Slashdot Mirror
Create Your Own Psuedo-RDRAM
ucsimon writes: "For those of you who like to overclock, this is truly some hardcore stuff that shows how to convert SDRAM to RDRam. I value my ram too much to try, but I wonder if anyone out there has tried this yet. " Having the same sentimental attachment to my RAM that most people do, I haven't even attempted this either - anyone know if this works, even in theory? Post below. [Updated 28th April by timothy:] Well, this won't actually convert your SDRAM to RDRAM of course, but it will add to your heat-slurping overclocking abilities.
This article has NOTHING to do with turning SDRAM into RDRAM.. All they did was put a copper shield on a stick of SDRAM, making it look similar to a real RDRAM module.
Doesn't Slashdot occasionally GLANCE at the links that are submitted before posting them?
The metal plate in Rambus isn't just an EMF shield--it also functions as a heat sink, as not only do RDRAM modules run hot, the heat production is concentrated in a very small area. I'm guessing that any speed increase the mad overclockers saw in their SDRAM was due more to improved heat dissipation than anything else.
"I'm not just an elder god, I'm cute too!"
See Rambus's RIMM design guide, section 8.2.1 "Mechanical Components - Heatspreader/Cover Reference Design". More info can be found on Intel's developer site, on the RIMM Module Reference Design page, under the section "Production Heat Spreader Reference Design".
Basically, RDRAM RIMMs give off a lot of heat, plus the heat is produced in a very small area. The aluminum covers are necessary to prevent hot spots. If you look at the figure 8-1 "Double-Sided RIMM Module Assembly-Exploded View", you'll see that the reference design calls for a thermal pad (Or thermal grease) between the heat spreader and the RDRAM chips.
This comes in really handy as I just converted my 16MHz i386 to 1.5GHz Willamette, which as we all know doesn't support SDRAM for now. Now if could just convert that 5.25" floppy drive to a DVD-RAM drive...
The article doesn't go into converting SDRAM into RDRAM at all. It's about making an EMI shielding for your SDRAM chips. This doesn't make the SDRAM into RDRAM although it make look like it. It's sort of like taking a saturn and putting a porsche body on top of it and expecting it to perform like a porsche.
RDRAM uses a 8 or 16 bit channel running at 300 to 400+ MHz with data being sent on the rising and falling edges of the clock. SDRAM runs at 100MHz on a 32bit bus that sends data on the rising edge only. Turning SDRAM into RDRAM would involve replacing the memory interface and control circuits on the SDRAM and changing the packing to fit in a RIMM. If you can do this, you'll probably do better to work as a chip designer for a semiconductor company and using the money you earn to buy some RDRAM.
"When you sit with a nice girl for two hours, it seems like two minutes. When you sit on a hot stove for two minutes, it
I know a few of these points have been brought up already, but let me summarize and add my own thoughts:
:)
1) The metal bracket on RIMMS isn't for EMI shielding, it's for heat. It's just like the heatsink on your CPU, which doesn't add to the EMI protection noticably.
2) A 150MHz SDRAM dimm isn't RDRAM. Rambus uses an entirely different protocol, and anyway... 150Mhz SDRAM is probably faster than rambus.
3) Your SDRAM won't work at 150MHz because it's simply too fast for the junctions to switch, not because of EMI. If you've got a serious EMI problem, you'll probably see it if you're overclocking or not.
4) That design shown probably wouldn't fix an EMI problem, either. It'll most likely act as an antenna, worsening the problem.
BUT... To whoever made that page.. Keep hacking. While this idea might not have worked, you may come up with a great supercooled DIMM refrigerator or something. Good luck.
Here is the working link...
The illegal we do immediately. The unconstitutional takes a little longer.
--Henry Kissinger
This article merely tells you to make an SDRAM module look like and RDRAM by giving it a shield agains electromagnetic emissions. This might have some benefit in very rare cases, but realistically, it's totally pointless.
The illegal we do immediately. The unconstitutional takes a little longer.
--Henry Kissinger
A metal shield over the chip will NOT reduce magnetic or radiation effects. The shield will simply SCATTER magnetic interference, NOT block it. RDRAM chips are engineered to protect against interference which encompases much more than tin foil. Personally, I prefer paying for RDRAM chips that actually do protect against interference rather than hacking a cheap DIMM chip and randomly scatter magnetic anomolies throughout my system RAM.
See, RDRAM has large latency (approx. 28 cycles @ 400MHz for a random access = 70ns) but for sequential accesses it spews the data out very very fast indeed (a 16-bit-channel RDRAM setup at 400MHz can spew out 16 bytes in 4 cycles = 10ns; a 32-bit-channel dual-chip setup - only found in consoles - can double this to 32 bytes in 10ns).
This is great for consoles, where you have lots of DMA going concurrently with semi-random CPU cacheline refills. For this application, RDRAM is much, much cheaper than the alternative (which would be 128-bit-wide, 100MHz synchronous RAM of some flavor). RDRAM also wins hands down on simplicity of system design since its data bus is only 16 bits, or 32 bits in dual-chip setups, plus only 8 more bits for addressing. It provides superb theoretical bandwidth, which can be approached quite closely in a well-designed console (er ... can someone build a well-designed console for me? ;)
However, for PCs, the case is much less clear cut. DMA seems to be far less important these days, and you certainly won't see half a dozen concurrent DMA channels going over a 200MHz FSB anytime soon. Even building a 200MHz FSB would be practically impossible in a modular system such as the desktop PC. Consoles are so non-modular, the current crop are internally dense as hell with heatsinks and EMI shielding, and the primary internal busses are no more than half an inch long.
To make RDRAM work great with a desktop PC would take, in my estimation, far, far better RDRAM controllers than are currently available, and integrated much more tightly with the CPU, and communicating over a synchronous bus of at least 128 bits.
RDRAM definitely wins on bandwidth, but its latency is so atrocious that you need a good prediction of what data is needed next ... easy to do on a dozen DMA channels but when the primary channel is the CPU it's near-impossible.
In summary, RDRAM is too application-specific to be suitable for desktop PCs, and the controllers are not yet mature enough to get anything like full speed from it in Real World PC applications.
Not to mention the fact that it's proprietary, so it intrinsically costs more to manufacture than other RAM designs.
--
It's a
-- Danny Vermin
This reminds me of high end audio freaks who paint the outsides of their CDs to prevent environmental light disturbing the laser, or fill their Amps with acrylic so the electrons in the circuits wont be affected by vibrations.
-- Newsraider.de - News aus dem Netz - German Geeknews Community
That page was about reducing EMI in your ram in case you are overclocking your system, not converting SDRAM to RAMBUS. They just mention in the beginning why RAMBUS has the shielding.
EK-Hack
remove the SPAM from e-mail address