View Full Version : What happened to SLDRAM?
Why did it dissapear? Was it too close to the design of RDRAM?
First ask this on RWT. Paul DeMone will likely recant a good history lesson.
From my understanding SLDRAM got canned when DRDRAM got the support by Intel rather than SLDRAM. SLDRAM from my understanding was a more open standard. Paul DeMone works for MOSAID and they were one of the contributing members to the SLDRAM consortium (I think that's the group).
Entropy
20-Aug-2002, 22:49
First ask this on RWT. Paul DeMone will likely recant a good history lesson.
From my understanding SLDRAM got canned when DRDRAM got the support by Intel rather than SLDRAM. SLDRAM from my understanding was a more open standard. Paul DeMone works for MOSAID and they were one of the contributing members to the SLDRAM consortium (I think that's the group).
Going from memory, SyncLinkDRAM was developed as alternative to RDRAM intended for JEDEC standardisation in order to avoid RAMBUS Inc. royalties and control. I distinctly recall some vicious name-calling between representatives for the two sides. :) With Intel backing RDRAM, the other players went on to support inertia (PC133 and DDR) rather than SLDRAM which would have added infra-structure costs similar to those of RDRAM albeit without the royalty costs and control concerns.
Again this is going from memory of old discussions, and I'll echo Saems' recommendation of Paul DeMone as one guy who might actually answer if you ask.
Entropy
Another detail was that on paper RDRAM was faster, of course later RDRAM prooved to not be manufacturable to their paper specs with the processes they promised it on.
I believe this was even brought to attention but supposedly RAMBUS was very good at convincing others that it would be possible. Interestingly enough, during the time when Rambus was making ludicrious promises without much in the way of proof of concept SLDRAM had been demonstrated to work at promised performance levels.
Rambus was running RAM at unheard of speeds in the early nineties. A lot of the stuff SLDRAM relied on was invented by Rambus.
So why would Intel choose Rambus if SLDRAM looked so good?
BTW, could you show me a link to a proof of concept SLDRAM device, Saem? All I've gotten were some distant forcasts before the consortium dissapeared from the radar screen.
Rambus was running RAM at unheard of speeds in the early nineties. A lot of the stuff SLDRAM relied on was invented by Rambus.
IIRC wasn't that in a far from mass production process? I'm not sure how to take the last statement. Is that an invitation to rehash the is Rambus fradulent in it's IP claims or are we talking about explicitly closed IP versus what was open supposed to be JEDEC, in the eyes of JEDEC?
BTW, could you show me a link to a proof of concept SLDRAM device, Saem? All I've gotten were some distant forcasts before the consortium dissapeared from the radar screen.
I remember someone mentioning it. I'll see what I can dig up.
So why would Intel choose Rambus
We made a big bet on Rambus and it did not work out. In retrospect, it was a mistake to be dependent on a third party for a technology that gates your performance...We hoped we were partners with a company that would concentrate on technology innovation rather than seeking to collect a toll from other companies
Everyone makes mistakes.
Marco
PS. SLDRAM's signalling technology is usually seen as an extension of SCI.
Everyone makes mistakes.
I don't buy it. 6+ years and millions of dollars were invested into the p4. I don't think they invested into RDRAM for the heck of it. The people who designed the P4s are not morons.
We made a big bet on Rambus and it did not work out. In retrospect, it was a mistake to be dependent on a third party for a technology that gates your performance...We hoped we were partners with a company that would concentrate on technology innovation rather than seeking to collect a toll from other companies
I'm not going to be baited into discussing this.
Out of curiosity, what is your degree in, Marco?
I'm not going to be baited into discussing this.
I can understand that you would not believe me when I say everyone can make mistakes, albeit I think its a touch naive to believe otherwise, but if you choose to ignore Craig barret too then I will have to do the same and stop discussing this :( I will just say that these kind of decisions are not solely taken by engineers, and with SLDRAM they could not have bought themselves a competetive "advantage" (well it would have been if RDRAM had been competetive from the get go) through an early alliance like they could with RDRAM.
Out of curiosity, what is your degree in, Marco?
EE.
You are way overselling Rambus's innovations. If Paul Demone's statements are not enough you might want to read the archives from sig-int mailing list and see how impressed experts in the industry have been with Rambus over the years.
I must admit that with yellowstone they seem to be turning it around, quite a sensible design IMO. Though of course in some ways it is closer to SLDRAM than to RDRAM ;)
Marco
I can understand that you would not believe me when I say everyone can make mistakes, albeit I think its a touch naive to believe otherwise, but if you choose to ignore Craig barret too then I will have to do the same and stop discussing this :(
I believe people make mistakes. I just don't believe that they didn't consider in depth what RAM type to use. The idea is fairly ludicris.
I will just say that these kind of decisions are not solely taken by engineers
Who else would make this kind of decision at Intel?!
EE.
How fast do you think SLDRAM would be at today?
You are way overselling Rambus's innovations.
Some things I know about Rambus (or think I know)
1) Rambus was running memory at very high speeds in the late eighties and early nineties.
2) Synchronization was made possible through some of Rambus' patents.
3) DDR SDRAM is just now starting to reach speeds that RDRAM reached in 1997.
If Paul Demone's statements are not enough you might want to read the archives from sig-int mailing list and see how impressed experts in the industry have been with Rambus over the years.
Link please? I will check it out.
I must admit that with yellowstone they seem to be turning it around, quite a sensible design IMO. Though of course in some ways it is closer to SLDRAM than to RDRAM ;)
So the same companies that were so close to providing amazingly fast SLDRAM can't use this same technology to make DDRII much faster than DDRI? I'm sorry, Marco. Some of the things you are saying are throwing warning signs up in my head.
We hoped we were partners with a company that would concentrate on technology innovation rather than seeking to collect a toll from other companies
Why does Craig Barret take this pot shot at Rambus? What company wouldn't protect their IP? I really don't understand the point he is trying to make through this quote?
I believe people make mistakes. I just don't believe that they didn't consider in depth what RAM type to use. The idea is fairly ludicris.
They considered it in depth, and then made a mistake.
I will just say that these kind of decisions are not solely taken by engineers
Who else would make this kind of decision at Intel?!
Like in any other company, management.
How fast do you think SLDRAM would be at today?
