Why doesn't Intel use on-die controllers?

[TheRyuu]

Why does Intel still rely on the old FSB? Can't they use a hypertransport bus with their cpu's? Or even move to an on-die memory controller?

All those technologies seem so much better then their current ones yet Intel doesn't impliment them, why not? I mean, if those Netburst CPU's are so bandwith hungry, WHY NOT GIVE THEM BANDWITH!?!?! Use the fast Hypertransport bus which would eleminate a bottleneck.

And what I really don't get, is NONE of these technologies are on Intels road-map for the future. Apparently in 2007 they will be supplying the Quad-Core Xeons or any other Quad core chip with a dual FSB. Don't really think that will be enough bandwith.

Why Intel, why!?!?!? Can you see the reasoning behind this?

 

[formulav8]

Arrogance is probably the #1 reason.

I also believe Intel did try the ondie controller and failed with Tejas? I do not remember the details anymore.

Arrogance is likely the most likely reason. Be that as it may, they are looked to as the leader in their industry and therefore, to keep their image, should NOT follow the other guy if at all possible. Doing that, would not keep your leadership image which is a HUGE marketing point.

Others will have more insight than me though.



Jason

 

[BrownTown]

Intel does have CSI and integrated memmory controllers on its roadmaps, they are just a couple of years away. They did have an integrated memmory controller on one chip awhile back, but it ended up failing because thememmory it supported ended up sucking. So Intel was probably anxious to try again with DDR in case some other memmory standard ended up being better. Intels NGMA is supposed to have better prefeteching in order to help alliviate the problem of not having an integrated memmroy contorller, but we have yet to see how that works. Also of note is that Intel is the one selling the northbridge, so if they integrate it into the CPU they are losing a big source of income.

 

[dmens]

LOL, arrogance. How do you come up with such bullshit and post it with seriousness?

The real reason is because the decoupling of the memory interface from the core saves design effort and allows a more flexible approach at the cost of latency. Oh, and the fact that an on-die memory controller is a non-factor (maybe even a liability) for the mobile and mainstream markets.

 

[Keysplayr]

Originally posted by: formulav8
Arrogance is probably the #1 reason.

I also believe Intel did try the ondie controller and failed with Tejas? I do not remember the details anymore.

Arrogance is likely the most likely reason. Be that as it may, they are looked to as the leader in their industry and therefore, to keep their image, should NOT follow the other guy if at all possible. Doing that, would not keep your leadership image which is a HUGE marketing point.

Others will have more insight than me though.



Jason

W......T.......F.......????

 

[TuxDave]

Originally posted by: formulav8

Others will have more insight than me though.

Jason

I sure hope so.

 

[openwheelformula1]

$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$

enough reasons? Here are more:

$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$

 

[BrownTown]

an integrated memmory controller isn't a universally good thing. Money is probably a big reason why Intel keeps the northbridge, but there are others. Having an integrated memmory controller means that you are pretty much tied to a specific type of memmory, If new innovations come out the core has to be redesigned instead of jsut the northbridge. So, it could take longer to adapt to new memmory technologies, and it would likely be mroe expensive since you will have to put the whole core through the validation process instead of just the northbridge. Also, you cannot take an old processor and put it in a new motherboard and have it work with new memmory. So, when new memmory tech comes out you have to upgrade the motherboard, memmroy and CPU all at the same time.

 

[formulav8]

Should have known the Intel cry babies would start up from what I said.

I simply stated my opinion. IF you don't like it you don't need to whine about it. How about stating your opinion on why Intel doesn't do what the subject of this thread is? Thats what I did.

Anyways, when I said arrogance, I actually meant pride, just didn't come out the way I meant it. I was thinking about them in the manner that for example, why they aren't implementing hypertransport. Why don't they use it? Maybe because they have to make their own or have to swallow some pride. My opinion, sorry if you don't like it.

Some people act like you've insulted their mother when you say something bad about amd/intel/nvidia, ect... I can't understand why anyone would be so attached to a company.



Jason

 

[formulav8]

Originally posted by: TuxDave

Originally posted by: formulav8

Others will have more insight than me though.

Jason

I sure hope so.

