Second-Gen DDR SDRAM On The Horizon

cplcap writes "This story in The Register picks up on Samsung's new DDR-II Chips, pushing DDR's speed up to 533 Mb/s and a 4.2GB/s memory bus. Prototype 512MB DIMMs are being produced, and IBM has developed a chipset to take advantage of the speed. There's a little more meat in Samsung's official press release."

Comments

the big question is...

[Lumpy]

Is it a whole new form factor so everyone had to redesign the motherboards and to force incompatability with older systems??

This is important because industrial and corperate-mission-critical is older equipment. and an upgrade path for ram is still important.

Re:the big question is...

[ToLu the Happy Furby]

Is it a whole new form factor so everyone had to redesign the motherboards and to force incompatability with older systems??

The answer is...yes and no. DDR-II does have a new form factor, with 232 pins, as compared to the 184 currently used in a DDR (I) DIMM. Similarly, a DDR (I) DIMM has a bit more pins than the 168 found in an SDRAM DIMM.

However, this isn't such a big deal for the mobo makers; it's just a matter of putting on a different slot and different traces on the board. One of the things that has made the transition from SDRAM to DDR more "evolutionary" than a transition to (say) RDRAM would have been, is that the same chipset can control both SDRAM and DDR, because they use the same (or at least backwards-compatible) commands. It's not often mentioned on the hardware-enthusiast sites which are only interested in benchmarking the fastest stuff around, but just about every DDR chipset is also available with SDRAM. You need a different motherboard--because the slots are incompatible--but the cost to the mobo makers for offering both versions is pretty small.

In a similar vein, the commands for DDR-II are a superset of the DDR-I command set, such that DDR-II chipsets should very easily be able to detect and use DDR-I as well, just like DDR-I chipsets currently use SDRAM as well. Furthermore there is talk (dunno if it will happen) of releasing DDR-I DIMMs in the 232-pin DDR-II form factor; that way, one could buy a motherboard and use either DDR-II or DDR-I in it, with no problems. Of course old sticks of DDR-I will not fit, and the new ones will not fit in current DDR motherboards.

So, while such a scheme doesn't get rid of all the headaches of an incompatible upgrade path, it does address some, albeit more on the mobo producer end of things than on the IT inventory end of things. It is indeed a pain that, while SDRAM (and the SDRAM form factor) enjoyed around 4 years as the mainstream memory type, DDR-I will only be on top for 2 before DDR-II takes over. And DDR-II will be lucky to have 3 years before something pin-incompatible comes along to replace it.

On the other hand, the SDRAM -> DDR transition probably would have happened a year earlier if Intel hadn't tried to transition to RDRAM instead. And, meanwhile, being stuck with SDRAM for 4 years meant DRAM bandwidth only doubled (and actual performance did less than that) over a period when CPU clock speeds increased by a factor of 5 or so. I think the DRAM industry wants to make speed increases more frequent than they were a few years ago, even if this means more inconvenience for corporate IT departments.

This is important because industrial and corperate-mission-critical is older equipment. and an upgrade path for ram is still important.

I'm not sure what you mean here. As far as the upgrade path for greater RAM capacity goes, standard SDRAM and DDR-I will still be made and sold for quite some time, even if they will eventually be more expensive than their newer and faster brethren. EDO RAM is still being made and sold today.

If you mean an upgrade path for higher-performance, of course you can't just buy faster DRAM and expect it to speed up (or even work in) a system that was built to use a slower type. The system clock sets the DRAM speed, and unless the system has been validated to run at the new DRAM speed, doing so amounts to overclocking. An performance upgrade path for RAM is always going to require purchasing at least a new CPU module and accompanying memory bus, if not a new machine.

geez...

[echosa]

something tells me that the "computer age" is getting a bit unbalanced... we're nowhere NEAR reaching the hardware limits that we already have (home useage speaking at least) and we're already surpassing whats more than enough? sounds a bit unneccesary, at least for the moment...

Clever Marketing Scheme

[goldspider]

How many people besides myself have noticed that there are hardly any backward-compatible hardware technology breakthroughs?

I understand that there are physical limitations well beyond my comprehension that are factors in all of this, but it seems that any time one of these improvements comes out, be it RAM or CPUs or any number of other upgrades, a new chipset has to be developed to support it?

I'll tell you why: because marketers understand that some people (many of which frequent this site) will pay plenty extra to have the latest technology, no matter what.

I, for one, am sick of it. For once I'd like to be able to upgrade my CPU or RAM without having to buy a new motherboard and re-install my entire OS.

Sorry for the rant, but I think the fact that every incremental hardware update requires a new chipset is noteworthy.

Re:Clever Marketing Scheme

[Yet Another Smith]

Simple. Next time you put together a computer, buy a chipset for DDR-II, but only put pc2100 memory in. Then, six months later, buy screaming fast DDR-II memory. Sure it'll cost you more in the long run, but you'll have upgraded your RAM successfully. Or you could do like everybody else and buy more RAM rather than faster to make your upgrade.

Of course, if you had one of the early AMD t-bird chips, you could buy an XP and run it on the same motherboard, so CPUs generally are upgradeable within the same form-factor. Same is true for a Duron to XP upgrade.

If you're having tp re-install your OS for a CPU or RAM upgrade, you've got bigger problems.

