Slashdot Mirror
DRAM Almost as Fast as SRAM
An anonymous reader writes "IBM said it has been able to speed up the DRAM to the point where it's nearly as fast as SRAM, and that the result is a type of memory known as embedded DRAM, or eDRAM, that helps boost the performance of chips with multiple core calculating engines and is particularly suited for enabling the movement of graphics in gaming and other multimedia applications. DRAM will also continue to be used off the chip."
Enough, just SCRAM! :)
to go for title of most patents filed in 2007
"Stallman says add to this code and you are one of us. Gates says use this code and you belong to us."
With all these improvements in processor and RAM speed, when can I expect a faster HDD? A solid state drive would be nice.
All chips wait at the same speed. Why not concentrate on the bottlenecks rather than what is already one of the fastest components in any system.
There is no "I disagree" mod for a reason. Flamebait, Troll, and Overrated are not substitutes.
IBM's new motto
Knowledge is how to play a game, intelligence is how to win, wisdom is knowing what game to play.
So SRAM and DRAM is fast? That's nothing... Wait until they combine it into SDRAM!
EDram.
I don't know about you, but I wouldn't want something named that in me...
The older fellas should get it (and their wives might, too)
Living With a Nerd
To those wondering why it would be good to have DRAM as fast as SRAM: SRAM doesn't need to be "refreshed" constantly, and is faster, but takes up many more transistors and is therefore much less dense and more expensive for the same amount of memory.
However with DRAM it takes quite a bit of power just to keep data in memory (because of the constant "refreshes"), which isn't the case with SRAM. So this discovery wouldn't take SRAM out of production for applications which require its low power usage.
// MD_Update(&m,buf,j);
doesnt the xbox360 already use Edram?
i'm looking foward to ZRAM more actually. gotta love weird side effects of amd/ibm's SOI process that allows one bit with just a transistor.. no capacitors. :D
but on topic.. almost as fast in what way, bandwidth or latency? i see the former being really easy. the latter... not so much. SRAM still beats the fastest DRAM in latency by an order of magnitude, easily.
when are we going to get solid state hard drives to keep up with the rest of the computer crap these days? I wish someone would quite lolly gagging and just produce them for christ sake. they have them.. now sell them. stop playing with ram and start playing with harddrives!
I don't get why this is news. Embedded-DRAM has been in heavy usage for many years now.
Both the title and the summary are quite misleading, since eDRAM is on-chip and that of course is much faster than external off-chip memory, be SRAM, DRAM or whatever.
Some big examples? PS2, Nintendo Gamecube, Wii, Xbox 360. All these consoles use eDRAM for their GPU's on-chip framebuffers to enhance their performance, and that goes back to at least the year 2000 when the PS2 came out.
Some will be quick to say "no, the Nintendo consoles use 1T-SRAM, not DRAM". Yeah, right, but even 1T-SRAM (despite its name) is a form of embedded-DRAM.
- Otaku no naka no otaku, otaking da!!!
If you could stick a crapload of this on the Cell, then those SPEs could have more than 256kB memory each, and utilizing them would become dramatically easier.
I'd guess the next revision of Cell will have a shitload of eDRAM on it. And it will either have more SPEs, or a new bus that allows multiple Cells to be used. The latter would be more expensive to implement, but probably result in higher yields than substantially growing the Cell to support more coprocessors - the yields are already poor if they just turn all the SPEs on, or else why would they be disabling one?
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
So I guess everyone should just drop what they are doing and work on the "bottlenecks". Even though IBM sold off their HDD division, they should still work on it anyway. Intel should stop making cpu's faster and instead redirect the company to do R&D on HD's?
EE Times article. Today SRAM is used for processor caches, but new multicore chips need massive (i.e. expensive) cache. Because eDRAM is much denser than SRAM, it allows chip designers to fit much more cache in the same size chip, increasing overall performance. IBM and AMD use silicon-on-insulator (SOI) technology, while the rest of the industry uses bulk CMOS; eDRAM for bulk has been available for a while (it's used in Xbox 360 and BlueGene/L for example), but now IBM has developed SOI eDRAM that can be used in IBM's future processors (and maybe AMD's).