[/quote]



Apparently YOU don't.


Jason

 

[dmens]

I don't give a crap when people insult intel, or any other entity. I just laugh at puerile fanboi opinions (e.g. yours). Of course, you are entitled to hold such opinions, just don't get pissy when others mock them, LOL.

 

[themusgrat]

I kind of agree with Jason. It would look like they have screwed up big time to come out with a copy cat version of hypertransport.

 

[BrownTown]

they own the rights to use HyperTransport, so they wouldn't have to copy it, they could put the exact same thing in their CPUs if they wanted.

 

[DrMrLordX]

Pride? Arrogance? Eh, maybe. It's more likely a marketting decision to avoid using on-die memory controllers. Your guess is as good as mine as to why they have avoided Hypertransport all these years, though in reality, it could take anywhere from 6 months to a year for Intel to convert all their platforms from a FSB system to a Hypertransport system(if not longer?). That's time they probably don't want to spend retooling all their platforms to take advantage of a technology that AMD has already implemented. My guess is they want to skip on to something they feel is better than Hypertransport.

Another reason why they've likely avoided on-die memory controllers is(or was) BTX.

 

[openwheelformula1]

I remember reading somewhere that Intel would have to pay $15/cpu to implement hypertrasport, is that correct?

 

[TuxDave]

Originally posted by: formulav8

Originally posted by: TuxDave

Originally posted by: formulav8
Others will have more insight than me though.

Jason

I sure hope so.

Apparently YOU don't.
Jason

I could say nothing and still sound more educated than you. My god "Arrogance"? What kind of decision process is that? It would almost impossible to use the "AMD came up with it first so we better abandon all ideas of implementing it because it would hurt our pride" argument. That's terrible. Anyways, read one post up. It's on its way. So much for your 'arrogance' argument.

 

[BrownTown]

Originally posted by: openwheelformula1
I remember reading somewhere that Intel would have to pay $15/cpu to implement hypertrasport, is that correct?

I think I might have heard that too, but no clue if its true.

Don't really matter though, theres no way Intel will use Hypertransport, they are developing their own technology to accomplish the same task.

 

[AnandThenMan]

In some form or another, arrogance DOES factor into it. Don't forget, Intel flat out was telling everyone that no one needs 64 bit on the desktop. Of course, in reality, they were telling people this because they didn't want to adobt AMD64 tech, and were hoping and praying that Itanium would become the defacto architecture.

Maybe arrogance is not really the right word, perhaps stupidity is more appropriate. Intel has refused to let go of the ancient interconnect technology, and it is really hurting them where it counts most, in the server market. You think AMD has the advantage when it comes to scaling, wait until we start seeing quad core. Without an onboard memory controller and a hypertransport like system, the tired old FSB mentality will kick Intel in the crotch even harder.

 

[hans007]

Originally posted by: DrMrLordX
Pride? Arrogance? Eh, maybe. It's more likely a marketting decision to avoid using on-die memory controllers. Your guess is as good as mine as to why they have avoided Hypertransport all these years, though in reality, it could take anywhere from 6 months to a year for Intel to convert all their platforms from a FSB system to a Hypertransport system(if not longer?). That's time they probably don't want to spend retooling all their platforms to take advantage of a technology that AMD has already implemented. My guess is they want to skip on to something they feel is better than Hypertransport.

Another reason why they've likely avoided on-die memory controllers is(or was) BTX.

i dont think on die memory controllers has anything to do with btx personally.

first off every intel chip for about the last 3-4 years has been running on basically the same bus. it is not super easy to just switch to a new one and get all the validation and such intel has to have.

an ondie memory controller was on the tejas chip (which eventually a lot of the technology in became the pentium M sans controller). it used rambus though and that did not pan out.

the ondie memory controller thing intel really is just making up for by using larger caches right now. in some ways a non on-die controller is an advantage, as intel can always take advantage of new memory technologies much more quickly and if cache can make up the difference then it is not TOO bad. intel probably will be able to take advantage of ddr3 and fb-dimm much sooner than amd for example. they also dont have to make 2 seperate chips just to support regisreted and non registered memory.


the bus issue is a bad one, but amd did invent hypertransport and intel will eventually go in that direction too. just give them a couple more years. CSI might actually be better than hypertransport who knows and intel's future roadmap isnt awful even with a normal fsb and non integrated controllers.

