Intel cancels plans for highest-speed Pentium 4 chip

[Simon F]

The only bad thing is that now we have to wait for programs made to be run on two processors to see a major speed increase because 1 thread will still be about the same speed, i suppose.

But that's what Folding is for

 

[Saem]

Except for playing games, everything I do requires many processes. Each of which can have many threads of their own.

 

[Fox5]

The only bad thing is that now we have to wait for programs made to be run on two processors to see a major speed increase because 1 thread will still be about the same speed, i suppose.

Very true. I've always thought that multicore CPU's were an admission of being unable to produce better single core technology. Dothan was supposedly coming along pretty well (with a low clock ceiling thought), and I was really excited to see what would come of it. Instead, both Intel and AMD have this lame idea to just stick two modern day CPU's together and call it a 6 GHz (2*3 GHz) CPU. This sort of thing won't boost performance until whole new programs come out to take advantage of multicore CPU's.

I think dothan has two disadvantages, low clock ceiling and expensive to produce.(what with all that cache)

Anyhow, by the time multicore cpus reach an affordable level I'm sure that there will be plenty of programs that make use of them...I'd still rather spend my money on an expensive cooling system and 1 really fast cpu with really fast memory though, unless the dual core cpus actually do offer close to twice the performance of 1.

 

[Saem]

THe thing with multicore is that there are synergies.

OSes when they're reporgrammed for multi-core will run much faster. And I'm not talking about more threading and processes even. But you can have lighter scheduling algorithms, in fact you can use some very simple light weight ones and have impressive performance, and reduced context switching per CPU, which means less time doing, "meta-work".

 

[Accord1999]

I think dothan has two disadvantages, low clock ceiling and expensive to produce.(what with all that cache)

The low clock ceiling hasn't really been proven, given the lack of a OCing platform for the Pentium-M. And the Dothan is probably the cheapest processor for Intel to manufacture, since its less than 90mm^2 and much of it is cache, which is relatively easy to manufacture due to simplicity and redunancy.

 

[Fox5]

I think dothan has two disadvantages, low clock ceiling and expensive to produce.(what with all that cache)

The low clock ceiling hasn't really been proven, given the lack of a OCing platform for the Pentium-M. And the Dothan is probably the cheapest processor for Intel to manufacture, since its less than 90mm^2 and much of it is cache, which is relatively easy to manufacture due to simplicity and redunancy.

I heard there are server motherboards that support dothans, no one has tried one of those?

BTW, are you sure dothan is that cheap with 2MB of L2 cache? A 1.5ghz dothan goes for $216 online, while a mobile 64 2800+ is $165.($197 for the 1MB cache version) A 3000+ is $178, or $236 with 1MB L2 cache. I believe a 1.5ghz dothan and a 2800+ about compete in performance, but $165 versus $216 is a bit more than the normal difference for an amd versus intel chip. Still not directly comparable though, as I guess intel could make a desktop centrino that wasn't low power, and pentium ms are probably harder to come by than mobile 64s. I'd expect to see 1MB cache if desktop pentium Ms if they came out now, since presumebly they could up the mhz to make up for the loss of cache.

 

[Killer-Kris]

I heard there are server motherboards that support dothans, no one has tried one of those?

Only problem is that server boards tend to not have any form of overclocking available on them either.

BTW, are you sure dothan is that cheap with 2MB of L2 cache? A 1.5ghz dothan goes for $216 online, while a mobile 64 2800+ is $165.($197 for the 1MB cache version) A 3000+ is $178, or $236 with 1MB L2 cache. I believe a 1.5ghz dothan and a 2800+ about compete in performance, but $165 versus $216 is a bit more than the normal difference for an amd versus intel chip.

The prices just an example of supply and demand, along side Intels ability to charge just about what ever it wants for its (now pretty much only mobile) chips.

But yeah long story short, cache is extremely cheap. Not only is it fairly small and regular, but that small size allows for a great deal of redundancy with out increasing the size of the chip. This lets Intel sell pretty much all Dothans as P-Ms, and in the extremely rare case they'll have to disable a significant portion of the cache and you get a Celeron-M. That's something that's not quite as easy to do with logic.

 

[T2k]

Well, at least now we know how long AMD has before the gauntlet gets thrown. They have until about 2006 to get their act together, or it's just another AthlonXP replay where AMD fritters away all the its market share gains and loses money for 9 quarters.

They have a year to do better than they've ever done before, even better than the lead with the Thunderbird vs. P3.

I really don't like their odds .

Or versus the early p4s...

But thunderbirds were really hot, whereas the athlon 64s are much cooler than the prescotts and the g5s, which makes them almost the ideal oem solution. However, I think amd's future basically depended on the timely arrival of 64 bit windows and applications to go with it, and since that hasn't happened and probably won't until intel has something 64 bit ready...well, maybe ibm will decide they want a piece of the x86 pie and buy amd or something.

Well it's a double catch: IBM already working together with AMD on the CPUs but at the same time it's a very well-rouned cooperation: IBM wants to see whatever AMD knows and compete with that from the PPC side...