Speed vs. power would an appropriate measure....
IBM said it has been able to speed up the DRAM to the point where it's nearly as fast as SRAM, and that the result is a type of memory known as embedded DRAM, or eDRAM, that helps boost the performance of chips with multiple core calculating engines and is particularly suited for enabling the movement of graphics in gaming and other multimedia applications.
:-) IBM's new speed achievement (what are they exactly???) has jack shit to do with the "eDRAM" name. The article author is simply clueless.
;-) And it's somewhat about the ratio of storage to logic whether a chip features "embedded memory" (inside logic) or "embedded logic" (inside memory)... But speed has no place in those definitions.
This is just plain wrong. Any "DRAM embedded into logic circuitry" is "eDRAM" (a type of "embedded memory"). And yes it's been around for over a decade. PS2 and GameCube made it commonplace
Of course, these definitions aren't absolutely clear-cut and crystal-clear. A DRAM chips in your DDR stick has lots of logic in it -- it's an electronic device, not a rectangular area of storage
Sure it's great if they have sped up eDRAM -- AFAIK it has remainded below the 1 GHz barrier so far.
While SRAM usually takes up to six transistors per cell (and an electrically different lithography process) compared to DRAM's just one, there's Mosys Inc's 1T-SRAM that (yup) sports only one transistor per SRAM cell, however the density hasn't been quite as good as with DRAM and it has significantly tailed behind SRAM's slockspeed. The article should at least mention this stuff -- 1T-SRAM is what GameCube used for "eDRAM".
Yup the Xbox 360 has the ATI Xenos GPU (accompanying the IBM Xenon triple-core CPU) which is actually a package with two chips: one for the GPU and northbridge, one for the smart framebuffer. The latter has a few megs of eDRAM, and besides providing humongous framebuffer bandwidth through a superwide chip-to-chip bus, is notably capable of performing the simplest graphics ops (Z comparing and culling, alpha blending, multisample blending for edge anti-aliasing) on-chip within the memory, much like Sun's ancient 3D-RAM. This eliminates a lot of ping-pong traffic between the 3D core and video memory which helps performance a lot -- mostly the GPU can use the GDDR3 shared memory only for texturing.
This really says it all about the quality of TFA:
Earlier this week, Intel Corp. said it has developed a research chip capable of performing calculations as quickly as a supercomputer while only consuming as much energy as a light bulb.
Who the heck are these analogies for? So there is exactly one (1) model of supercomputer ever made and some single universal standard wattage for all lightbulbs, eh?
Since this is used for cache memory it may be possible to eliminate the refresh cycles. A cache row can always be re-fetched from main memory. All you need is some reliable method to tell if has expired. Any cache row which hasn't been accessed long enough for it to expire is, pretty much by definition, not very critical to performance anyway.
Stop worrying about the risks of nuclear power and start worrying about the risks of not using nuclear power.
Note that striping (RAID-0) gives no benefit at all for writes, and destroys *latency* for reads. So it's only beneficial for streaming large files (say, in video editing) -- and of course it doubles your risk of data loss (as one failing drive zaps *all* your data) so it's really only useful for a work/scratch space for your large video/audio/CAD files.
;-)
Better to have a single 10K Rappy (or better a piece of 15K SCSI/SAS goodness -- where are 15K SATA drives already???) as a "system/apps/work cache" and then large 7K SATA drives for RAID-0 "scratch" pair and a RAID-1 "save" pair and even then burn all the important stuff to DVD weekly...
Of course the "more FPS!!!oneone" kiddies will ignore this advice and add blue case bottom lights too
There are 2 areas of latency for a cache, the first is the performance of the actual data cells, and the second is the speed of doing a lookup in the cache. The larger the cache, and the higher the degree of set associativity, the longer the lookup takes. Thus you're unlikely to see this eDRAM used for L1 caches, and probably not for L2 caches either, as more cache would slow them down, even if the cells are just as fast as SRAM. The sweet spot will probably be for L3 caches, that are already slow by cache standards, but a whole lot faster than system memory. Since L3 caches are large, the cost savings for switching to eDRAM would be largest there.
