32bit rdram.

Galvin

Distinguished
Dec 31, 2007
107
0
18,680
I thought intel was going to use 16bit 1066rdram. Then after reading toms articles I see the next 850chipset will use 32bit rdram. I guess I was mislead by the public.

I guess 32bit rdram will be a little better since you have two channels and terminator on one module. But does anyone know how much of a performance this will add?
 
There will be both dual-channel 16-bit PC1066 and single-channel 32-bit PC1066. The only change is that the single-channel will not have to be used in pairs, but will have higher latency.

All of this because of morons who'd rather have worse performance than buy two sticks of RAM (even though most people do anyway).

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>
Don't step in the sarcasm!
 
How do you figure higher latency.
But two channels on one module, this cuts down on trace lengths.
I don't see how the latency can be higher ?
 
The dual-channel technology is what cut down on latency, since it reads from both channels at once. A single 32-bit module wouldn't function the same.
Someone feel free to correct me, but I believe that's the case.

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>
Don't step in the sarcasm!
 
Well by that assumption then it's not dual channel on one module. From what you said it's sound as if Rambus is using two channels as one channel. Yet on their datasheets a 32bit ram module puts out just as much bandwith as two 16bit modules on dual channel.

You can't get that much bandwith without going dual channel either on one module or two.

I wish I knew more about this 32bit technology.
 
There will be both dual-channel 16-bit PC1066 and single-channel 32-bit PC1066. The only change is that the single-channel will not have to be used in pairs, but will have higher latency.

All of this because of morons who'd rather have worse performance than buy two sticks of RAM (even though most people do anyway).
I think it isn't so much to please the morons as it is just making the headway for the next technology. If they jumped to dual-channel 32-bit right away, then they would have to actually double the number of paths on the motherboard, which is never easy. So while they work on that, they offer single-channel 32-bit. Then in six months or so, they offer dual-channel 32-bit with double the theoretical bandwidth.

As for the latency, I suppose it depends on how they are designed. The design could just be the equivalent of two 16-bit single-sided RIMMs squished together into a single RIMM. If so, then the latency would be no different than before. It might even be a tiny bit less.

<pre>If you let others think for you, you're the
only one to blame when things go wrong.</pre><p>
 
You can, but it won't be dual-channel (effective 64-bit). That will come later.

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>
Don't step in the sarcasm!
 
We'll have to see a benchmarks with 32bit rdram single channel, and 16bit rdram dual channel.

If 32bit rdram performs even worse, then it won't sell very well. First off I think 32bit rdram boards will have 2 slots.
This makes it more expensive if you want to use a gig of ram. a 512meg module costs far more than than two 256meg modules from what I seen in pricing so far with existing rdram.

So most people will probably use the 2.4A northwoods with 400mhz 16bit dual channel boards instead of 32bit rdram.

Double the datapath increase the speed but double the latency. heh.
 
Quad channel it on the roadmap for rambus
As 64 bit module

cheap, cheap. Think cheap, and you'll always be cheap.AMD version of semi conducteur industrie
 
There's just way too much work over RDRAM, more channeling, when a dual-channel DDR solution that nVidia has already, and could have provided to Intel, would give the P4 all it needs quickly and cheaply.
DDR 400, Dual Channeled would rule, and that is of no doubt.

--
For the first time, Hookers are hooked on Phonics!!
 
Intel is already working on dual-channel DDR SDRAM chipsets:
<A HREF="http://www.intel.com/products/server/processors/server/xeon_mp/index.htm" target="_new">http://www.intel.com/products/server/processors/server/xeon_mp/index.htm</A>

The problem is that dual-channel DDR SDRAM is difficult to accomplish because it requires double the paths of single-channel DDR. Where as dual-channel RDRAM (or even quad-channel) is much easier to achive because RDRAM inherantly uses a mere fraction of the paths that even single-channel DDR SDRAM uses, so even at quad-channel, RDRAM still uses less paths than dual-channel SDRAM.

Dual-channel DDR SDRAM might give the P4 all of the bandwidth that it needs. However, it will always be through an asynchonous connection. Where as quad-channel RDRAM could give more bandwidth than a DDR SDRAM, and still remain synchronous to the FSB.

<pre>Join PETT.(People for Equal Treatment of Trolls)
Trolls:Keeping bridges clean 'n safe.</pre><p>
 
This makes it more expensive if you want to use a gig of ram. a 512meg module costs far more than than two 256meg modules from what I seen in pricing so far with existing rdram.
According to PriceWatch.com, a 512MB PC800 RIMM costs $128, and a 256MB PC800 RIMM costs $64. So a 512MB RIMM is <i>exactly</i> twice the cost of a 256MB RIMM.

Alternatively, also according to PriceWatch.com, a 512MB PC2700 DIMM (only at cas2.5) costs $155. (A true CAS2 from Corsair costs $216.) And a 256MB PC2700 DIMM (only at cas2.5) costs $77. (A true CAS2 from Corsair costs $107.) So anyone looking to build a PC2700 DDR SDRAM solution will be paying more <i>per memory stick</i> than someone building a PC800 RDRAM system. (So Intel DDR330 fans, think about that one.)

So most people will probably use the 2.4A northwoods with 400mhz 16bit dual channel boards instead of 32bit rdram.
This could very well be true. I have also heard that a lot of home users have also been able to use 1.6A Northwoods with 400MHz 16-bit Samsung RDRAM, to overclock their FSB to 133 with a high success rate.

Double the datapath increase the speed but double the latency. heh.
Which is an odd statement from someone who earlier specifies single-channel 32-bit and double-channel 16-bit, meaning the <i>same</i> number of paths.

<pre>Join PETT.(People for Equal Treatment of Trolls)
Trolls:Keeping bridges clean 'n safe.</pre><p>
 
To add to this, (as I've said many times), a dual-channel DDR platform would require a 128-bit datapath, or 128 wires embedded in the motherboard.
To get the same bandwidth from current RDRAM (comparing PC2100 and PC800 here), you'd only need 42 wires.

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>
Don't step in the sarcasm!
 
I did some reseach on it. So it's not odd.

The reason intel went dual channel was to reduce latency and increase bandwith. Now intel is taking a step backwards.
 
The real question that I'd like to know is: Are they?

For all we know, the single-channel 32-bit RDRAM might simply be two dual-channel 16-bit RDRAM sockets and RIMMS squished together into a single socket and RIMM, and then named single-channel 32-bit for marketting purposes. In which case, single-channel 32-bit RDRAM is <i>exactly the same</i> technologically as double-channel 16-bit RDRAM, just in a smaller package. (And thus also in theory a hotter package.)

And in which case, as you predicted, single-channel 32-bit RDRAM motheboards would only have two RIMM slots.

<pre>Join PETT.(People for Equal Treatment of Trolls)
Trolls:Keeping bridges clean 'n safe.</pre><p>
 
Interestingly, McKinley is slated to have 6.4GB/s of bandwidth available to it. That would suggest dual-channel 32-bit (effectively 64-bit) RDRAM.

However, the bus is 128 bits wide, which could suggest PC400, or simply un-double-pumped and dual-channel with 64-bit modules? Or quad-channel with 32-bit modules.

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>
Don't step in the sarcasm!
 
mckinley work normaly with quad channel DDR 200

cheap, cheap. Think cheap, and you'll always be cheap.AMD version of semi conducteur industrie