 

[abomb]

Alright, I?m going to try to field this question the best I can. Once upon a time Intel had a processor called Tejas, it had an integrated memory controller that integrated with RAMBUS ram. Bad idea, RAMBUS went the way of the dodo and Intel got burned. I was actually told that even back in the days of the 386 Intel had dabbled in integrated memory controllers. So just to let everyone know, Intel has been there done that. The reason Intel continues to stick with the FSB is obvious, so there cannot be a repeat of the RAMBUS and I would say that is a smart choice. AMD has really lucked out that a new type of memory has not taken hold, although they are now (finally) going to support DDR2, about a year or so after Intel first supported it. As its been covered above integrated memory controllers can be a very bad thing, if memory changes the controller has to change, I don?t need to get into that further its obvious. As for hypertransport I really hate you burst everyones bubbles, but to my knowledge it has not changed since its inception on the T-Bred Athlon? (Feel free to correct me if I?m wrong) As it is FSB technology will catch up to any benefit that hypertransport has minus a little latency very soon, thanks to DDR2.
With that covered let us summarize for the uniformed AMD zealots out there. AMD?s integrated memory controller was a gamble and it worked. Hypertransport has not changed since its inception. FSB continues to evolve pretty consistently. Intel has made 65nm chips, and is currently working on the 35nm manufacturing process. AMD has yet to even manufacture 65nm chips.

The next point I will make is Netburst. Now to everyone who says it is ?inferior? to AMD?s architecture. I hate to rain on your parade but your wrong, unless of course you provide a scope to your criticism. The Netburst pipeline is designed for a different kind of programming then the AMD (?short pipeline?) processors. Today?s programs are much more efficient on the ?short pipe? processors, but there may come a time when the longer pipeline will be an advantage and Netburst might show its ugly face again. If someone would like to explain exactly how that works they can go ahead.

Also I am not an Intel fan boy. In fact I own both brands of processors. In the past I had owned an Athlon 2400+ and I absolutely loved that box, which sadly recently died. In fact come to think of it as I upgrade I alternate company by some happenstance, so I am really not partial to either company.
I would like to also point out I do not want AMD or Intel to go out of business. I do like seeing the market share even out, that?s a good thing. This shift of market share requires that Intel be competitive and that requires them to innovate. Look at AMD for example, they haven?t done anything since 2003. Intel is even worse, and AMD capitalized by not innovating and just churning out the same old thing to make profit, a possibly cause why they have not switch to 65nm a possible cause why they wont support DDR2 until their next processor is near ready? I would hazard a guess to either they have trouble with these technologies or that?s the answer.

The last thing that troubles me is the extreme anti-intel attitude that certain technology review/preview services have taken. Misrepresentation of facts, and a lot of journalistic license have really started to irk me. That and the simple irrelevant comparisons that I see benchmark wise and the conclusions people draw from them. One recently brought to my attention was a 2.0 ghz P-M vs. 2.0 ghz Turion. All seems well correct? Nay my friend, The Turion laptop was running DDR 400 (very nice DDR400 if I may add), and the P-M laptop was running DDR2@533. Now I think most of you will know that DDR 400 is significantly faster then DDR2. So how can this be a ?comparison? of the two processors. I would say it would be a good comparison of the two notebooks, but not of the performance of the processors. Feel free to correct me if I?m wrong by saying the DDR is significantly faster then the DDR2. Unless I?m mistaken and I could be DDR2@800 should be pretty equivalent to DDR400.
Thank you and I hope you appreciated this small tome of mine, I did enjoy writing it as well.


I was silly to not to comment on something significant. AMD has really done one thing to note significantly better then Intel and they launched a mainstream consumer grade 64 bit processor. I do give them that, they accomplished that very well. They are by no means the first, I believe either Motorola or IBM hold that crown. But that is one accomplishment I give to them, while Intel only slapped on their 64bit support rather lackadaisically only because AMD did first.