As for power concerns, DRAM is higher than SRAM, but a larger L3 cache may reduce the traffic through the memory controller, and out to the DIMMs, which will probably more than make up for any increase in power density in the cache.
No, I'm not a fan of patent trolls; but this isn't patent trolling. IBM has created a new, better way to embed cache RAM on the CPU die, at a signifigant cost in both manpower and materiel. This isn't like they patented "a method to check customers out with one click" or something similarly banal. This is a real, new technology which took a great deal of time, energy and work to create. No "prior art", no "trivially obvious" - this is exactly the kind of technological advancement which patents should protect.
DRAM will also continue to be used off the chip.
Oh, good! They had me worried that I could no longer keep my DRAM in the water cooler. And how could I get through my day without a bit of chipless DRAM floating in midair above my keyboard?
Goodness. What next? They'll try to take away my off-chip flatware?
Typical soft errors (e.g. bit flips due to cosmic neutrons, which is a particular problem on aircraft and/or satellites) occur in any memory device. But because of all this 'refreshing' that DRAM endures, it's soft error rates have been declining while those for SRAM have been increasing. So perhaps SRAM isn't the holy grail of all memory applications after all.
Well, it IS pretty late, but I read the headline as: "DRM faster then "SPAM". Quite a disappointment, really...
"can't run, can't hide...oh well, return 0"
Sure, we'll get 3THz RAM, and it will be $150 for a 1GB stick. That's not what I want, nor what I expect. What I expect is that I get a 2GB stick for what was the price of a 1GB stick 12-18 months ago. By now 4GB sticks should be $75.
In the last couple years prices haven't dropped hardly at all and new stuff is no bigger than before. That doesn't happen in IT unless someone isn't playing fair. So who is it and how do we get them to stop?
Help stamp out iliturcy.
I am in no way an expert, but I read about other upcoming types of RAM which also sound interesting:
Z-RAM. One cell is a single transistor. Faster than SRAM, which uses 6 transistors per cell. http://en.wikipedia.org/wiki/ZRAM
TTRAM. One cell contains 2 transistors. As fast as SRAM, according to Wikipedia. http://en.wikipedia.org/wiki/TTRAM
Good lord! I've always wondered what happend to those COAST (Cache On A STick) modules back in the Pentium 1 days. Brings back memories...
Life is not for the lazy.
DRAM, that's fast!
I'll wait for the development of SPeDRAM
ARAM is of course still fastest. However it's good to see DRAM get some distance from the horribly slow FRAM and GRAM.
Oh shit! I forgot to click "Post Anonymously"...
That's what's happening. Used to be 100KB was massive. Gates once said that 640KB was enough solid state memory and that the rest would be stored on disks. Later came 1.4MB floppies, 5MB hard drives, then 80MB, 120MB, not so long ago 500MB, and today >500GB. Magnetic drives will continue to carry the major, giga- and tera-byte data stores of the world, especially when low power consumption, ruggedness, and speed are not required. While solid state memory continues to move in from the small and fast(and expensive) to the ever larger and less costly.
An unusual example of this is my Nokia 770 which uses the removable flash card for virtual memory paging space. That is an application that would have been rejected out of hand as a possible use of solid state memory just a year or two ago. Today it works just fine and gives a nice performance and flexibility boost to a machine with limited(64MB) RAM space.
Good judgement comes from experience, and experience comes from bad judgement.
- W. Wriston, former Citibank CEO
Just make them support HyperTransport.
And make them in BTX board. Or even better : make them Socket F/F+ and Socket AM2/2+/3 compatible and people will be able to drop them as vector accelerators in their existing multisocket motherboards.
(... hum, technically, that'll require making both the front-end bus HyperTransport *AND* the memory controller DDR (2/3) instead of current Rambus controller).
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
I just read about an investigation into price-fixing among manufacturers of flash memory devices. It includes any USB memory stick or flash card businesses or consumers may have purchased.
Apparently, if you purchased a memory stick, or flash memory device you could be included if the investigation leads to a class-action suit.
The press release said go to www.hbsslaw.com for more information.