Re:Clever Marketing Scheme

[Tackhead]

> My point was that every time I've upgraded my CPU or RAM, I've had to buy a new motherboard as well. Hence the OS re-install.

Depends on the upgrade.

As a lark, when I moved from a Pentium I (430TX chipset) to the 440BX-based system (new mobo, Celeron CPU, new chipset, new video card, new sound card, new RAM), I tried transferring (after Ghosting :) a Win9x system to it. After several reboots and requests for the install CD, it actually ran.

(Then, of course, I wiped it and reinstalled from scratch to be sure I had a decent config and drivers, but it's theoretically possible. I was amazed it worked at all.)

That said, I chose the 440BX because it had headroom for growth. That lowly Celeron-366 (66 FSB oc'ed to 100 for 550 MHz) is now running a Celeron 800 at over 1 GHz (FSB at 124.) It could probably run at an FSB of 133, except that I've got mismatched sticks of PC133.

Am I getting as much out of that PIII-1G on a 440BX chipset as I could? Of course not. My 5400 RPM drives are still running ATA-66. It's still SDRAM. It's still the same PCI frequency.

But the upgrade was $50 for the CPU, gave me another two years out of the system, and (most importantly) required no time-consuming OS or driver changes, be it Win9x, 2000, or Linux.

I think we might be at a similar point with the P4 Northwoods. Buy a cheap Northwood 1.6A now, and a mobo with a chipset (SiS 645DX or Via P4X333) with some FSB headroom. Throw some fast DDR-I into it.

Two years down the road, I think you'll probably be able to plunk in another "$50 CPU and $20 stick of RAM" behind the OS's back, giving you decent performance for another year.

There are no guarantees, of course, but by paying a $50 premium for quality parts today, you can often get better than 50/50 odds of saving $500+ two years from now. That's a good risk, IMNSHO.

Getting old...

[Little Dave]

Hell, it makes you feel old when 512Mb is being bandied around as the standard memory size. I remember getting all excited about a 512k(!!) trapdoor expansion for my Amiga, for which I paid 100 quid for.

Oh yeah, and this whole website was fields back then... far as the eye could see. ;)

A few facts

[Anonymous Coward]

Intro: I am D. C. Sessions, and I'm the chair of JC-16 (one of the committees which participates in the DDR standard). Here are a few facts:

The two big reasons for the generational change are

• Voltage change: DDR II is going to 1.8 volts to allow thinner gate oxides and denser, faster devices.
• Internal timing. First-generation DDR has some architectural timing issues which make it impossible to go much beyond 333 in volume production.

Yes, this makes for backward-compatibility problems.

Yes, the Committee (JC-42.3) put a huge amount of work into making DDR-II as backward-compatible as possible

Yes, we're starting work on DDR-III. You'll have to wait until 2006 or so.

Target speeds for DDR-II were set at 600 MT/s for fully-loaded systems and 800 MT/s for embedded stuff like graphics.

The signal-integrity issues for DDR-II are ugly, but we met the margin specs with lots of conservativism thrown in, so once we get hands-on time with systems you'll probably see the numbers exceeded just as the original DDR targets were.

Flame away. You can get more info at JEDEC or Advanced Memory International.

IBM Chipset for which CPU?

[ayden]

The article says that IBM made a chipset to take advantage the new memory speed, but what CPU does the chipset support? Athlon? P4? G4/G5? Anyone? Bueller? Bueller? Frye?

fucking marketting!

[Cyno]

Gah! I fucking hate marketting! Why can't they just call it QDR, because that's what it is. Quad Data Rate fucking DIMMs. God damn it, motherfucking, sun of a bitch!

EXA Anyone?

[OS24Ever]

Have you not seen the new intel servers IBM is releasing like the x360 and x440?

Need more PCI? Add a drawer of 12 and plug in a cable. Need more processors? Buy another four way and plug them together, you have an eight way.

Hot swap a failed memory dimm lately? You can in a x440.

There are a lot of cool tech coming from IBM in the xSeries servers. There are only so many marketing guys out there

But it sure is easy to bash IBM, so people do. They are changing. You think the layoffs of the last year or two are getting rid of the good people and not the middle management?

SDRAM - DDR - DDR-II transitions

[dpilot]

I suspect that there were really two factors at work in the short DDR-I lifetime.

First, Intel muddied the waters with the big exclusive Rambus push. While there was DDR work going on prior to the Rambus push, there was some very real contention in carrying both programs through development. This doesn't even mention quite a bit of "wait and back the winner," at many levels of the industry.I suspect that the success of the Athlon competing with PIII had almost as much to do with DDR success as Rambus prices.

Second, there were very real signal integrity issues that had been skirted for quite some time, and really came to the fore with DDR. That took some time, but more thought has been applied forward to DDR-II, so it shouldn't be as painful.

Re:Any news on their 210Ghz transistor?

[RevRigel]

210GHz is the unity gain bandwidth of that transistor. That means that when switching at 210GHz, it only puts out as much as you put in (in layman's terms). In terms of chip production, that means you can only have a fanout of one. Therefore, it's impossible to construct basic circuit elements at that speed. Once you throw in a requirement for a fanout of say, 10, you've thrown 10 gate capacitances in parallel with that transistor, which is going to knock its bandwidth down quite a bit. To what, I don't know, but it wouldn't surprise me if it was by at least an order of magnitude.