Is a Rambus Comeback in the Future?

ApesMa · Feb 27, 2002

The answer to the rhetorical question in a recent column here appears to be "No." /. references Anandtech who quotes EBN, which writes: "Intel Corp. in the second half of this year will drop its final Direct Rambus DRAMs support in new computer products, it was learned Tuesday at the Intel Developers Forum."

FatBurger · Feb 27, 2002

Well I heard from my uncle's cousin's wife's barber that the nephew's sister's cat's friend's owner's mother talked to the CEO of Intel's son's girlfriend's sister's babysitter that Intel is going back to EDO.

Intel is using DDR for it's Xeon servers at the end of the year, that is the only base for all of the wild claims flying around the net.

Quarter Pounder Inside
Don't step in the sarcasm!

slvr_phoenix · Feb 27, 2002

ROFL @ FatBurger!

Hey, I have an old P1 that actually uses EDO.

Hell, for that matter I still have a C=64.

I lost my TI-85 though. That still depresses me to think about. I never did manage to perfect my 'Breat Out' clone on it, but I was so close.

<pre>AROOOOOOOOOOOOOOOOOOOOOOO!</pre>

redstar · Feb 28, 2002

i just read the article that says maybe tomshardware was wrong to bash RDram in the past. RDram is the future afterall so they now think.

My mind boggles at this lightening turnaround.

This site has gone from calling the RD people the anti- christ to labelling them future saviour of P4 tech.

It reminds me of their review of the MSI kt266 pro-r MB.
At first they were all aglow with praise and oohing over the obvious quality. Then, after i buy the darned board, they diss the old board and praise the new kt266A chipset.

I have always viewed tomshardware as an independent and credible source.

But, it seems i may just have to re-assess that view and do a 180 just like tom!

ath0mps0 · Feb 28, 2002

IMNSHO, Intel will do what ever makes financial sense.

They can't release server boards for RDRAM because they need to be able to support up to 4GB of RAM and this is not currently possible as 512MB modules are currently the max (something of a connundrum if RDRAM is the "future"). Outfitting a motherboard with 4GB would require 8 RIMM slots and this is not currently possible or advisable because:
1) This would increase the trace length to exceed specifications or require exotic placement of the RIMM slots causing huge motherboard design headaches.
2) This would effectively quadruple latency - as RDRAM latency is the product of the latency of the slowest module times the number of modules in a channel.

Please note that the new 32bit RIMMS do not solve this problem; they are simply double-sided 16bit RIMMs - already a very common configuration in the SDRAM market; they will only help by allowing a user to install one at a time and taking up less space on the motherboard. The only thing that can solve this problem is a higher density RDRAM chip module. If higher density RIMMs become widely available, Intel may later release a RDRAM server chipset.

Current desktop motherboards only have two slots per channel (or 2GB limit with current 512MB RIMMS) so latency is only doubled in comparison when adding extra RAM. RDRAM currently only makes sense in short bus scenarios. Rambus is looking at widening the 16bit bus to 64bits to compensate for the degradation in latency that occurs in long/narrow busses; it has yet to be seen, however, whether or not this will lead to excessive heat and/or clock ramping issues. Prior to widening the bus Rambus will have to solve the current problem of RDRAM chip module density; they will have to double and then quadruple the density to remain competetive.

With current DDR-333 bandwidth already 66% faster than current shipping RDRAM PC800 bandwidth and 1GB DIMMs commonly available, Intel made a business decision with its server chipset.

The question that no one - not even Intel - seems willing to answer definitively, is whether or not Intel will take action to supplant their current P4 performance leading i850/i860 desktop and workstation chipsets with an equivalent performing dual-channel DDR version. Despite all of the arguements to the contrary, several vendors have shown that the physical limitations of a 128bit bus can be relatively easily overcome - the e7500 even has 6 DIMM slots. Current Intel marketing dogma says that RDRAM is faster for the P4 - although true in current shipping desktop & workstation configurations, not true in the purest sense; the server market will set the matter straight.

