Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Register for free to join our community of investors and share your ideas. You will also get access to streaming quotes, interactive charts, trades, portfolio, live options flow and more tools.
Don’t forget to watch MTV this evening.
I don't have MTV unfortunately, but probably it will be plastered all over the media.
Also saw a rumour today that PS3 might ship in 2005 instead of 2006 and also saw another anti-rumour disclaiming this.
Seems the marketing wars are on:)
Cheers
Cor
Threejack: The future can be hard to predict...
Indeed, and it is very dangerous to try to bend that prediction into one particular direction
Btw, I don't think you should take "niche memory" as such a negative thing. As we have seen here, graphics memory is about 7% of the total memory market, so they qualify as niches memories. Still 7% is quite a lot of bucks (7% of $ 30 Billion...).
Even within these 7% many ordinary memory chips are now being used successfully (DDR a.o.). On the other hand there will be other applications than PCs which use high bandwidth memory. Whether they will dominate against PC memory? Hmmm, this has been thought before.
I don't think in investment decisions it is all that important to know exactly where all the product and revenue is going. As long as one can predict healthy growth overall, the Company is a BUY. Whether the stock price goes to $ 40 or $ 100 in a few years is not really al that important, unless you have a list of Companies which are going to do that growth and ave to make choices Anyway, some diversification is necessary, even if you think your favorite stock is going to the stars:)
Cheers
Cor
Paul
Maybe AMD will put XDR on its roadmap and give Intel something to worry about :)
It's always possible
But some info I saw today (posted at TMF) makes it likely that AMD has a programmable DDR/DDR2 memory controller in it's future line-up including the dual core chips.
Read some of that dual core stuff which came out of NDA today. Interesting is the quite large performance increases (at same clock speed, which AMD has managed for the x2 chips and Intel hasn't because of heat problems) of these x2 chips with the same memory interface as before. The benchmarks have been run with DDR400 against Intel's with DDR2-667 (at least some I have seen)
Makes one scratch one's heads about one's assumptions about those bandwidth needs caused by multicores, doesn't it?
Cheers
Cor
Paul
benchmarks for the dual-core AMD processors are out:
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2410
and
http://techreport.com/reviews/2005q2/athlon64-x2/index.x?pg=16
Enjoy
Cheers
Cor
Still not as fast as the 1066MHz FSB
We are not trying to get serious about the Pentium-EEs here are we?
They are an (expensive) joke.
Cheers
Cor
Hi Paul,
I have seen this debate, but Sony and Toshiba have plans to use XDR in HDTVs with Cell.
I have seen that claim too. It won't be just a HDTV TV chip that needs it. Maybe something advanced such as a HDTV with DVD burner, gaming platform etc. One $ 100 Cell would wipe out all the other costs on electronics on a HDTV
Sony also has plans for a home server using Cell.
We'll see when. The guy responsible for these ideas at Sony just got demoted.
True, so the real world advantage will be somewhere between the maximum numbers and the performance without it.
I wasn't referring to microthreading there but to XDR in general (bandwidth centric).
DDR2 is 600MHz</I>
DDR2 at 800 MHz is available now. As available as XDR
and:
Samsung shows their 6.4GHz XDR with a 32n prefetch. That can change if Core speed can increase..
Everything is possible. Faster cores are more expensive and consume more power. Don't know if that is the way to go.
Cheers
Cor
Hi Paul,
I think the microthreading is indeed an issue and it is for the PS3. This is where it originated no doubt. They have been workign with it (Nvidea, Sony) and have been finding that they are not getting the real bandwidth anywhere near the theoretical bandwidth with the memory due to two issues: latency and granularity.
You concentrate exclusively in your arguments on (max) bandwidth, but there are other parameters which have an effect on the overall performance.
So I think (just a supposition), that Nvidea/Sony would very much like to have microthreading added to XDR. Whether those who have to spend the money redesigning the chips, redoing the masks, documentation etc. (Samsung, Elpida, Toshiba) are willing to do it, remains to be seen.
(We will know for sure within about 6 months, the PS3 demonstrating in a week's time will have "ordinary" XDR which will be interesting enough
You say:
FSB speeds are increasing faster than memory bus speeds.
I assume you meant to say that cpu speeds are increasing faster. The FSB speeds are not increasing all that fast. (maybe they need a Flexio bus like Cell The question is whether for normal PC computing everything is not fast enough anyway as it is and they should not concentrate more on getting the power down and the cost lower.
(AMD has just launched a 1 Watt processor at 1 GHz, maybe that becomes the trend)
I imagine that there is a market for it though. That market will likely be Game consoles, servers, and HDTVs.