Probably not faster than Rambus, but I think it would have had less start up problems ... and without Rambus there would have been a more collaborative environment to speed along adoption of a sucessor standard.
1) Rambus was running memory at very high speeds in the late eighties and early nineties.
Let's make that early nineties.
2) Synchronization was made possible through some of Rambus' patents.
That and it was not meant for memory modules. We are talking the original RDRAM here, what we now call RDRAM was originally DRDRAM ... the followup standard.
3) DDR SDRAM is just now starting to reach speeds that RDRAM reached in 1997.
Quite impressive how far they can flog a dead horse, all considering.
Link please? I will check it out.
Signal integrity mailing list archives, try google :)
So the same companies that were so close to providing amazingly fast SLDRAM can't use this same technology to make DDRII much faster than DDRI?
Evolution versus revolution.
We hoped we were partners with a company that would concentrate on technology innovation rather than seeking to collect a toll from other companies
Why does Craig Barret take this pot shot at Rambus? What company wouldn't protect their IP? I really don't understand the point he is trying to make through this quote?
He apparently thinks differently on just how far their patents should extend.
Also I think he just doesnt like them very much on a personal level :) Intel became big because of technical excellence, not through protecting their IP (they tried, and failed mostly). Intel makes real products, its not hard to take pride in that and value it over just milking patents ... especially not if the company who owns the patents uses selective licensing to try to ensure they control the memory standardization process (which has the tactical advantage that it chills R&D from competitors).
Like in any other company, management.
And who do you think their management is made up of? Look at Craig Barret . . . what do you think his degree is in? :wink:
and without Rambus there would have been a more collaborative environment to speed along adoption of a sucessor standard.
I don't know about this. RDRAM was running at 800MHz before DDR SDRAM came out.
Those JEDEC guys are not all that collaborative, anyway. How much IP would one invest in an open standard? Not too much benefit from my point of view.
I'm still looking at the archives.
An engineer engineers, a manager manages. Management had to make decisions based on incomplete data shaped partly by internal politics and partly by the predictions of Rambus themselves (which did not turn out to be accurate). They also took into account what the success of Rambus could mean for them if they allied themselves with them ... so they had reason to make optimistic rather than pessimistic assumptions.
As long as it was still mostly paper even the best assesment Intel's engineers could make had to be partly based on guesswork ... and management relied on a wrong set of assumptions, neither them being engineers in the past nor being advised by engineers allows them to predict the future.
They made a mistake.
Money is made by companies making actual products, for them there is always an advantage in being first and for that they have to be actively involved in progressing the standards (in the case of memory).
Marco
PS. on reflection I should add to all this that while Intel backing SLDRAM instead of RDRAM might have made it succeed ... SLDRAM might have failed just as RDRAM, albeit a little less spectacularely (I dont think Intel would have had the same spectacular technical failures they experienced with RDRAM, which both delayed introduction and were horrid PR). Some of the biggest reasons why the industry did not like RDRAM apply to SLDRAM too, basically it just did not offer enough advantage to justify its added costs in die size and other costs related to breaking away from SDRAM.
Well I don't agree with you, but that's ok. I'm pretty sure Intel knew to almost the exact fps scored in Quake at different resolutions before the P4 even became silicon. As I understand it, it is actually the P4 design team themselves that have been pushing for RDRAM from the beginning. Many of their managers aslo have EE degrees, including the ones that would make the decisions.
He apparently thinks differently on just how far their patents should extend.
Also I think he just doesnt like them very much on a personal level :) Intel became big because of technical excellence, not through protecting their IP (they tried, and failed mostly). Intel makes real products, its not hard to take pride in that and value it over just milking patents ... especially not if the company who owns the patents uses selective licensing to try to ensure they control the memory standardization process (which has the tactical advantage that it chills R&D from competitors).
I'm not sure where you are comming from on this? The capitalistic system depends on IP. Intel is extremely fierce in protecting its' IP and makes quite a good amount of money on licensing alone. They would have done the same thing in Rambus's situation and probably would have won in the courts by now. How can you be angry at a company for protecting its' IP?
Even if Barret didn't like them, how come Intel signed a cross licensing agreement with them?
What does performance when they finally got the RDRAM to work to spec have to do with anything? That doesnt sell product ... RDRAM's cost and RDRAM's ability to operate in a complex system were the problem. Most of the reasons for the cost were not outside of the realm of engineering (and a little plain common sense on how markets operate) and all of its problems with operation which Intel ran into were down to engineering. Yet they failed to foresee both.
If cost did not matter someone could have easily designed a memory standard using an eDRAM process which would beat RDRAM in per pin bandwith for quite a while now ... we already have interfaces (but not memory specific) which are faster than RDRAM.
The capitalistic system depends on competition, on one hand patents will ensure companies competing to innovate where without patents the price of innovation might not be warranted by market benefits ... on the other it promotes companies to rest on their laurels and to try to use their patents to actually prevent competition (I assume I dont need to point out the myriad of ways in which that can happen, never forget that patents represent government granted monopolies). Wether it is a net win is anyone's guess, impossible to proove one way or the other.
I think everything we depend on in IT would have been discovered regardless of patents ... do you really think it would have taken 17 years for someone to commercialize synchronous signalling in memory standards without Rambus? (Let alone some of the more trivial stuff they claim intellectual ownership of.)
Intel makes money selling good products, the rest is just bonus ... their biggest competitor has access to pretty much all their IP, and that fact has done us as consumers a whole lot more good than bad Id say.
As for why they went into buessinuess with them, if others could have executed well when implementing their technology it would have been a good deal regardless of how "evil" they are :) You asked why he would say that, I said probably because he doesnt like them ... but as a manager he wouldnt let that influence his decisions if there was money to be made ... that was where it went wrong though.
What does performance when they finally got the RDRAM to work to spec have to do with anything? That doesnt sell product ... RDRAM's cost and RDRAM's ability to operate in a complex system were the problem. Most of the reasons for the cost were not outside of the realm of engineering (and a little plain common sense on how markets operate) and all of its problems with operation which Intel ran into were down to engineering. Yet they failed to foresee both.