I think that Intel will continue to support RDRAM until it is no longer financially beneficial to them (and the same goes for any technology - including DDR). With RDRAM and the Rambus "alliance," Intel attempted to gain control of the CPU/RAM question, a battle that they arguably lost. Since it is not likely that they will take up the battle again soon - at least not in the open, it is anyone's guess where we will be in 2 years time.

I thought a thought, but the thought I thought wasn't the thought I thought I had thought.

FatBurger · Feb 28, 2002

2) This would effectively quadruple latency - as RDRAM latency is the product of the latency of the slowest module times the number of modules in a channel.

Do you have proof that the latency multiplies with the number of modules? I know it goes up, but I have a very hard time believing that it's multiplied.

Please note that the new 32bit RIMMS do not solve this problem; they are simply double-sided 16bit RIMMs -

No they're not. I have double-sided 16-bit RIMMs sitting in my computer right now.

it has yet to be seen, however, whether or not this will lead to excessive heat and/or clock ramping issues.

I doubt it will, as simply adding more pins shouldn't affect the internal workings of the chips. In fact, if it gets spread out as a result, it's possible that they would run cooler. But you could be right, we'll have to wait and see.

With current DDR-333 bandwidth already 66% faster than current shipping RDRAM PC800 bandwidth

Hmm...2.7GB/s vs. 3.2GB/s. Nope, RDRAM has more bandwidth. Sorry.

is whether or not Intel will take action to supplant their current P4 performance leading i850/i860 desktop and workstation chipsets with an equivalent performing dual-channel DDR version.

It would be much easier and more economical to simply introduce dual-channel using 32-bit modules, but if Rambus drags their feet introducing higher pin count memory, then Intel may have no choice. Rambus appears to be doing just that, with no new memory technology introduced in quite a while (excluding the 32-bit modules shown yesterday at IDF, which of course haven't been released to the public).

Quarter Pounder Inside
Don't step in the sarcasm!

ath0mps0 · Mar 1, 2002

Do you have proof that the latency multiplies with the number of modules? I know it goes up, but I have a very hard time believing that it's multiplied.

It is simple mathmatics and part of the RDRAM spec. RDRAM is a serial memory technology, so any signal to the higher memory addresses has to traverse the other modules. The RDRAM spec requires pre-negotiation that limits EVERY call to the speed of the slowest possible call.

<A HREF="http://www.realworldtech.com/page.cfm?section=news&AID=RWT110799000000&p=3" target="_new">To put the "random access" back into a DRDRAM-based memory system, Rambus Inc. designed into each memory chip the capability of delaying the output of read data onto the channel beyond the normal 20 ns page read access latency by a programmed amount of 2.5, 5.0, 7.5, or 10.0 ns using the TPARM control register. When a DRDRAM-based computer system is powered-on or reset, the processor and memory controller ASIC perform an elaborate initialization ritual for each DRDRAM in the system. As part of this effort the read round trip delay for each memory device is measured and the longest delay is determined. Then the processor and/or ASIC attempt to equalize the round trip read access time for all devices by programming extra read delays into DRDRAMs closest to the ASIC. The net result is all DRDRAM devices appear as equally slow as the farthest device.</A>

Please note that the new 32bit RIMMS do not solve this problem; they are simply double-sided 16bit RIMMs -

--------------------------------------------------------------------------------

No they're not. I have double-sided 16-bit RIMMs sitting in my computer right now.

Current double sided RIMMs do not function like current double sided SDRAM (DDR or SDR)DIMMs. Double sided DIMMs are usually dual bank; Current double sided RIMMs are still serial. The upcoming 32bit RIMMs more closely match current double sided DIMM technology; you get a full bank from each side. In the case of RIMM technology this means 16bit on each side (16 + 16 =32)<A HREF="http://www.anandtech.com/showdoc.html?i=1590&p=5" target="_new">The board layout is relatively simple, a single 16-bit RDRAM channel is routed to one side of the RIMM slots while another channel is routed to the opposite side of the slots.</A>

I doubt it will, as simply adding more pins shouldn't affect the internal workings of the chips. In fact, if it gets spread out as a result, it's possible that they would run cooler. But you could be right, we'll have to wait and see.