I don't think servers need it. below a link of a record sustained performance for a 2U server by Tyan; as you can calculate it's not the memory which holds it back, probably the disks. (sustained bandwidth for 48 hours
http://www.tyan.com/html/pr05_nna.html
The 6.5 gigabit data rate was measured as 96 client computers of Neal Nelson's Multi-Node Test Bed performed FTP "Gets" and "Puts" of multi-megabyte files. The client traffic was evenly spread over six separate gigabit ports in the server machine.
...
"Never before has a single 2U chassis run real application programs like FTP at sustained data rates that are this high," continued Nelson. "When this is viewed in the context of continuous operation for 48 hours and a total of 1 petabit of data transferred this is truly a record-setting accomplishment."
As for HDTV, most HDTV chips now use sdram and some DDR. They don't need the extra bandwidth. TV is actually quite slow. TV also wants to use the cheapest product possible. It's like the car industry, every cent is turned around a few times.
Workstations maybe. They could use the bandwidth, but also need low latency and large capacity memory. (for example for cracking large matrix operations to keep them in memory, think also of second order Laplace eqations etc that sort of thing)
I remember a few years ago somebody on the AMD board who desperately needed something which would support a memory of 2 GBytes, kit speeded up his calculations by several orders of magnitude.
Anyway I am waiting "with bated breath" for the workstation with Cell (and hence XDR), which IBM has apparently promised.
Cheers
Cor
Elixe, it is very simply and a first grader understands your statement that granularity equals bus width (in bytes or bits whichever metric you like most) times prefetch.
The fact that calbiker wants another formula (which may also be correct) involving core speed is just because he does not like your simple formula. Of course there are always derivative formulas but your:
granularity equals bus width times prefetch
is a simple FACT.
Also I find the whole discussion whether Intel would go for XDR with microthreading very theoretical. Now we speak not only of a memory type which is in low supply (XDR), but also of one which is still on the drawing board (XDR+microthreading).
I agree with you: it just is not going to happen in main PC memory in this decade (and maybe even not after that).
Cheers
Cor
Just ordinary DDR.
Ah, what I suspected.
BTW, I suspect the first company to create an XDR graphics solution for the PC will be nVidea.
Yeah, it won't be Intel
Maybe nVidea feels they need microthreading (because of the pesky prefetch and Rambus has to convince Samsung first to put it inside their XDR, but Samsung won't do it until Rambus gives them a good deal on the renewal of the contract which runs out in two months .....
Anyway only about a week to go to see the new graphics boxes. Will be interesting to see how many games they can show already.
Cheers
Cor
Wow, Intel needs to announce chips out "late in 2006", they must be panicking
Must say, that Charlie is quite capable of sarcasm below the surface (and sometimes sticking it's head up).
If Intel really needs that long (say 1.5 years) to get something which competes with the present AMD offering, maybe I should become an AMD fanboy again
AMD should be coming out with their new product before that probably. Now if they would get their production capacity up to what the customers want, they would finally make a profit again and I can buy the shares again, which I sold at $ 21.
Cheers
Cor
BTW, can I assume you’re still gathering inputs for your Rambus report?
Hey doesn't hurt to check
Thanks for the breakdown btw.
re:
From what I’ve seen, performance graphics memory represents about 4 - 5% of W/W DRAM revenues.
(About $1 billion this year).
I thought we were at a slightly higher number 7% in the last discussion. Or are you now separating "performance" graphics from the ordinary graphics? (which might be using DDRx rather than GDDRx).
I have a Radeon 9600 on this computer. Does that use GDDR or just ordinary DDR?
Cheers
Cor
meet friends for cinco de Mayo at a Mexican place.
Hey Paul,
5th of May is liberation day in Holland, mostly related to WW2 from the Germans and also the Japanese in Indonesia, but in recent years is is more like a general independence thing, like yours.
(Have to be careful today what I say, my wife is German
What is it in Mexico?
Luego
Cor
Hi Elixe,
From what I read I think you expect to be XDR a one product wonder for the first few years, say 2006-07, i.e. Cell only.
Is it true that you don't expect one of the major graphics guys (ATI, Nvidea) to switch from GDDR to XDR before 2008?
Does not seem too unreasonable to me, but I just like to verify.
There will I think, be some applications where a narrow bus or having one memory chip only is important or at least convenient. As it was with certain rdram applications.
Cheers
Cor
The parts aren't cheap.
Hey Cal,
I also looked up these parts. It is not clear whether they contain the Rambus PHY, as these are parts pre-existing to the announcement today.