 

[Rock Hydra]

Originally posted by: formulav8
Arrogance is probably the #1 reason.

I also believe Intel did try the ondie controller and failed with Tejas? I do not remember the details anymore.

Arrogance is likely the most likely reason. Be that as it may, they are looked to as the leader in their industry and therefore, to keep their image, should NOT follow the other guy if at all possible. Doing that, would not keep your leadership image which is a HUGE marketing point.

Others will have more insight than me though.

Jason

I don't know about arrogance, but I do remember seeing prototype pics of a chip with a northbridge or something built into the package about 1/4" away from the die itself.

 

[pm]

The i386SL had an integrated memory controller. It was released back around 1990.

 

[hans007]

Originally posted by: abomb
Alright, I?m going to try to field this question the best I can. Once upon a time Intel had a processor called Tejas, it had an integrated memory controller that integrated with RAMBUS ram. Bad idea, RAMBUS went the way of the dodo and Intel got burned. I was actually told that even back in the days of the 386 Intel had dabbled in integrated memory controllers. So just to let everyone know, Intel has been there done that. The reason Intel continues to stick with the FSB is obvious, so there cannot be a repeat of the RAMBUS and I would say that is a smart choice. AMD has really lucked out that a new type of memory has not taken hold, although they are now (finally) going to support DDR2, about a year or so after Intel first supported it. As its been covered above integrated memory controllers can be a very bad thing, if memory changes the controller has to change, I don?t need to get into that further its obvious. As for hypertransport I really hate you burst everyones bubbles, but to my knowledge it has not changed since its inception on the T-Bred Athlon? (Feel free to correct me if I?m wrong) As it is FSB technology will catch up to any benefit that hypertransport has minus a little latency very soon, thanks to DDR2.
With that covered let us summarize for the uniformed AMD zealots out there. AMD?s integrated memory controller was a gamble and it worked. Hypertransport has not changed since its inception. FSB continues to evolve pretty consistently. Intel has made 65nm chips, and is currently working on the 35nm manufacturing process. AMD has yet to even manufacture 65nm chips.

The next point I will make is Netburst. Now to everyone who says it is ?inferior? to AMD?s architecture. I hate to rain on your parade but your wrong, unless of course you provide a scope to your criticism. The Netburst pipeline is designed for a different kind of programming then the AMD (?short pipeline?) processors. Today?s programs are much more efficient on the ?short pipe? processors, but there may come a time when the longer pipeline will be an advantage and Netburst might show its ugly face again. If someone would like to explain exactly how that works they can go ahead.

Also I am not an Intel fan boy. In fact I own both brands of processors. In the past I had owned an Athlon 2400+ and I absolutely loved that box, which sadly recently died. In fact come to think of it as I upgrade I alternate company by some happenstance, so I am really not partial to either company.
I would like to also point out I do not want AMD or Intel to go out of business. I do like seeing the market share even out, that?s a good thing. This shift of market share requires that Intel be competitive and that requires them to innovate. Look at AMD for example, they haven?t done anything since 2003. Intel is even worse, and AMD capitalized by not innovating and just churning out the same old thing to make profit, a possibly cause why they have not switch to 65nm a possible cause why they wont support DDR2 until their next processor is near ready? I would hazard a guess to either they have trouble with these technologies or that?s the answer.

The last thing that troubles me is the extreme anti-intel attitude that certain technology review/preview services have taken. Misrepresentation of facts, and a lot of journalistic license have really started to irk me. That and the simple irrelevant comparisons that I see benchmark wise and the conclusions people draw from them. One recently brought to my attention was a 2.0 ghz P-M vs. 2.0 ghz Turion. All seems well correct? Nay my friend, The Turion laptop was running DDR 400 (very nice DDR400 if I may add), and the P-M laptop was running DDR2@533. Now I think most of you will know that DDR 400 is significantly faster then DDR2. So how can this be a ?comparison? of the two processors. I would say it would be a good comparison of the two notebooks, but not of the performance of the processors. Feel free to correct me if I?m wrong by saying the DDR is significantly faster then the DDR2. Unless I?m mistaken and I could be DDR2@800 should be pretty equivalent to DDR400.
Thank you and I hope you appreciated this small tome of mine, I did enjoy writing it as well.