Yeah, we will have to wait and see. The issue that causes more heat is more power to drive higher densities of transistors (what is required to develop higher density RDRAM chip modules). Unless the RDRAM process is drastically scaled down (SOI and/or .09 micron, etc.) the heat generated by the higher density chip modules (a RIMM is made up of chips that are in and of themselves modules) will require active cooling.

Hmm...2.7GB/s vs. 3.2GB/s. Nope, RDRAM has more bandwidth. Sorry.

<A HREF="http://www6.tomshardware.com/mainboard/00q1/000315/rambus-01.html" target="_new">See Table</A>

Again, do the math: PC2700 DDR333 has 2.667GB/s bandwidth per stick (333.333MHz * 8bytes = 2666.667MB/s). PC800 RDRAM has 1.6GB/s bandwidth per stick (800MHz * 2bytes = 1600MB/s). DDR SDRAM has more bandwidth. OTOH, as I stated above, RDRAM chipsets (i850, i860) double that bandwidth by combining two channels into one (1.6 * 2 = 3.2GB/s). The extra bandwidth is due to the chipset, not the memory technology used. The Intel e7500 chipset does the same thing with older PC1600 DDR200 SDRAM. So, DDR technology currently offers 66% more (2.667GB/s vs. 1.6GB/s) bandwidth.

<A HREF="http://www.anandtech.com/chipsets/showdoc.html?i=1588&p=3" target="_new">The memory controller in the E7500 is validated for use with both DDR200 and DDR266 SDRAM however the bus will only operate at 100MHz (DDR200 speeds). This means that although you can use DDR266 SDRAM in it, your memory will always run at DDR200 speeds. Intel's reasoning behind this that dual DDR200 channels yield a theoretical 3.2GB/s of bandwidth to main memory which is perfectly matched up to the 3.2GB/s FSB. As we've seen in the past (take the KT133A chipset for example), a synchronized FSB and memory bus generally yields lower latency CPU/memory accesses than an asynchronous setup. It is very clear however that when Intel does move to a 133MHz (533MHz quad-pumped) FSB, a future successor to the E7500 chipset will support DDR266 SDRAM.</A>

It would be much easier and more economical to simply introduce dual-channel using 32-bit modules

For Intel, Rambus and Samsung, maybe. But for consumers, when 32bit PC1066 * 2 (4.2GB/s) modules finally ship, how much will it cost? We can't even buy PC1066 yet and if Rambus tradition holds, it will cost an arm and a leg. Current DDR333 would already provide 25% greater bandwidth in comparable dual-channel configurations (5.3GB/s). DDRII will also probably be available around this timeframe.

<A HREF="http://www.ee.umd.edu/~blj/papers/memwall2000.pdf" target="_new">http://www.ee.umd.edu/~blj/papers/memwall2000.pdf</A>

So, I stand by my statement: it is nearly impossible to tell what memory technology will be in the forefront in 2 years time.

I thought a thought, but the thought I thought wasn't the thought I thought I had thought.

bum_jcrules · Mar 1, 2002

I still have my Commedore 64 and my Commedore Vic 20. Donkey Kong still looks great on it.

"Taurelilomea-tumbalemorna Tumbaletaurea Lomeanor" - Treebeard :lol:

FatBurger · Mar 1, 2002

Do you have actual latency test results, though? I'm just trying to make sure that practice matches up with theory, since often it doesn't.

The issue that causes more heat is more power to drive higher densities of transistors (what is required to develop higher density RDRAM chip modules).

True, I hadn't thought about higher power requirements.

Again, do the math: PC2700 DDR333 has 2.667GB/s bandwidth per stick

Per stick? How about we pick something even more arbitrary and meaningless? How about bandwidth per watt of heat created? Or bandwidth per BMW owned by the executives of the manufacturer?

how much will it cost? We can't even buy PC1066 yet and if Rambus tradition holds, it will cost an arm and a leg.