There of course two Ramboid positive explanations
1) The original parts did not work well and are now going to be redesigned with a Rambus PHY
2) Rambus PHY was used and this is one of the licensees which declined to be mentioned in the past.
I am glad they show prices as there have been some ridiculous prices been bandied about by a certain elderly gentleman whose name starts with Ha... This makes it about $ 1 selling price per extra channel (38-26)/12, which is even lower than the channel prices I researched some years ago for RaSer.
Cheers
Cor
My last post here. Got better things to do.
Elixe, there are a lot of things better than posting, but I for one will miss you:( Thanks for a lot of great and informative posts.
And I don't intend to take over your sparring with "change the subject" Calbiker
Cheers
Cor
Hey Cal and Elixe,
I am of course some kind of nitwit in this discussion, but it seems to me that the prefetch has a deciding effect on latency.
Uh, maybe we were not talking about that, but wth the subject which is not posted here at the top of the posts tends to morph somewhat.
Cheers
Cor
The price slide is amazing, Elixe.
But apparently after the fantastic 2004 year for memory vendors we now have a bit of a lull and overcapacity.
Dramexchange still shows $ 6.94 for the 512 DDR2 part.
Gotto stick with one numbers supplier otherwise I go crazy:)
Cheers
Cor
Looking for Mr. Hughes to begin his dance with Wall Street by delivering .04-.05 in earnings.
Hey Threejack
If he did exceed by a few pennies after the updated guidance, Wall Street would tie a heavy anchor chain around his ankles and throw him overboard
My guess: between 3.4 and 4.3 cents...
Cheers
Cor
For somebody who wants to build a business around Cell, this Trounson does not seem to have much of an intimate contact with IBM.
Before venturing there, one would expect one to have technical information to some depth and face to face meetings with the ones selling Cell. (as I noted on TMF the Hypertransport manual has 1100 pages, that would be about the minimum]
Clearly Cell is not a part number yet at IBM.
Will be interesting to see if they produce a cheaper version with DDR
And then there is "Cell light" of course:))
http://boards.fool.com/Message.asp?mid=22293927
Cheers:)
Cor
I think this will be a tough year for DRAM. Capacity will exceed demand unless there is attrition in the makers. That never seems to happen, though.
Has happened in the past when the Japanese more or less bailed.
I saw a recent report from people connected to IFX that yields on DDR2 would now be about 50% and on DDR about 75%. DDR2 probably also has a small die penalty, but negligible. I agree MHz come free, except when they are from Rambus
So if the changeover from DDR to DDR2 goes faster than so far, there might be shortages of the desired types/speeds.
Cheers
Cor
Elixe: That would put graphics devices at ~6% revenue share.
That seems low if the bit share is also 6% that would mean they would command no premium on the asp??
(I thought I was the lowballer here:)
ASP of $4.30. About the same price as a 512Mb DDR2 for the year.
Means you are expecting a very steep ramp down for DDR2. Right now, at the beginning of qtr2 we have an asp of $ 8.18 (512 Mb, DDR2-533, spot); in edit today $ 7.98 that was 2 days ago;)
Spot for 256 Mb DDR-400 $ 2.66.
Cheers
Cor
Hi Elixe
I had seen this presentation too.
So for the future they see about 5% for graphics chips and 1% for game consoles (additional?) for a total of 6% in terms of bits.
What would that be in terms of dollars?
If we assume GDDR to be twice the price of DDR (this may be too low) it would be approx. 10% of the dram revenue market.
Do you agree?
Cheers
Cor
No need for Hypertransport because PCIe is the evolving standard; sorry, Cor. Mr. Kumar thinks embedded graphics will be good enough for mainstream PCs.
Hi Threejack,
This is what they say at Intel, of course. NIH! (not invented here)
Meanwhile there is (has been for some time) Hypertransport inside about 20% of PCs, a higher percentage of network processors, all Apples etc...
Als Hypertransport is enabling the interconnection of 2-way and 4-way AMD processors without any additional chips.
Of course after 5 years one can design a better bus, but I can tell you, if Hypertransport had been Rambus' there would have been major royalty income from this
Cheers
Cor
Rambus found the tapes sometime prior to March 17, the date it informed Outside Counsel;
That's 4 days before the Infineon deal was announced.
If one were a suspicious character, I would say that that explains the haste and terms of the IFX deal
As far as I remember (from various briefs) there were all sorts of missing emails requested by parties. Maybe these are there. The spooky thing is that the period just before the start of the IFX trial is covered, that time in 1998/99 where Payne ruled spoliation had occurred in a planned manner leading up to litigation.