:-? For what they had to work with, I think they made a good choice. The i850, IMO, is one of the top chipsets ever made. Look at all the crap that DDR SDRAM went through. The mobos were nine months late and they still sucked. There is still inherent istability with DDR SDRAM chipsets. You think Intel wanted anything to do with it? What are the other alternatives? SDRAM? An EE told me at Sharkys that the definition of an engineer is, "Someone who can do for a penny what any fool can do for a pound." Something has to fill in the gap before the perfect system is created. How would you have gone about it, MfA? It's pretty easy to create the ideal product in your head. It's very hard to make that product successful in the real world.
If cost did not matter someone could have easily designed a memory standard using an eDRAM process which would beat RDRAM in per pin bandwith for quite a while now ... we already have interfaces (but not memory specific) which are faster than RDRAM.
So you want to compare RDRAM with DDR SDRAM which brings in no profit?
The capitalistic system depends on competition, on one hand patents will ensure companies competing to innovate where without patents the price of innovation might not be warranted by market benefits ... on the other it promotes companies to rest on their laurels and to try to use their patents to actually prevent competition (I assume I dont need to point out the myriad of ways in which that can happen, never forget that patents represent government granted monopolies). Wether it is a net win is anyone's guess, impossible to proove one way or the other.
You have a better system in mind, then?
I think everything we depend on in IT would have been discovered regardless of patents ... do you really think it would have taken 17 years for someone to commercialize synchronous signalling in memory standards without Rambus? (Let alone some of the more trivial stuff they claim intellectual ownership of.)
Who cares? Rambus did it first so they have licensing rights. If you don't agree with this system you can buy a house in Cuba. Look how successful their system is.
Intel makes money selling good products, the rest is just bonus ... their biggest competitor has access to pretty much all their IP, and that fact has done us as consumers a whole lot more good than bad Id say.
Intel fiercelydefends their IP. Sometimes it helps consumers, sometimes it doesn't. There are many large companies that make huge amounts of money from IP (IBM comes to mind). No, I don't think this is overly negative to consumers.
I cannot put myself in their mindset, we neither have their knowledge (even after all this time) nor have I their experience ... but given mr. Barret's comments I would tend to believe that a more cautious management should have decided on a two pronged approach.
As for how justified betting the farm on RDRAM was given its future "potential" ... well Im sure Intel and its shareholders did not view their lost sales due to lack of timely DDR chipsets quite as complacently as you. Disregarding that though in the end I think DRDRAM will dissapear without ever having made the investments back ... and you can hardly call it a step up to Yellowstone, hell that might not even end up becoming the winner against Intel's own ADT initiative (although I think thats flagging a bit) unless they can force the issue through licensing.
As long as money is being pumped around there is always people profiting, and DDR does it a lot faster than RDRAM.
I dont quite see why you equate patents with capitalism ... its quite trivial to see you can have a free market economy without patents, and Cuba is not that so it in no way prooves your point. Im not a communist, I believe in social democracy (I think its only just as long as you dont have a libertarian government anyway, and I think libertarianism while admirable in some ways just would not work ... although as far as patents are concerned Ill agree with libertarians :). I would like to see patent protection eroded, and if it turns out there are major areas of R&D which benefit society but which dont get funded that way ... they will just have to be collectivized ;) Or more likely just maintain patent protection in those fields. IT definetely wont be one of them though IMO (and not just because its benefits to society are so small either :).
Noone ever said monopolies cant make good money (talking about the patents themselves, not the individual companies who own them). That says very little about their net benefit to society, I shouldnt have used Intel as an example sorry ... should have stuck to "impossible to definetely proove one way or the other".
MfA,
could you elaborate on ADT?
Thread topic,
I believe Intel wished to use DRDRAM to basically take controll of the market to a greater degree. They had a lot of money in Rambus and would have made a huge amount of money. It didn't work so they're cleaning up the mess that's left behind. Mr. Barret's words are likely the PR damage control/clean up.
As for patents, I'm with you MfA.
Google a bit, on a technological level they do not have a lot to show and there seems to be a lot of friction with JEDEC ... it might end up disbanding and just reintegrating their efforts with JEDEC (all the ADT members are JEDEC members anyway). On a political level prophetic (http://www.ebnonline.com/story/OEG20000717S0022>this</A>).
PPS. or on second thought it shows that considering it is all they are doing, still ...
Heh, I should have thought to google, my only concern was that an acronym like ADT would take me everywhere.
In anycase, I followed a link, which ended up being and EBN article covering the formation of the group.
As for you PS, MfA, the first EBN article you linked to hinted towards such a thing. I hope it goes through.
mboeller
26-Aug-2002, 09:22
The capitalistic system depends on IP.
Couldn't be farther from the truth. IP prevents competition. So the prices stay high. So consumers have to pay far higher prices and economy suffers. Most of the time even the IP-holder suffers, cause they sit on their IP milk it and don't improve/innovate. So when the tides turn they go under.
Most of the internet is open-source. Expecially the standards. All/Most the propretary interfaces/standards are gone. Without open-source their wouldn't be an internet. So their wouldn't be all this hype (which would have been good) but no internet-economy too (really bad, if you see how much money is made and how much people are employed ).
For private people IP is deathly even. Patents only "allow" you to protect "Your right" in court, but this is so expensive that you loose always, even when you win the battle.
lol . . . I should have known I was going up against people who believe in Socialist Democracies! :)
Anyway . . .
As for how justified betting the farm on RDRAM was given its future "potential" ... well Im sure Intel and its shareholders did not view their lost sales due to lack of timely DDR chipsets quite as complacently as you. Disregarding that though in the end I think DRDRAM will dissapear without ever having made the investments back ... and you can hardly call it a step up to Yellowstone, hell that might not even end up becoming the winner against Intel's own ADT initiative (although I think thats flagging a bit) unless they can force the issue through licensing.
I tend to desagree with you on this. I think that the average American consumer is not all that stupid and when he realizes the cost savings he will have by going with a 2.6GHz system on RDRAM that runs at the same effective speed as a 3GHz system with DDR SDRAM, RDRAM will win out. But that is JMO.
I also see a movement towards 10GHz at an incredible pace. In the future, all that will matter are BW and latencies. As the President of Samsung said, "Sure, DDR SDRAM is nice at the speeds we have now, but will we use it in the 3GHz range? I don't think so." (I can't remember the quote exactly, but it was pretty close to this statement). I don't see how DDR SDRAM will stay even close to RDRAM in the future. These 128-bit busses that are planned will cost quite a bit and will still not give the BW we need at 5GHz+.