I was silly to not to comment on something significant. AMD has really done one thing to note significantly better then Intel and they launched a mainstream consumer grade 64 bit processor. I do give them that, they accomplished that very well. They are by no means the first, I believe either Motorola or IBM hold that crown. But that is one accomplishment I give to them, while Intel only slapped on their 64bit support rather lackadaisically only because AMD did first.

i think actually tejas was a much more next gen cpu.

i believe the cpu you are talking about was called the timna. it was supposed to be celeronesque and have an integrated rambus controller and even work with integrated graphcis for really low cost PCs.


had rambus not completely failed it was basically going to be a pentium M like cpu. probably was a good idea too as it would have made for very simple motherboards (with a serial ram type and no memory controller on the northbridge the board would have been very simple).

unfortunately rambus sucked (not from a technical standpoint, but from a we are a bastard company patent standpoint, because technically it was a good idea).


and you do make a very good point about netburst. the main problem with netburst was that people at intel had not anticipated that leakage current would well become such a problem. the netburst architecture is really made for things like streaming media which really doesnt brand much, which well are just finally starting to take off. and had prescott say scaled to 5ghz, i think intel would still be in an ok position where no one would be bashing them at every turn.

even apple did not know that the 90nm architecture would not scale well (hence the g5 never hitting 3ghz) yet no one bashes them (thena gain most hardware sites do not cover macs).


yes it is very popular to bash intel now (i'm assuming because they are large and goliath-like, unlike say apple or amd), i mean it is easy to in hindsight, but in general they are a really good engineering company. they do sometime do things that area little "ahead of their time" such as netburst with streaming media (hence the not coincidental arrival of sse / sse2/ sse3 which is designed for decoding those things). but i still believe had it not been for the leakage issue that it wasnt that bad an idea.

i mean another thing that intel did that was "ahead of its time" was the 386. at the time of the 386 almost NO ONE used 32bit apps. at most people ran windows 3.0 which was all 16bit (i think only windows for workgroups 3.11 was 32bit at all, and that was basically the "PRO" version at the time, thus the "workgroups" part). so yeah.. the 386 was 32bit before anyone needed it, and that may have played a part in intel NOT putting out 64bit right away since they thought mabye they could wait a while and release itanium first and let it trickle down. i am not sure itanium was the ideal architecture either, but it is probably better than x86 from a general "not a kludge" perspective, and as a software engineer i tend to feel much less awful about that since x86 is pretty awful at this point and extending it yet again is even more awful.

it probably would have been BETTER had amd not extended it, so we could finally get rid of all the total garbage that x86 has. not to mention no one in the consumer desktop space even uses it still. the itanium now if it was manufactured at 65nm would actually be fairly small and economical if it did not have say 6mb of cache. a 2mb variant at 65nm would actually be affordable but it never took off obviously. i suppose intel is a bit more forward thinking than amd, since well in a way what AMD has done has somewhat set back cpus though they gave the public what they wanted.

so yeah i think all the intel bashing is unjustified, but well people like to jump on bandwagons and be part of the "winning team" so to speak. oh well whatever. and like the last poster who i applaud for also taking this unpopular stance i'm trying to be a little bit objective here. i know i know, wtf right?

also i'd like to say i have had both types of cpus. my first own computer that i got when i was 14 was a p100. my 2nd computer after that was a k6-200 engineering sample. after that a k6-2/450, a bunch of athlon, some p3s, some p4s, an athlon 2500. i have even had an athlon 64 2600 engineering sample. i've had an athlon 64 3000 as well and a pentium D 820 currently (the board in the amd64 actually shorted out, and i didnt bother to replace that machine). the last guy is right again on the AMD hasnt done anything since 2003 (whcih is when opteron came out). they may fall behind soon, as intel has a LOT of new stuff on the horizon and amd just has many more varients of the same architecture on a more antiquated production proces. so it will be interesting to see if people start jumping bandwagons again. and well based on that conclusion i actually just bought some intel stock as a long term investment.