I highly doubt that. RDRAM used to cost more than it's weight in gold, and dropped significantly by the time PC800 came out. Nobody would buy PC1066 if it cost too much, it should release within 25% of PC800 prices.

Current DDR333 would already provide 25% greater bandwidth in comparable dual-channel configurations (5.3GB/s).

And would also take up four times as much space on the motherboard as current RDRAM configurations. Of course, you're also assuming that there will be CPUs that can use that much bandwidth (or would you simply split the bandwidth, like on nForce?). For a P4 platform, that may work. Definitely not for an Athlon system (unless they up the bus to 166, like most people hope).

So, I stand by my statement: it is nearly impossible to tell what memory technology will be in the forefront in 2 years time.

Oh, I completely agree. We've seen in the past year or two that consumers have a huge say in that (why else would Intel release the i845?), and since so many people hate Rambus the company, RDRAM's development may be choked simply by lack of sales, for one example.

Quarter Pounder Inside
Don't step in the sarcasm!

slvr_phoenix · Mar 1, 2002

and since so many people hate Rambus the company, RDRAM's development may be choked simply by lack of sales, for one example.

I'm surprised that no DDR SDRAM zealot has as of yet gone and blown up Rambus HQ like some religious zealots have done to abortion clinics.

<pre>AROOOOOOOOOOOOOOOOOOOOOOO!</pre>

bum_jcrules · Mar 1, 2002

Since you like using "quote" I will too.

Again, do the math: PC2700 DDR333 has 2.667GB/s bandwidth per stick (333.333MHz * 8bytes = 2666.667MB/s). PC800 RDRAM has 1.6GB/s bandwidth per stick (800MHz * 2bytes = 1600MB/s). DDR SDRAM has more bandwidth.

Your math is incorrect.

PC2700 is DDR 166MHz. The end result was correct but the calculation is incorrect. 166MHz x 2(DDR) x 8bits = 2.672Gb/s

Now saying that DDR has "more" bandwidth is correct in theory, but not in the "real world." RDRAM is more efficient when it comes to "actual throughput" when looking at "theoretical vs. actual."

"Taurelilomea-tumbalemorna Tumbaletaurea Lomeanor" - Treebeard :lol:

Guest · Mar 2, 2002

if RDRAM is so damn good, then why did Rambus need a special agreement with Intel to protect it from DDR when Pentium 4 was first released? trying to deny competition with sneaky tricks like that does not gain my trust. what i dont trust, i dont buy. i think a lot of other people feel the same, and that is all there is to it.

rickd59 · Mar 2, 2002

If DDR prices keep skyrocketing (mind you, it costs nothing to make), they're going to piss off someone, and keep the gate open for RDRAM. Someone better get them back into check.

-Rick

ath0mps0 · Mar 2, 2002

Do you have actual latency test results, though? I'm just trying to make sure that practice matches up with theory, since often it doesn't.

No, but Rambus does - try prying them from their cold, dead fingers....

In reply to:
--------------------------------------------------------------------------------

Again, do the math: PC2700 DDR333 has 2.667GB/s bandwidth per stick

--------------------------------------------------------------------------------

Per stick? How about we pick something even more arbitrary and meaningless? How about bandwidth per watt of heat created? Or bandwidth per BMW owned by the executives of the manufacturer?

'cause then it would be something arbitrary, like say, the things you listed...hehehe :tongue: Seriously, call it arbitrary if you like, but we are talking about existing technology here; existing RDRAM is 16bit - future RDRAM chip modules are still going to be 16bit - they are just planning on effectively "slapping" two RIMMS back to back and calling it a 32bit module. Measuring memory bandwidth stick by stick (or RIMM vs DIMM) is the only true and valid measurement one can take to answer the question "which one is faster?" Any other mesurement relies on other factors - all of which would have to be equal or equivalent to answer the question truely and scientifically.

Don't get me wrong here, I do know that current i850/i860 chipsets outperform current i845D chipsets - but that's the chipset, not the RAM.

Nobody would buy PC1066 if it cost too much, it should release within 25% of PC800 prices.