This proves my other point in a way. The gigantic companies that contribute to JEDEC won't be able to be even close to the technology that Rambus is coming out with despite the fact that they are open source. If Infineon had just decided to pay Rambus royalties, they just might not be losing 900 million dollars a quater. Samsung paid Rambus royalties and are making huge profits on one of the few profitable mainstream memories.
It doesn't get more black and white than this: IP has allowed huge advances in technology compared to the backstabbing open source companies at JEDEC despite the lopsided investment into RandD at these huge companies. JEDEC is open source and look at the inherent stability problems that DDR SDRAM has compared to RDRAM. JEDEC created a shoddy product.
I dont quite see why you equate patents with capitalism
Capitalism with IP is much more effective. Without encouragement to invest into RandD, technology will simply stagnate.
I believe Intel wished to use DRDRAM to basically take controll of the market to a greater degree.
Here's my opinion: Intel chose RDRAM because it is superior to DDR SDRAM. Pretty simple opinion, eh? :)
Couldn't be farther from the truth. IP prevents competition. So the prices stay high. So consumers have to pay far higher prices and economy suffers.
If the consumers want to pay the price, they will. If they don't want to pay the price, they won't (basic economics). In America, the economy doesn't suffer with higher prices, it actually improves (if the consumer wants to pay for the product). I don't see where you are coming from on this one.
Most of the time even the IP-holder suffers, cause they sit on their IP milk it and don't improve/innovate. So when the tides turn they go under.
In the US, companies are allowed to add inovations to existing products and thus extend the amount of time companies are allowed to collect royalties on patents. Thus, there is large incentive to innovate (this is especially true in the medical industry).
Find me a country with a good IP system and a bad economy (I can't think of one off the top of my head).
What is your political colour BTW? Probably not a libertarian, because government granted monopolies are pretty hard to unite with that without some major mental blinders ... so fiscal conservatism? As long as you support big government we arent really that far apart you know :)
I can trade quotes for quotes, Intel's Pat Gelsinger on Rambus early this year :
great technology but more than likely to remain a niche technology due to legal and industrial dynamics
I hope by the time we are at 5 GHz that for really high bandwith requirements external memory will loose out to embedded. If/Where we do need high bandwith external memory I hope DRDRAM wont be the interface of choice, maybe Yellowstone ... maybe something else (probably something serial if Rambus's synchronous memory patents stick around till then).
Profit margins on commodity memory are non existant because of competition and over capacity, if RDRAM was the commodity memory standard those underlying causes would not go away.
RDRAM is not a huge advance in technology, if it was an order of magnitude better Id agree with you ... but what are we talking about in practice? 30% more bandwith, about 5% higher performance IRL and about 100% higher price. Rambus was commercialized before its time, it offered no real advantages up till only very recently.
If Rambus had wanted to make it a success they could have always gone with a RAND patent licensing policy, or put their money where their mouth was and got their own manufacturing capacity for that matter. Talking about return on investments, the memory manufacturing companies are the one's taking the real monetary risks ... why should most of the profit margins (because everything besides the IP licensing costs are just going to melt away due to competition) go to a company which invested the least? :) They have no obligation to do Rambus's work for them.
Countries which can effectively trade on the global market place are pressed into IP laws by WTO, so thats a bit of a self fulfilling prophecy. Still plenty of countries not doing so hot at the moment, Argentina for one ... on the other side we have China as an example of a country with shoddy protection of IP and a good economy but you would just say that they profit using other people's R&D ... which goes to proove my point, the present situation is unique and no definite proof either way can be given using the situation itself since that just ends up in circular arguements. Its not like we have a second world to compare this one against where we can experiment with different settings.
For some thats a reaon to stick to the status quo, if it aint broke dont fix it. Being broke is such a relative concept though.
mboeller
27-Aug-2002, 07:49
Couldn't be farther from the truth. IP prevents competition. So the prices stay high. So consumers have to pay far higher prices and economy suffers.
If the consumers want to pay the price, they will. If they don't want to pay the price, they won't (basic economics). In America, the economy doesn't suffer with higher prices, it actually improves (if the consumer wants to pay for the product). I don't see where you are coming from on this one.
So You don' know so much about marketing. Reduce the price to 50% then you will sell 5-10times as much (at least most of the time) look at Nvidia for example. They had to lower the price of the GR-Ti-Series (GF4Ti4200) and sell now far more high performance parts then before. This works nearly everywhere the same. So lower prices increase the demand, higher prices lower the demand. In the end lower prices means better business for the companies.
Find me a country with a good IP system and a bad economy (I can't think of one off the top of my head).
maybe the US when the software/standards-patents are in widespread use, because they only protect ideas and no products.
Entropy
27-Aug-2002, 15:25
I don't see how DDR SDRAM will stay even close to RDRAM in the future. These 128-bit busses that are planned will cost quite a bit and will still not give the BW we need at 5GHz+.
What, exactly, do you base this statement on?
128-bit wide DDR boards utilize four layer motherboards. Dirt cheap. All chipset manufacturers will have 128-bit products out in Q402, giving 5.3GB/s nominal bandwidths with standard 333MHz DDR.
As far as roadmaps show, 128-bit DDR and DDR-II will meet or exceed projected requirements.
Entropy
I believe the move to RDRAM with the pentium 4 was similar to the move to slot 1 from socket 7 by Intel in the Pentium 2 days..
basically they are large enough to exclude the competition by creating and maintaining a proprietery platform.
after the socket>slot switch intel admitted the slot was a fucked up idea and moved back to socket.
Intels done this before and everytime they do the industry standards get thrown for a loop.
remember when you could choose between pentium and k62 on a single motherboard standard?
think about it, if microsoft told the industry that the sucessor to windows XP would only run on systems with a special bios installed (palladium?) do you have any doubt that the chip would be on every motherboard coming out of asia the following year?
and if this chip stopped installation of any non-microsoft operating system would manufacturers dare remove it?
monopolies can force standards, but obviously intel miscalculated the power of it competitors in popularizing an open standard.