 

[openwheelformula1]

boy I'd like to write a bunch of off topics and tell you every cpu I've ever owned and how innovated my favorite company has been over the last 20 years and how cute my new puppy is....but I won't put you through that cause the above post is enough.

Getting to the point: From a consumer's point of of view, we want every company to impliment the latest and greatest immediately. That only happens in a perfect world unfortunately. There are engineering minded company, there are marketing minded company, and then there are those in the middle. Problem is that Intel seems to be shifting from engineering minded, to somewhere in the middle, to marketing minded. There seems to be great things coming, but is it just more marketing?

Saying AMD "hasn't done anything" since 2003 is like say Intel hasn't done anything since Pentium3. Just because there isn't a complete revamp of cpu architecture doesn't mean they "haven't done anything." AMD's 130nm to 90nm trasition for example can pretty much top anything Intel has done since the introduction of netburst. What hyperthreading? The advantages of Winchesters/Venice... were much greater than something that very few could actually take advantage of.

 

[hans007]

Originally posted by: openwheelformula1
boy I'd like to write a bunch of off topics and tell you every cpu I've ever owned and how innovated my favorite company has been over the last 20 years and how cute my new puppy is....but I won't put you through that cause the above post is enough.

Getting to the point: From a consumer's point of of view, we want every company to impliment the latest and greatest immediately. That only happens in a perfect world unfortunately. There are engineering minded company, there are marketing minded company, and then there are those in the middle. Problem is that Intel seems to be shifting from engineering minded, to somewhere in the middle, to marketing minded. There seems to be great things coming, but is it just more marketing?

Saying AMD "hasn't done anything" since 2003 is like say Intel hasn't done anything since Pentium3. Just because there isn't a complete revamp of cpu architecture doesn't mean they "haven't done anything." AMD's 130nm to 90nm trasition for example can pretty much top anything Intel has done since the introduction of netburst. What hyperthreading? The advantages of Winchesters/Venice... were much greater than something that very few could actually take advantage of.

hyperthreading actually is a decided advantage when a process locks up the cpu for something stupid. it is one of the reasons trying to say transcode a dvd on a HT enabled p4 single core, is stilla lot faster than doing it on an athlon. intel has also made significant advances in the mobile sector.

also intel has always had really good marketing btw. and amd has always had well horrible marketing. should intel not market their products? i mean a smart business knows that marketing is a tool to make profits.

the only thing amd has done since 2003 is the current architecture. their roadmaps seem to indicate they will just basically ride on it until it dies.

how does amd transitioning to 90nm top any technical achievement intel has made. in the same span intel has transitioned from 130nm all the way to 65nm. i mean this amd love is just amazing. most of the people pumping amd , analysts included do not know 1/100th of what matters to a cpu.

case in point, thousands of analysts and "tech" sites saying that an onboard memory controller and hypertransport are where the speed is coming from... and the other posters was completely right, the reason netburst is not as good, is it is not as good with branched code because of pipeline stalls.

has anyone even noticed that the dothan chip is clock for clock the equal of a single core athlon, lower power at 90nm AND uses a non integrated memory controller and a pathetically slow according to most people 533mhz 64bit bus?

and i haevnt even started on the yonah. 90mm^2 die, uses less power, as fast at the same speeds, STILL uses a non integrated controller and still uses a 667mhz fsb and wait holy crap its as fast as the apparently beloved ddr2 am2 athlon64 x2s will be on a 220 mm^2 die @ 90nm?

seriously , the crap and myths about on die memory controllers and hypertransport that EVERY SINGLE tech site seems to reprint shamelessly as well as half the "technology analysts" the litter sadly respected publications like business week are just sad.

yes netburst sucks with current code. netburst with an ondie memory controller and hypertransport would STILL probably suck with short branched code. that is why intel doesnt care enough to put the on die controller on it.

and if anyone accuses me of being some sort of intel fanboy, i can produce for you pictures of watching "thresh" (dennis fong) playing quake2 painfully slow on k6-2s at amd's "i've seen the light with amd" gospel singing pathetic sadness of a comdex booth in 1997. hell i am wearing an AMD racing team tshirt right now. im just tired of the amd love fest bias.