Keep in mind: IBM is much bigger than AMD and yet couldn't compete with Intel as AMD does...

IBM emphasized the traits that people hate about intel much better than intel did. Intel was the underdog when they took the market from IBM, but when intel took the market it was much smaller and probably more tech savvy on the whole.(after all, who needed computers 20 years ago?)

Edit: Amd's share in the market now is probably larger than the entire market was when IBM was still in it.

IBM is MUCH-MUCH bigger company than AMD.

 

[T2k]

THe thing with multicore is that there are synergies.

OSes when they're reporgrammed for multi-core will run much faster. And I'm not talking about more threading and processes even. But you can have lighter scheduling algorithms, in fact you can use some very simple light weight ones and have impressive performance, and reduced context switching per CPU, which means less time doing, "meta-work".

Exactly. Eventually the multicore is an evolutional step - it's actually well-known in server or mainframe world.

 

[T2k]

I think dothan has two disadvantages, low clock ceiling and expensive to produce.(what with all that cache)

The low clock ceiling hasn't really been proven, given the lack of a OCing platform for the Pentium-M. And the Dothan is probably the cheapest processor for Intel to manufacture, since its less than 90mm^2 and much of it is cache, which is relatively easy to manufacture due to simplicity and redunancy.

Large cache always makes any CPU much more expensive - and Intel is getting famous about its lousy design plus their profit mrgin shold be much higher in order to feed this beast-sized something, slow mammoth.

Intel has a lot to do within months, I'm telling you.

BTW I already said on this board around Spring and I reiterate it: AMD has told me as early as this March they'll release a multicore Opteron which will be pin compatible with my then-new workstations.

So far it seems they were right, their confidence was based on something... .

 

[T2k]

I heard there are server motherboards that support dothans, no one has tried one of those?

Only problem is that server boards tend to not have any form of overclocking available on them either.

BTW, are you sure dothan is that cheap with 2MB of L2 cache? A 1.5ghz dothan goes for $216 online, while a mobile 64 2800+ is $165.($197 for the 1MB cache version) A 3000+ is $178, or $236 with 1MB L2 cache. I believe a 1.5ghz dothan and a 2800+ about compete in performance, but $165 versus $216 is a bit more than the normal difference for an amd versus intel chip.

The prices just an example of supply and demand, along side Intels ability to charge just about what ever it wants for its (now pretty much only mobile) chips.

Not really. I see more and more AMD mobile in around me: when it comes to desktop replacement, Intel is nowhere near to AMD's Mobile A64-based laptops like the very best eMachines M6811 or the HP ones.

In fact I see more and more ppl leave Dell and go to HP - and end up with AMD-config.

But yeah long story short, cache is extremely cheap. Not only is it fairly small and regular, but that small size allows for a great deal of redundancy with out increasing the size of the chip. This lets Intel sell pretty much all Dothans as P-Ms, and in the extremely rare case they'll have to disable a significant portion of the cache and you get a Celeron-M. That's something that's not quite as easy to do with logic.

This is one side - on the other hand cache is EXPENSIVE. Why? Because of it HUUUUGE size. That kicks up the wafer costs (less core from the same sized 'cookie').

OFF
As I said, I believe Intel can't do too much until doesn't have anything against AMD's integrated memory controller - that's just simple amazingly fast. That was AMD's best step ever.

 

[epicstruggle]

Large cache always makes any CPU much more expensive - and Intel is getting famous about its lousy design plus their profit mrgin shold be much higher in order to feed this beast-sized something, slow mammoth.

Ive heard this before, but why is that?

epic

 

[3dilettante]

Larger cache with no other design changes equates to a larger die size. Since chips are fabricated from a wafer of a fixed size, fewer larger chips can be manufactured from a wafer. Larger chips also tend to be more likely to accumulate faults, though cache is highly redundant for this reason.

This doesn't always translate into higher costs, though it is a factor. If the higher cache makes the chip competitive, it is often better to sell all of a run of larger chips than to sell 3/4 of a run of smaller ones. Unsold inventory costs and unused fab capacity are a constant drain on the budget, so it isn't always a bad idea to increase die size.

 

[T2k]

Large cache always makes any CPU much more expensive - and Intel is getting famous about its lousy design plus their profit mrgin shold be much higher in order to feed this beast-sized something, slow mammoth.

Ive heard this before, but why is that?

epic

Look above: they eat up large space of the die - in case of Dothan (2MB cache like P4EE aka average Xeon) it's bigger than half the die size:

OFF
This is what Intel will offer for the market?

It's gotta be a joke...

 

[3dilettante]

Half of Dothan's die is cache, but with the 90nm feature size, it is only slightly larger than the original P-M.

Right now with the way leakage current is going, a large cache is what balances out performance gains vs. heat output.

Pretty much every processor out there has 1/2 or more of its die dedicated to cache.
Dothan just happens to cram it in very well.