What is your political colour BTW? Probably not a libertarian, because government granted monopolies are pretty hard to unite with that without some major mental blinders ... so fiscal conservatism? As long as you support big government we arent really that far apart you know :)
[US PERSPECTIVE]
I'm not sure what your definition of a libertarian is. To me, it is someone who advocates free will. I'm somewhat of a libertarian. I believe that seatbelt laws are stupid, but drugs should still be illegal. I am also somewhat of a fiscal conservative. A private debt is not as bad as a public one, but it is still not as efficient as having no debt. Money that could be spent on specified programs is instead spent freely in piecemeal amounts. Instead of Social Security, the government should pay for medical companies RandD, which they already do to some degree. A certain amout of welfare is needed to take care of people who can't take care of themselves. The current hospital situation in America is pretty good. In general, people get attention when they need it. From what I've heard, patients, in general, have to wait longer for treatment in England and genrally have less high tech equipment. Certain industries, like power and public transportation, should be regulated because they are necessary.
[/US PERSPECTIVE]
Profit margins on commodity memory are non existant because of competition and over capacity, if RDRAM was the commodity memory standard those underlying causes would not go away.
You are probably right, but things would be better, at least. Computers would be more stable and the industry wouldn't have to support two different memory types which would bring down costs for manufacturers and consumers.
They have no obligation to do Rambus's work for them.
Of course not. But if they want to make some money, they will pay Rambus royalties so they can make RAM. Paying a dollar to Rambus for every RAM stick sold is not that bad considering the fact that Rambus invested millions of dollars into research.
So You don' know so much about marketing. Reduce the price to 50% then you will sell 5-10times as much (at least most of the time) look at Nvidia for example. They had to lower the price of the GR-Ti-Series (GF4Ti4200) and sell now far more high performance parts then before. This works nearly everywhere the same. So lower prices increase the demand, higher prices lower the demand. In the end lower prices means better business for the companies.
If something has high enough demand, people will pay for it. Illegal drugs are a good example.
What, exactly, do you base this statement on?
128-bit wide DDR boards utilize four layer motherboards. Dirt cheap. All chipset manufacturers will have 128-bit products out in Q402, giving 5.3GB/s nominal bandwidths with standard 333MHz DDR.
As far as roadmaps show, 128-bit DDR and DDR-II will meet or exceed projected requirements.
http://www.rambus.com/rdf/presentations/1_05_RambusRoadmap_Fox.pdf
pg. 22
I've got more roadmaps if you want them, but take a look at the design of both RAM types. RDRAM is on a 32-bit bus right now and still exceeds the BW of DDR SDRAM with a 64-bit wide bus. Which do you think will hit the ceiling first?
128-bit wide DDR boards utilize four layer motherboards. Dirt cheap. All chipset manufacturers will have 128-bit products out in Q402, giving 5.3GB/s nominal bandwidths with standard 333MHz DDR.
Such implementations are not known for their stability, the nForce even had a stability mode which downclocked the memory bus because of stability issues. Other wide bus implemenations are usually done with a very large amount of layers.
To add to what Saem said, the DC DDR solutions will not be dirt cheap. The Granite Bay has over 1000 pins! Even with Intel selling their motherboards at a loss, the Granite Bay will be very expensive. The same will go for the 4 layer DC DDR solutions.
By 2005, RDRAM will be at almost twice the speed that DDR SDRAM reaches. And I think the DDR camp is being a little optomistic.
Entropy
28-Aug-2002, 21:50
What, exactly, do you base this statement on?
128-bit wide DDR boards utilize four layer motherboards. Dirt cheap. All chipset manufacturers will have 128-bit products out in Q402, giving 5.3GB/s nominal bandwidths with standard 333MHz DDR.
As far as roadmaps show, 128-bit DDR and DDR-II will meet or exceed projected requirements.
http://www.rambus.com/rdf/presentations/1_05_RambusRoadmap_Fox.pdf
pg. 22
I've got more roadmaps if you want them, but take a look at the design of both RAM types. RDRAM is on a 32-bit bus right now and still exceeds the BW of DDR SDRAM with a 64-bit wide bus. Which do you think will hit the ceiling first?
That is not a roadmap. That is a straight line drawn in an Excel chart in a promo document for RAMBUS. There is nothing whatsoever that says it has anything to do with Intels (or AMDs) plans. And incidentally, the "need" it projects is covered nicely by 128-bits of DDR, which is why they conveniently only compare it with (conservative) estimates of 64-bit DDR. DDR is hitting a practical ceiling just about now, btw. So it evolves into DDR-II. And there are already techniques to move beyond that as well.
Retreating a bit from the immediate issue, there is little point in trying to make long term extrapolations anyway, since one of the endearing properties of the PC platform is that if there's a need, eventually there will be a dirt cheap kludge to fill it. Judging by a 80386, you would extrapolate to 3 GHz fast page mode RAM. Bus multipliers, L1, then L2 cache, wider busses, new memory technology, multiprocessors with independent memory channels... - the need you would have projected has been adressed in ways other than just extrapolating existing solutions.
128-bit wide DDR boards utilize four layer motherboards. Dirt cheap. All chipset manufacturers will have 128-bit products out in Q402, giving 5.3GB/s nominal bandwidths with standard 333MHz DDR.
Such implementations are not known for their stability, the nForce even had a stability mode which downclocked the memory bus because of stability issues. Other wide bus implemenations are usually done with a very large amount of layers.
Hard to say much about their stability at this point in time since the nForce is the only current example. It's users seems happy enough with the latest bios upgrades, particularly considering that it is nVidias first effort in the mainstream chipset business. It works. It's cheap. Hardly statistics enough to say anything in general about 128-bit DDR on four layer boards though. Since nVidia and all other players are introducing similar new solutions I would be surprised if there were general problems particularly as the ground has already been broken. (I could make a low blow against RDRAM chipsets, but the 850 actually seems OK). Of course, although they will try to make as much as possible from these chipsets as long as they are news, just as nVidia did, we are dealing with el cheapo PC motherboards, where the total cost of components is very low indeed, and profit margins are slim. There will be wine and there will be vinegar in such a market. Buyer beware, as always.
Entropy
Retreating a bit from the immediate issue, there is little point in trying to make long term extrapolations anyway, since one of the endearing properties of the PC platform is that if there's a need, eventually there will be a dirt cheap kludge to fill it. That is not a roadmap. That is a straight line drawn in an Excel chart in a promo document for RAMBUS. There is nothing whatsoever that says it has anything to do with Intels (or AMDs) plans. And incidentally, the "need" it projects is covered nicely by 128-bits of DDR, which is why they conveniently only compare it with (conservative) estimates of 64-bit DDR.
