit? Considering AMD is having trouble keeping up with earning estimates (like every other tech company), why would they increase the core size, slow production, and raise costs? If AMD was indeed planning on continuing with the Athlon as their only means of making money, they should go with an even smaller fabrication process and then triple the cache. See, the thing about Athlon is, they're so small that they approach a point where they almost have to make the core bigger than it needs to be, just so they can connect it electrically! This would be like a computer manufacter making a PC so small, that he actually had to make it a little bigger, just so the power plug would fit! That said, AMD could make a chip so small, and then beef up the size a little with a considerable cache. "What AMD will do next?" is a totally different question, however. They will release Hammer, and they will add cache to Barton. I also suspect that Barton will also be optimized in other ways. I find it difficult to believe double the cache is going to justify 600 more performance rating points. And, basically, that's it for now.

Jul 14, 2002, 05:36 AM	#11
JavaFox E Pluribus Unum Join Date: May 2002 Posts: 3,203 Rep Power: 0	Oh my frickin' Lord, I cannot believe three people voted to "space transistors out for better heat dissipation." One, the P4 does not have "spaced out transistors" -- it has MORE transistors, and that's what makes it larger. They are not spaced out, and anyone that thinks that only need compare the transistor count with that of the Athlon to see how dumb that assertion is. Secondly, WHY should AMD do this? To decrease production? To increase costs? 'Cause a larger die will do both of those things.

Jul 14, 2002, 08:29 AM	#13
JavaFox E Pluribus Unum Join Date: May 2002 Posts: 3,203 Rep Power: 0	Let's see... so you propose that we take a Thoroughbred, and space its transistors out? Well, AMD already made that. It's called the Palomino. The P4 Northwood, at 55 million transistors and a die-size of 146 mm^2, has some 380,000 transistors per square millimeter. The Palomino, in comparison, has barely 300,000. So there you have it, Yeti. There's a chip with "spread out transistors." Spread out transistors don't make the P4 scale well. Its architecture lets it hit high clock speeds. The reason the Thoroughbred can't scale well is not heat -- if that were true, we'd see good o/cs out of watercooling folk-- it's youth. It's not mature enough yet.