We'll see - I think that it will be closer to 33-50%. Also, I was refering to 32bit (PC1066 * 2) for the arm and a leg part.

In reply to:
--------------------------------------------------------------------------------

Current DDR333 would already provide 25% greater bandwidth in comparable dual-channel configurations (5.3GB/s).

--------------------------------------------------------------------------------

And would also take up four times as much space on the motherboard as current RDRAM configurations. Of course, you're also assuming that there will be CPUs that can use that much bandwidth (or would you simply split the bandwidth, like on nForce?). For a P4 platform, that may work. Definitely not for an Athlon system (unless they up the bus to 166, like most people hope).

Actually, it would only take up more space for the traces - something already proven doable and feasable - with the nForce (3 DDR slots with room for a forth) and the Intel e7500 P4 Xeon server board (6 DDR slots - two more than any RDRAM based i850 mobo). The slots themselves are no bigger than RIMM slots. I'm specifically talking about the P4 here - It's the only platform that uses RDRAM. The Athlon has no need for dual-channel; just like with the P3, the extra bandwidth goes to waste. Athlon w/166MHz FSB will be paired synchronously with single-channel DDR333 which perfectly matches its needs, just like dual-channel PC800 or PC1600 (DDR200) synchronously match with the 100MHz(400 effective) and dual-channel PC1066 or PC2100 (DDR266) will synchronously match with the 133MHz (533 effective) FSB of the P4 desktop and P4 Xeon server chips respectively.

I thought a thought, but the thought I thought wasn't the thought I thought I had thought.

ath0mps0 · Mar 2, 2002

I will double quote you:

Since you like using "quote" I will too.

In reply to:
--------------------------------------------------------------------------------

Again, do the math: PC2700 DDR333 has 2.667GB/s bandwidth per stick (333.333MHz * 8bytes = 2666.667MB/s). PC800 RDRAM has 1.6GB/s bandwidth per stick (800MHz * 2bytes = 1600MB/s). DDR SDRAM has more bandwidth.

--------------------------------------------------------------------------------

Your math is incorrect.

PC2700 is DDR 166MHz. The end result was correct but the calculation is incorrect. 166MHz x 2(DDR) x 8bits = 2.672Gb/s

Now saying that DDR has "more" bandwidth is correct in theory, but not in the "real world." RDRAM is more efficient when it comes to "actual throughput" when looking at "theoretical vs. actual."

I don't know where you went to school, Bum - sounds like maybe not at all...but you need to check your math. The number you gave was for 167MHz (NOT 166MHz) * 2(DDR) * 8bits = 2.672Gb/s - no existing memory spec I know of. The DDR multiplier was a given in my calculations - for both RDRAM and DDR - as they both are double-data-rate specs.

The actual 166Mhz spec is 166 2/3 or 166.6666666666666666666666666 etc. just like 133MHz really is 133.3333333333333333333333333 ad infinitum. 166.6666666666666666666666666MHz * 2(DDR) * 8bytes = 2.6666666666666666666GB/s as I previously stated (but only to 3 decimal places.

Also, you stated "Gb/s" which is gigabits per second - not "GB/s" which is gigbytes per second - and the real answer, in that case, would have been 166.667 * 2(DDR) * 64bits = 21,333.333 Gb/s as DDR DIMM are 64bits wide and are rated in "bytes" not "bits."

On the question of Theoretical vs. Actual, I would have to say that DDR has proven itself just as capable of using its theoretical bandwidth as RDRAM. In <A HREF="http://www6.tomshardware.com/cpu/02q1/020225/p42666-11.html" target="_new">Tom's latest tests</A> the P4 was, at most, able to use 78.5% of the available RDRAM bandwidth (3346/4266 and 2518/3200). The Athlon was able to use 97.7% of the DDR bandwidth (2080/2133). Plus, DDR latency is lower than RDRAM latency (by almost half).

These tests seem to indicate that if anything, DDR is more efficient.

I thought a thought, but the thought I thought wasn't the thought I thought I had thought.