(March 17, 2004- 220 F.R.D. 264 (E.D. Va. 2004), p. 287)
I think settlement before the 11th conf., but I have been wrong before
Cheers
Cor
Hey, I also have old backup tapes in all sorts of formats in my garage:) Even some on fullsize half inch 1600 bpi magtape...
This will almost certainly delay things for several months, don't you think?
Cheers
Cor
Sorry, Cor, but there isn’t much available for Spain.
I know, I tried it a few months ago when Google acquired the keyhole guys. It's there but not at the good resolution.
Cheers
Cor
For example, in an interactive shoot-em-up what would fall under that category?
Hey Elixe,
Not a gaming expert, but it would seem that a game with curvacious females would need lots of little triangles
Cheers
Cor
Denali memory report (March 05) is out.
(needs registration)
Some discussion of "pure play IP" mddels, see for an excerpt:
http://www.semireview.com/info/viewtopic.php?t=15
Also quite a bit about future adoption of FB-DIMM solutions (outside Intel
Cheers
Cor
XDR pricing.
I think we should have a concerted effort to determine at which price XDR is being offered as compared to GDDRx and DDR2.
I have tried to sind some price information, emailed with several (Samsung and Elpida), but have not found anything other than the one article which said the Elpida 512 Mb XDR chip was sampling at $ 50/pc.
Here is the text I got back to a multiquestion email I sent to Elpida IR:
...
Secondly, with regard to price of the XDR,
we are sorry that we don't disclose prices by product.
We are currently sampling the XDR.
We will set the price by considering the needs as a result of sampling.
Hopefully that would be helpful for you....
Cheers
Cor
Hi Elixe,
On the triangles, would like to know more, so you estimated a full triangle (one not "stripped") was 96 bytes, i.e. 32 bytes per vertex?
Could it be (just a hunch), that Sony is running into this (efficiency) problem with the Cell games development and hence it got to Rambus and back and forth etc?
Cheers
Cor
Wonder how long before some benchmark tests are available?
Threejack, it would be nice to have some benchmarks anyway of XDR in any form. These do not exist afaik.
Cheers
Cor
Something we agree on is that Eulau needs some grooming
And he has to do his job better, like the 15% error in guidance for qtr4 for the tax rate was inexcusable and has never been explained.
We'll probably see next weke how things pan out.
Cheers
Cor
IMO it's intended for XDR.
Yes I understood that Elixe.
1) There is more advantage for XDR
2) XDR probably has much of the muxing logic already in place
3) GDDR needs it less and the MMs will not be convinced to do it easily (for a smaller benefit, more types etc, working on GDDR4 anyway etc)
Cheers
Cor
and the next gen XDR will read 8x the data.
Thanks Elixe,
So the prefetch is going to be 32? Is this from a certain speed onwards, where the core cannot keep up again?
Cheers
Cor
That's why it will be interesting to watch Mr. Hughes operate. Think he lowered expectations either to take the sting out or to deliver a small surprise. He succeeded in lowering your expectations, anyway ; )
Hey Threejack,
On previous occasions when you called me a lowballer I was right
On previous occasions when nobody (except Patrick Newell) accused me of lowballing, I was much too optimistic
Looked at that PR again:
http://www.rambus.com/news/newsroom/pressrelease.cfm?id=180
snip:
it expects revenue to be approximately $37 and $38 million and total spending, which includes cost of services and direct expenses, to be between $33 and $36 million for the first quarter of 2005. The guidance given in January 2005 was for revenue to be in the range of $38 to $42 million and total spending to be in the range of $30 to $34 million. The increased range in total spending is driven by higher than expected litigation costs. In addition, the company expects that net interest and other income will now be in the range of $2.1 and $2.3 million.
So it turned out I stuck one wrong number into my spreadsheet and the expectation should really be 4.4 cents/share!
So sorry if I spooked you. Not that 4.4 is fantastic, but it is not as gloomy as I thought. (that's what you get if nobody checks my numbers:(
Cheers
Cor
PS based on 109 M diluted shares, if these are higher because of teh convertible the number is a bit lower.
With guidance lowered a couple weeks ago, earnings announcement in a couple weeks should be ho-hum unless there is a surprise. Will be interesting to watch how Mr. Hughes operates.
Hey Threejack
That earnings day (soon I guess) is not going to be good unless there is some stellar guidance or maybe a good-bad-good sandwich with a settlement in one of the outside pieces.