Touche.
Of course, I personally believe that today's processors could easily soak up the BW from a RIMM9600. In fact, I wouldn't be suprised if the 2.8GHz P4 could eat up the BW from a 40GBs RAM stick. The BW needs at 10GHz will be much more important than the needs we have today and people will be willing to pay more for RAM.
http://www.rambus.com/about/rdf_taiwan_2002/Main_4_Samsung_RDFT.pdf
pg. 22
DDR is hitting a practical ceiling just about now, btw. So it evolves into DDR-II. And there are already techniques to move beyond that as well.
And they look a lot like RDRAM!
Retreating a bit from the immediate issue, there is little point in trying to make long term extrapolations anyway, since one of the endearing properties of the PC platform is that if there's a need, eventually there will be a dirt cheap kludge to fill it.
One aspect of engineering is that there must be a product to fill in the gap before the perfect product can come out. Engineers must choose the best product they have at the moment. We can't wait ten years for some magical product to give us massive BW at extremely low costs. And in the next 10 years what solutions do we have? We have RDRAM, and we have DDR SDRAM which looks more like RDRAM every day.
Judging by a 80386, you would extrapolate to 3 GHz fast page mode RAM. Bus multipliers, L1, then L2 cache, wider busses, new memory technology, multiprocessors with independent memory channels... - the need you would have projected has been adressed in ways other than just extrapolating existing solutions.
Your point?
Hard to say much about their stability at this point in time since the nForce is the only current example.
The laws of physics dictate that using a four layer motherboard with cheap components will result in poor stability. You can't have something for nothing. There are always tradeoffs. If the current high end 64-bit six layer DDR SDRAM motherboards have stability problems, I don't see how the 128-bit wide mobos on four layers using cheap components are going to do much better. Maybe the channels will have to be at right angles to each other. But that is just another tradeoff which results in decreased room for periphials on the mobo.
From my understanding DDR boards are 4 layer and until recently DRDRAM boards were 6 layers, they've moved to 4.
DRDRAM's offered bandwidth isn't chewn up, yet. However, DDR's is not only chewn up but falls short and then some. This can easily be seen in benches where you can see 2.6-2.8GHz P4s blowing pas 3.0GHz ones. Tom did a fair number of OC tests.
Even if we pair ANY P4 performance class MPU with PC 2700 and a 64 bit interface we'd still be in a fair amount of trouble when it comes to being starved for bandwidth, to the point that the effects of latency would take a back seat.
With all this said and done. I don't believe Rambus has claims to the IP they've "grandfathered" and claim to own. SLDRAM, seems like it would have been a better choice.
Personally, I hold little hope for DDR II, since that entire multiplexing the bus and reducing memory array clocks to get more bandwidth doesn't sit well considering the fetch size will be a wee bit big. I'm sure people will use it, but it'll suck donkey balls for the most part.
What I am looking forward to is more intelligent memory. I think there has been too much centralization of "intelligence" in machines and I'm of the firm belief that this is the wrong way to go.
Entropy
29-Aug-2002, 17:51
Of course, I personally believe that today's processors could easily soak up the BW from a RIMM9600. In fact, I wouldn't be suprised if the 2.8GHz P4 could eat up the BW from a 40GBs RAM stick. The BW needs at 10GHz will be much more important than the needs we have today and people will be willing to pay more for RAM.
Heh. I'm all for more bandwidth. IMO the PC-platform got stuck at PC133x64-bits for far too long. On the other hand, there are several areas that don't benefit from better main memory subsystems. And most benchmarks tend to do a lot of work on small data-sets, rather than increasing the problem size, which is what I often see as causing a need for increased processing capability. Thus bandwidth doesn't carry the marketing clout it could have, if we only had looked at problems that really DO benefit from higher performance, rather than estimating office or internet use performance (* cough *).
Your point?
Simply an example which my meager language skills failed to make very clear.
Hard to say much about their stability at this point in time since the nForce is the only current example.
The laws of physics dictate that using a four layer motherboard with cheap components will result in poor stability. You can't have something for nothing. There are always tradeoffs. If the current high end 64-bit six layer DDR SDRAM motherboards have stability problems, I don't see how the 128-bit wide mobos on four layers using cheap components are going to do much better. Maybe the channels will have to be at right angles to each other. But that is just another tradeoff which results in decreased room for periphials on the mobo.
We'll see. I'd tend to agree in general, but the proof of the pudding is in the eating, and if these manufacturers ship kit that works, I'll accept that they have gotten good enough. I have some faith in Intels validation programmes. Actual implementation could well matter quite a bit, so I would be conservative in my MB brand choice, if I was in that market. But that may just be my suspicious nature.
DRDRAM's offered bandwidth isn't chewn up, yet. However, DDR's is not only chewn up but falls short and then some. This can easily be seen in benches where you can see 2.6-2.8GHz P4s blowing pas 3.0GHz ones. Tom did a fair number of OC tests.
Even if we pair ANY P4 performance class MPU with PC 2700 and a 64 bit interface we'd still be in a fair amount of trouble when it comes to being starved for bandwidth, to the point that the effects of latency would take a back seat.
This is both application dependent obviously, and dependent on just how large differences in latency you are talking about. There is little doubt that the on-chip memory controller of the Hammers will be a huge benefit. Preliminary data say that latency will be roughly half that of the Athlons, and that is enough to have a major impact. Of course, Real Men will buy Opterons with 128-bit busses. :)
With all this said and done. I don't believe Rambus has claims to the IP they've "grandfathered" and claim to own. SLDRAM, seems like it would have been a better choice.
I'm not touching that nest of vipers with a ten-foot pole. I stood at the sidelines enjoying the fight when these technologies slugged it out, and they seemed to offer similar potential. They have both lost in the marketplace now of course, but good ideas tend to hang around and be implemented at a later date.
Personally, I hold little hope for DDR II, since that entire multiplexing the bus and reducing memory array clocks to get more bandwidth doesn't sit well considering the fetch size will be a wee bit big. I'm sure people will use it, but it'll suck donkey balls for the most part.
Are we feeling grumpy today? :) I'm not with you on the fetch size. I have the option of changing the burst length on this motherboard, and my experiments have shown only minor differences for actual applications. I'd be more concerned about lack of critical word first bursting, but while that should have an impact, it is relatively limited compared to the (projected) clock benefits.