FatBurger · Mar 4, 2002

Measuring memory bandwidth stick by stick (or RIMM vs DIMM) is the only true and valid measurement one can take to answer the question "which one is faster?"

No, it's not. Because there aren't really any single-channel RDRAM platforms or dual-channel DDR platforms. Measuring by stick means nothing.
Measure either total bandwidth available in a platform, or bandwidth per pin. Measuring 16-bit RDRAM against 64-bit SDRAM might show what current platforms consist of, but it falls far, far short of showing what the differing technologies are capable of.

Quarter Pounder Inside
Don't step in the sarcasm!

Guest · Mar 5, 2002

I have seen at least two companies, Dell and Colfax international offering 4 GB of RDRAM. Why do you say that it isn't possible?

FatBurger · Mar 5, 2002

Can you provide a link?

Quarter Pounder Inside
Don't step in the sarcasm!

Guest · Mar 5, 2002

Check these:

http://www.dell.com/us/en/hied/products/model_precn_precn_530.htm

http://www.colfax-intl.com/NHome.html

FatBurger · Mar 5, 2002

Thanks for the link, they use a riser to provide 8 RIMMs.

Quarter Pounder Inside
Don't step in the sarcasm!

Guest · Mar 5, 2002

I don't see any mention of risers for this motherboard, unless I just missed it.

http://www.tyan.com/products/html/thunderi860.html

FatBurger · Mar 5, 2002

From the Dell link, configure the PC and pull down the RAM menu to where it says 4GB. It says "riser" in that option, and I think one or two others as well. I didn't check the other link, I assume it's the same.

Quarter Pounder Inside
Don't step in the sarcasm!

Guest · Mar 5, 2002

I don't think it's the same. At least not that I can tell.

FatBurger · Mar 5, 2002

The Colfax one has 8 RIMMs, so it must use a riser.

Did you even look at the Tyan board? The second picture is of a RIMM riser.

Quarter Pounder Inside
Don't step in the sarcasm!

ath0mps0 · Mar 5, 2002

Because there aren't really any single-channel RDRAM platforms or dual-channel DDR platforms.

The i820 and i840 are both examples of single-channel RDRAM platforms. The nForce Athlon platform and the e7500 and Serverworks GCLE P4 Xeon platforms are all examples of dual-channel DDR platforms. Isn't it strange that none of them support the P4 desktop platform? Dual-channel (and quad-channel - GCHE) DDR platforms are the highest performing memory platforms available - with the notable exception of the nForce (the only non P4 platform listed).

Intel is obviously avoiding using the higher performing DDR dual-channel on P4 desktop and it has nothing to to with technological limitations.

Measure either total bandwidth available in a platform, or bandwidth per pin.

LOL

People don't buy pins of RAM, they buy sticks. Bandwidth per pin is only a valid comparison with similar technology; RDRAM is serial and DDR-SDRAM is parallel. That's like saying that Serial ATA 133 is higher performing than UDMA 133 because Serial ATA only takes 4 pins while UDMA 133 takes 40. Serial ATA may scale higher than UDMA because it has fewer pins, but there aren't yet any platforms to prove it. RDRAM should have the same benefits, but it hasn't yet been proven. DDR sticks and Dual-channel DDR platforms already outperform dual-channel RDRAM in pure memory bandwidth. Oh, and about the arguement of 1GB RDRAM sticks and 2GB maximum, multi-channel DDR platforms offer up to 16GB.

This is obviously (unless you're wearing Intel/Rambus blinders) a political issue.

We'll just have to wait and see how long Intel holds out.

I thought a thought, but the thought I thought wasn't the thought I thought I had thought.

Is a Rambus Comeback in the Future?

Distinguished

Illustrious

Splendid

Distinguished

Distinguished

Illustrious

Distinguished

Distinguished

Illustrious

Splendid

Distinguished

Guest

Guest

Distinguished

Distinguished

Distinguished

Illustrious

Guest

Guest

Illustrious

Guest

Guest

Illustrious

Guest

Guest

Illustrious

Guest

Guest

Illustrious

Distinguished

Share this page