The guidance given recently points at about 2.5 cts/shr for the qtr. (there is a mysterious decrease of revenue from the guidance given in jan, we haven't quite figured out what that is, could be Samsung trouble)
The problem is that the analysts which are counted by WS have not reduced guidance sufficiently, see:
http://finance.yahoo.com/q/ae?s=RMBS
5 cents average, 3 analysts, low 3 high 7 well you can compute the other one
30 days ago it was 6 cents average.
this is still too high so marketwatch etc 's headlines after the earnings announcement will read something like "Rambus misses by 2-3 cents". Not good. (will present a good buying opportunity for everybody who is not maxed out yet though!)
Cheers
Cor
A bit further into understanding this "microthreading" pse:
Current XDR would prefetch 512 bytes. (256 bytes on a 128 bit bus).
Does this mean, that by splitting (or multiplexing) the address bus into, say, four 32 bit wide memory busses we could get four separate triangles simultaneously from the four side by side parts of the memory? (assuming of course that the software is smart enough to put them spread out properly)
Or are we talking about a changed core of the memory chips as well?
How does NVIDEA do this now? It seems they could have four separate memory channels, each 32 bits wide anyway...
Cheers
Cor
PS Cal thks for the formula. I see you are preparing for all sorts of weather now that you know you cannot trust the 30 year averages any more
Rambus Unveils Micro-Threading In DRAM Cores
http://www.rambus.com/news/newsroom/pressrelease.cfm?id=181
Delivers up to 4x performance improvement in 3D graphics and other applications
LOS ALTOS, Calif. - April 4, 2005 - Rambus Inc. (Nasdaq:RMBS), one of the world's premier technology licensing companies specializing in high-speed chip interfaces, today unveiled an architectural breakthrough of applying micro-threading to DRAM cores. This innovation significantly increases memory subsystem efficiency, resulting in up to four times greater performance when compared to a traditional DRAM in applications such as 3D graphics, advanced video imaging, and network routing and switching.
Micro-threading increases memory system efficiency by enabling DRAMs to provide more usable data bandwidth to requesting memory controllers. A single core operation of a typical mainstream DRAM provides a larger amount of data than needed by many applications. As a result, large amounts of memory bandwidth are used to deliver a small amount of relevant data. Micro-threading enables the DRAM to provide several smaller relevant pieces of data in place of a single larger piece of data, resulting in higher memory bandwidth efficiency while minimizing power consumption.
"By applying micro-threading to a DRAM core, we are continuing our tradition of innovation by designing and developing advanced technologies to improve the bandwidth between DRAMs and their associated memory controllers," said Laura Stark, vice president of Platform Solutions at Rambus. "We look forward to working with our various DRAM partners to bring this exciting new technology to the market in high-volume applications."
With the application of micro-threading to a DRAM core, separate addresses are provided to different DRAM core partitions, enabling the requesting controller to generate multiple micro-RAS and micro-CAS operations in the same time it would take to generate a single RAS or CAS command to a standard DRAM. Simultaneously accessible banks allow for concurrent retrieval of data, which are then bundled into a single transmission.
Realistic interactive 3D applications such as games and visualization software require more complex scenes, composed of increasingly smaller textured polygons or triangles. A Rambus analysis showed that a standard GDDR SDRAM being used in a 3D application can deliver between 50 and 125 million triangles per second. If the same GDDR SDRAM were to be enhanced with micro-threading, the rate of delivered triangles would increase to between 100 and 500 million triangles per second. By enabling higher triangle rendering rates, micro-threading, as it is applied to a DRAM core, benefits the end user experience by providing richer visuals.
Micro-threading may be applied to existing DRAM cores with relatively low incremental cost. To benefit from the performance increase, DRAM controllers interfacing with micro-threaded DRAMs need to be optimized for the new technology. This patent pending micro-threading technology is available for licensing today. For more information on micro-threading and other Rambus innovations, please visit www.rambus.com.
(how many bits/bytes is a "triangle"?)
Cheers
Cor
I looked in and it appears to me that Cryptscript is a good guy.
Agree, Elixe. Another relatively but active and tech savvy poster is scfiller (Scott). Between the three of us we keep some tech threads going. If you joined it would be even better
Btw did you see the new PR from rambus about "micro threading" inside drams? We are still figuring out what it means exactly, a pdf from Rambus would be in order, I think.
A bit analogous to "staircase seeking" on hard disks, I believe, see:
http://boards.fool.com/Message.asp?mid=22306556
I'll post the PR in the next post, saves Threejack the trouble.
Cheers
Cor
Hi Elixe, there is indeed more info in that original post of yours than I remembered, so I have corrected and combined with a reply from Crypt where you were accused of the classic Rambus bashing trick ...
see my new post at:
http://boards.fool.com/Message.asp?mid=22305033
Cheers
Cor