What I am looking forward to is more intelligent memory. I think there has been too much centralization of "intelligence" in machines and I'm of the firm belief that this is the wrong way to go.
Check out Session Four on this years MicroProcessor Forum.
"Session Four: Extreme Processors
Max Baron, Senior Editor, Microprocessor Report; Principal Analyst, In-Stat/MDR Active Memory Device Delivers Massive Parallelism
Graham Kirsch, Chief Architect, Active Memory Program, Micron Technology, Inc.
Micron will present its Active Memory device architecture featuring closely interfaced on-chip processors and DRAM memory. Micron's Active Memory architecture achieves high-performance by taking advantage of on-chip distributed private memories that can deliver fast access times and high bandwidth."
Interesting, no? ;)
It is regrettable that the current model for PC-evolution doesn't suit radical changes well. Optimums that require major redesign or rethinking of either hardware or software, (particularly together!) are unlikely ever to be reached, likewise if backwards compatibility is compromised.
Entropy
From my understanding DDR boards are 4 layer and until recently DRDRAM boards were 6 layers, they've moved to 4.
There are quite a few DDR SDRAM boards that are manufactured on six layer boards. The majority are pretty unstable. Going to a 128-bit bus makes verification insane.
DRDRAM's offered bandwidth isn't chewn up, yet.
RAM runs at 1066MHz while the P4 is currently at 2.8GHz. There is quite a lot of room for improvement and this is why caches are so necessary. If we had RAM that could provide 40GBs of BW right now it still wouldn't be fast enough.
Even if we pair ANY P4 performance class MPU with PC 2700 and a 64 bit interface we'd still be in a fair amount of trouble when it comes to being starved for bandwidth, to the point that the effects of latency would take a back seat.
The difference in latency between DDR SDRAM and RDRAM is so small most of the time as to be rediculous. If latency was really all that important, RIMM4200s would have much higher scores than RDRAM.
Personally, I hold little hope for DDR II, since that entire multiplexing the bus and reducing memory array clocks to get more bandwidth doesn't sit well considering the fetch size will be a wee bit big. I'm sure people will use it, but it'll suck donkey balls for the most part.
It's a good way to extend the ceiling of DDR SDRAM.
Interesting, no?
Not without the actual link.
Entropy
29-Aug-2002, 21:51
To sign up for MPF in San Jose, go here
https://www.mdronline.com/corpreg/eventreg_router.asp?eur=eventreg_frm_new.asp&nur=e ventreg_frm_new.asp&npr=eventreg_frm_update.asp&mo de=MPF
To see the schedule, go here
http://www.mdronline.com/mpf/mpf-brochure.pdf
Entropy
Are we feeling grumpy today? I'm not with you on the fetch size. I have the option of changing the burst length on this motherboard, and my experiments have shown only minor differences for actual applications. I'd be more concerned about lack of critical word first bursting, but while that should have an impact, it is relatively limited compared to the (projected) clock benefits.
I was a big grumpy. Hehehehe.
The fetch size is more important outside the PC realm. So I guess it's not really a valid point here.
There are quite a few DDR SDRAM boards that are manufactured on six layer boards. The majority are pretty unstable. Going to a 128-bit bus makes verification insane.
The nForces are 4 layer and run fairly well. Same with the i845s and so on. At least that's my understanding. I'm not sure where you're getting 6 layers from, unless we're heading into workstation and lowend server.
This is both application dependent obviously, and dependent on just how large differences in latency you are talking about. There is little doubt that the on-chip memory controller of the Hammers will be a huge benefit. Preliminary data say that latency will be roughly half that of the Athlons, and that is enough to have a major impact. Of course, Real Men will buy Opterons with 128-bit busses.
In my opinion the integrated memory controller with DDR is a bit pointless. The latencies are already pretty low, effective bandwidth will see a slight improvement and the real gains are to be had with faster DDR, which they don't have currently. I'd hate to say it, the only mass market technology they could really go with is DRDRAM, but the royalties = suck. Seeing as they're on both interface and memory. Then again, the MPU would look like a poor man's Alpha all over again and we know how much the AMDroids would hate that. ;) Okay, enough teasing, seriously, the integrated memory controller seems like the only way AMD can hope to keep up with Intel's insanely good memory controllers. I doubt, Via, SiS, ALi, AMD and even Nvidia can really compete. Then again, Intel isn't a weaklying, but a long shot.
mboeller
30-Aug-2002, 08:10
RAM runs at 1066MHz while the P4 is currently at 2.8GHz.
the Interface of RDRAM runs at 533 MHz. RDRAM is DDR like normal DDR-SDRAM.
The difference in latency between DDR SDRAM and RDRAM is so small most of the time as to be rediculous. If latency was really all that important, RIMM4200s would have much higher scores than RDRAM.
If the Latency wouldn't be important RDRAM would have been much faster then SDRAM from the beginning. Most Benchmarks show that Latency is really important. Latency means that the CPU is idle for as long as it takes to feed the first data.
Latency is important in context. That is to say, in the context of a P4 system where latency is well addressed and hidden rather well, it's a lesser issue -still important.
In the context of a P!!! such is not the case. I doubt the same would be the case for the Athlon and even possibly the Hammer, but the Hammer has the luxury of the integrated memory controller which would have made interesting possibilities with DRDRAM. Then again, my politically inclined side is going to start yelling if DRDRAM was the memory of choice for the Hammer.
Entropy
30-Aug-2002, 15:46
Latency is important in context. That is to say, in the context of a P4 system where latency is well addressed and hidden rather well, it's a lesser issue -still important.
In the context of a P!!! such is not the case. I doubt the same would be the case for the Athlon and even possibly the Hammer, but the Hammer has the luxury of the integrated memory controller which would have made interesting possibilities with DRDRAM. Then again, my politically inclined side is going to start yelling if DRDRAM was the memory of choice for the Hammer.
The Alpha 21364 is designed with integrated DRDRAM memory controller, and offers (best case) latencies of 75 ns, almost identical to those of the Hammer.
Latency improvements of this magnitude are important. At 2.5 GHz, reducing the main memory latency from 160 to 80 ns is the difference between waiting 200 instead of 400 processor cycles. Even with very good cache hitrates, that will still have a major impact on the average access latency of the CPU(s).