 

[T2k]

Larger cache with no other design changes equates to a larger die size. Since chips are fabricated from a wafer of a fixed size, fewer larger chips can be manufactured from a wafer. Larger chips also tend to be more likely to accumulate faults, though cache is highly redundant for this reason.

This doesn't always translate into higher costs, though it is a factor. If the higher cache makes the chip competitive, it is often better to sell all of a run of larger chips than to sell 3/4 of a run of smaller ones. Unsold inventory costs and unused fab capacity are a constant drain on the budget, so it isn't always a bad idea to increase die size.

The only reason behind Intel's crazy chache size is they HAVE to show something against AMD's integrated memory controllers - AMDs have EXTREMELY low latency.
Currently Intel's only bet to put as much memory as possible right next to the core to avoid their outdated memory subsystem.

 

[T2k]

Half of Dothan's die is cache, but with the 90nm feature size, it is only slightly larger than the original P-M.

Right now with the way leakage current is going, a large cache is what balances out performance gains vs. heat output.

Pretty much every processor out there has 1/2 or more of its die dedicated to cache.
Dothan just happens to cram it in very well.

That's bigger than half, sorry.

 

[3dilettante]

Larger cache with no other design changes equates to a larger die size. Since chips are fabricated from a wafer of a fixed size, fewer larger chips can be manufactured from a wafer. Larger chips also tend to be more likely to accumulate faults, though cache is highly redundant for this reason.

This doesn't always translate into higher costs, though it is a factor. If the higher cache makes the chip competitive, it is often better to sell all of a run of larger chips than to sell 3/4 of a run of smaller ones. Unsold inventory costs and unused fab capacity are a constant drain on the budget, so it isn't always a bad idea to increase die size.

The only reason behind Intel's crazy chache size is they HAVE to show something against AMD's integrated memory controllers - they have EXTREMELY low latency.
Currently Intel's only bet to put as much memory as possible right next to the core to avoid their outdated memory subsystem.

I agree for the P4 EEs that cache is being thrown in to combat the on-die memory controller. Though technically they aren't throwing cache onto a desktop P4, just rebranding a Xeon.

For Dothan, I'm not so sure. At the time of its design I don't think it was meant to compete directly with a desktop A64 or FX. The successor to Dothan might be more informative.

 

[Fox5]

Well, at least now we know how long AMD has before the gauntlet gets thrown. They have until about 2006 to get their act together, or it's just another AthlonXP replay where AMD fritters away all the its market share gains and loses money for 9 quarters.

They have a year to do better than they've ever done before, even better than the lead with the Thunderbird vs. P3.

I really don't like their odds .

Or versus the early p4s...

But thunderbirds were really hot, whereas the athlon 64s are much cooler than the prescotts and the g5s, which makes them almost the ideal oem solution. However, I think amd's future basically depended on the timely arrival of 64 bit windows and applications to go with it, and since that hasn't happened and probably won't until intel has something 64 bit ready...well, maybe ibm will decide they want a piece of the x86 pie and buy amd or something.

Well it's a double catch: IBM already working together with AMD on the CPUs but at the same time it's a very well-rouned cooperation: IBM wants to see whatever AMD knows and compete with that from the PPC side...

Keep in mind: IBM is much bigger than AMD and yet couldn't compete with Intel as AMD does...

IBM emphasized the traits that people hate about intel much better than intel did. Intel was the underdog when they took the market from IBM, but when intel took the market it was much smaller and probably more tech savvy on the whole.(after all, who needed computers 20 years ago?)

Edit: Amd's share in the market now is probably larger than the entire market was when IBM was still in it.

IBM is MUCH-MUCH bigger company than AMD.

Yeah, but when IBM was in the x86 market the market was much smaller.

But yeah long story short, cache is extremely cheap. Not only is it fairly small and regular, but that small size allows for a great deal of redundancy with out increasing the size of the chip. This lets Intel sell pretty much all Dothans as P-Ms, and in the extremely rare case they'll have to disable a significant portion of the cache and you get a Celeron-M. That's something that's not quite as easy to do with logic.

Cache seems to give a lot of performance though, especially for cpus like the p4s(extreme editions), so if it is cheap why not put 2MB L2 cache on every cpu?(or at least have cheaper extreme editions, which use L3 cache anyhow)

For Dothan, I'm not so sure. At the time of its design I don't think it was meant to compete directly with a desktop A64 or FX. The successor to Dothan might be more informative.

I think it competes with the mobile athlon 64s, especially the 35w mobiles.

 

[T2k]

Cache isn't cheap at all, that's completely wrong.
He's confusing the design complexity related costs - which are already paid: your engineer are FT not extra personnel - with manufacturing cost.

The latter really matters, believe me: imagine when you introduce a chip with 2MB cahce but you also have a 1MB version. Whay you're going to do?
A) keep manufacturing two kind of chips, in order to maintain higher per-wafer output in case of the 1MB version?
B) make only one 2MB version and disable half of it for the 1MB plus use some of the nonworking 2 megs as working 1 meg?

In case A) you get much higher costs AFAIK, so you'll make on chip - but that's still more expensive than make 1MB chip ONLY.