DDR Memory vs. System Bus

Camman

Now, this may come off as a newb question to some of you who are more experienced with overclocking and memory timings and such, but for some strange reason the concept of DDR memory has always been kind of lost upon me. When I had to switch over from all my 100/133 SDRAM I really just knew that DDR was the new thing on all the boards.

Now, I'm not a total newb on this like I understand that like the data is transfered twice per clock so it's really like 133x2 for DDR266 memory, but like I said, it's still a bit of a mystery to me.

Right now in my machine I have a stick of PC2100 and a stick of PC2700 crucial memory. I am running an AMD AthlonXP 2800+ which touts 333mhz system bus on the box, now, am I correct in saying that I would need to run only PC2700 memory in my machine to take advantage of the 333mhz bus, and also, running higher than PC2700 would have no advantage because the bottleneck would lie at the 333mhz bus on the processor, right?

Again, I consider myself pretty computer saavy, so don't feel like you have to dumb it down for me or anything, but some of the DDR stuff is just something that never really "clicked" with me, perhaps if someone can explain the answers to those two questions I will be able to graps it. Thanks in advance!

-Cam

leishi85

if u run your ram higher than ddr333, it will have advantage if u run your cpu FSB in sync with the ram.

ddr ram basically just transfer data on both side of a signal

this graph is really easy to understand

muddocktor

Your PC2100 memory might (most probably will) hold you back on running the memory in synch with the processor's fsb speed, since it most probably can't run at 166 fsb. With an AXP machine and nforce2 chipset, you get the best performance by running the fsb and memory in synch. You run into problems both ways (mem either running slower or faster than fsb speed) due to timing issues, which gives lower performance. If you are running the memory and fsb in synch presently, then you have a good overclocking stick of PC2100 in your machine. If you are having to run the memory speed slower than the fsb speed because the PC2100 can't run at 166 fsb, then you would be better off selling the PC2100 and buying another stick of PC2700, preferrably matching your present PC2700.

mmonnin

Yeah its the same way with PC100 and PC133. Running PC2100 on a system that needs PC2700 is like running PC100 on a 133MHz FSb system.

EMT

That's quite a good question Camman. The speed of newer memory chips doesn't explain changing from rise to rise+fall rather than doubling the frequency of the RAM - perhaps rise+fall works where twice as fast rise wouldn't? However, for the RAM data to get pushed through at the rate it's operating, you would need a DDR (rise+fall) bus to the Northbridge/CPU (I'm not sure which) or one twice as fast...

Perhaps we can find a technical page to explain it. I'm going to bed though

Thrax

You come up with an 800MHz memory chip, bus and PCB that can support it, and I'll show you the fastest SDRAM ever.

madmat

I think the reason they just didn't double the speed of SDram to make the troughput faster is that the limitation in clock frequencies would've jumped up and hit them in the heads and if you think about it, the uber fast DDR has to run relaxed timings to allow that kind of speed in the first place i.e.; 3-4-4-8 cas numbers for 270 and up FSB so to achieve a 266 SDram speed you'd be looking at similar if not much higher cas frequencies to attain those speeds.

To run the ram in step with the FSB without DDR for 333 and up FSB speeds I really doubt we'd have seen that happen or if it did it would've been pretty costly. You gotta remember, we're talking about the ram's clock rate and the higher the clock rate, generally, the more expensive the ram.

Straight_Man

Right, they hit thier heads on internal latency and FSB mitigation combined in RAM and the need to route data to RAM (combined, CPU does not write directly to RAM in reality, RAM is not bussed to CPU directly-- it is bussed through bridge chip) that can only process data through from CPU to RAM chips so fast. That is why the limits. IF we wanted to pay for video card RAM, we would be close to unity with FSB-- IF CPU could write to RAM like it writes to cache over a very short bus and at native FSB rates. BUT, for P4 chipsets other than 925X which can as designed use DDR3 RAM, CPU is clocked 4X for data pushing and pulling and RAM is pushed and pulled at 2X. Essentially we would need a CPU<->RAM bus at DDR4 rates, and RAM is not there yet.