Entropy
Deep buffers, prefetching, caching, speculative data loading and really high clock rates will help you overcome much of the latency problems. But to overcome them you'll need bandwidth, because you move data before you need it, not after, ideally it arrives right when you need it.
Entropy
30-Aug-2002, 23:24
Deep buffers, prefetching, caching, speculative data loading and really high clock rates will help you overcome much of the latency problems. But to overcome them you'll need bandwidth, because you move data before you need it, not after, ideally it arrives right when you need it.
"ideally"
Entropy
Fine, fine, "IDEALLY."
Nevertheless, we have to trade-off in different places to get desired performance. Of course, I'm a big believe in the right tool for the job and if it was upto me, I'd make multiple types of MPUs as functional units within the main MPU and have different memory types to handle different situations. But that's getting expensive. ;)
Entropy
31-Aug-2002, 00:15
Fine, fine, "IDEALLY."
Nevertheless, we have to trade-off in different places to get desired performance. Of course, I'm a big believe in the right tool for the job and if it was upto me, I'd make multiple types of MPUs as functional units within the main MPU and have different memory types to handle different situations. But that's getting expensive. ;)
:)
Yup. I've been fortunate in that scientific computing can afford and justify more specialized hardware, most of what is new in PC-space have been used for a long time in more expensive computers. Applications differ though, and microprocessors have had the edge in straight scalar non bandwidth dependent performance for some time. As I'm sure you are aware, all the features you bring up have been used for a long time in scientific/engineering computers, with much larger caches particularly relative to the core-main mem speed gap. The situation with PC CPUs is much much worse in terms of the CPUs outspeeding the memory systems that feed them.
So I'd consider latency to be quite important along with bandwidth.
It will be interesting to compare the Hammer with its 333MHz bus Athlon siblings, as it doesn't have all that much to gain in IPC over the Athlon according to AMDs presentations and nominal bandwidth will be identical. Most of the gain we see will be from the improved main memory latency, so it will give us some data to chew on pertaining to the relative importance of different parameters in the memory hierarchy, over a range of applications.
To me, this is known as "fun".
To others, the above sentence might justifiably be called "perverse". :)
Entropy
I think it's fun as well.
BTW, there should be significant IPC gains in the way of the FPU on the Hammer, since that unit is horribly imbalanced when you take into consideration the pipe that feeds the processor. An Athlonesque FPU on the P4 would be disgustingly strong. And I think the reason Intel really got rid of the second FPU pipe is because if would have spanked IA64 processors so bad, it's not funny.
To sign up for MPF in San Jose, go here
https://www.mdronline.com/corpreg/eventreg_router.asp?eur=eventreg_frm_new.asp&nur=e ventreg_frm_new.asp&npr=eventreg_frm_update.asp&mo de=MPF
To see the schedule, go here
http://www.mdronline.com/mpf/mpf-brochure.pdf
Entropy
What I meant to say was that I really don't think I'll see anything interesting from Micron. Maybe it will do well in some specialized market segment or something.
The nForces are 4 layer and run fairly well. Same with the i845s and so on. At least that's my understanding. I'm not sure where you're getting 6 layers from, unless we're heading into workstation and lowend server.
The nForces have stability problems. There are a number of DDR SDRAM boards that are six layer, IIRC.
If the Latency wouldn't be important RDRAM would have been much faster then SDRAM from the beginning. Most Benchmarks show that Latency is really important. Latency means that the CPU is idle for as long as it takes to feed the first data.
Which benchmarks? RIMM4200 can have quite a bit less latency than PC1066, but the gains are meager at best. The P4 masks latency fairly well. Unfortunately, it is hard to mask low BW. If the BW of RAM was anywhere close to the needs of the processor, latency would become more important.
the Interface of RDRAM runs at 533 MHz. RDRAM is DDR like normal DDR-SDRAM.
:roll:
BTW, there should be significant IPC gains in the way of the FPU on the Hammer, since that unit is horribly imbalanced when you take into consideration the pipe that feeds the processor. An Athlonesque FPU on the P4 would be disgustingly strong. And I think the reason Intel really got rid of the second FPU pipe is because if would have spanked IA64 processors so bad, it's not funny.
Have you seen the SPEC scores for the Itanium? The FPU scores are incredible. Even with a second FPU, I would have a tough time seeing the P4 score higher in SPEC.
Elmic,
You're kidding me right? Have you seen the throughput on the FPU in the P4? The extra pipe would have made a fair bit of difference, I'm sure there are enough operations in the SPEC tests that would benifit from the P4 not lagging behind due to a non-pipelined operation. Too bad I can't profile the code, but I'm pretty sure it would have done some serious damage to Itanium -it's beat it, but if the pipeline was there then it would have been a thrashing. Itanium 2 is a slightly different story, the reason here being a very large cache (which would help on the tests with large data sets) and double the system bandwidth of the P4. Even then, the P4 wouldn't have been that far behind with comprable performance. The Xeons with the extra cache (L3) might have been really powerful.
CFP2000:
http://www.spec.org/osg/cpu2000/results/res2002q3/
Athlon 2600+
Base: 655
Peak: 710
Itanium2 1GHz
Base: 1356
Peak: 1356
You would need dual P4 Xeons to beat the Itanium. The integer tests are much less impressive, however.
Why are you posting Athlon scores? I've already stated it's being held back by it's system bandwidth.
As for the P4, I believe the 2.8 scores in the high 800s. Too lazy to check.
mboeller
05-Sep-2002, 12:02
CFP2000:
http://www.spec.org/osg/cpu2000/results/res2002q3/
Athlon 2600+
Base: 655
Peak: 710
Itanium2 1GHz
Base: 1356
Peak: 1356
You would need dual P4 Xeons to beat the Itanium. The integer tests are much less impressive, however.
Pentium4 scores :
P4 2.8 GHz :
SPECint_2000 base : 1020 ( using PC1066 )
SPECfp_2000 base : 1041 (using PC1066 )
Crap, I didn't expect it to be that good. Heh.
Didn't see P4 scores where I was looking. Of course it is kind of silly to compare the two since their market segments won't overlap much.
Yeah, one is in an existing and the other in a non existing one :)
vBulletin® v3.8.6, Copyright ©2000-2013, Jelsoft Enterprises Ltd.