Any advantage to running FBS & memory clock synchronously?

[G3rain1]

In the past I've known this to be true, but does it matter on the newer architectures. Specifically a Core2 Duo on a P35 board that uses a 1066 FSB.

Should I go and buy DDR2-1066 Ram for my system or wont it make a difference if I use the 800 Intel says I should.

 

[SEALBoy]

Your FSB is quad-pumped, so to run RAM at a 1:1 ratio, you'd actually have to run it at 266MHz (533MHz effective, or PC4200 DDR2). I'm not sure, but I think you'd be better off running PC6400 DDR2 at 400MHz (800MHz effective) unsynchronized.

 

[jackluo923]

I think you should run ddr2 3200 duo channel or ddr2 6400 single changle for core2duo E4xxx series and ddr2 533mhz duo channel or ddr2 1066mhz single channel for c2d E6xxx except for E6x50s. In C2D E6x50s, you need to get DDR2 5400 or 667mhz in duo channel or 1333mhz in single channel. You could also use ddr3 if you want on p35 chipsets, but it think it would be very expensive and underperform fast ddr2 module.

 

[SEALBoy]

Huh? DDR3? And don't the E6x50's use the same 1066MHz bus?

 

[jackluo923]

E6x50 is 1333mhz FSB. Its not out until July 22nd which is the price drop. DDR3 is new type of ram supported by P35 chipset. It's true that intel uses quad pumped bus, but you match the ram with the quad pumped bus not the basic FSB. E.G 1:1 Ram ratio for E4300 is 200mhz x 4 ( FSB) : 200mhz x 2( duo channel) x 2 ( DDR2 ) THis is the case for DDR2 3200

 

[SEALBoy]

Sorry, I was confusing E6x20 with E6x50.

Are you sure it won't be faster to run, say, dual channel PC6400 RAM rather than force it to run dual channel PC3200 for the sake of 1:1 ratio?

 

[jackluo923]

The difference between PC3200 and PC6400 isn't a lot if you're not overclocking. In fact PC3200 might even be faster at 1:1 ratio because you could set the timing to 3.3.3.8 or something. 1:1 is the most effective and most efficient and most money saving to achieve great performance because neither RAM or FSB bottleneck each other. If The speed is faster than 1:1, the data from the RAM is too much for the FSB thus the extra speed does not effect performance much. On the other hand, timing can change performance by a lot because the faster the timing, the faster the chipset's able to deliver Data from Ram to CPU. BTW, PC3200 is nearly half the price of PC6400. The best way to get great performance is buy PC5400 (667Mhz) and downclock it to PC3200 @ lowest timing. Crucial Rendition Ram (667mhz) 1GB is 29.99 @ 4,4,4,12 timing. I think its cheaper than buying PC3200@ 3,3,3,8 timing which is around 50$.

 

[Mondoman]

The difference between PC3200 and PC6400 isn't a lot if you're not overclocking.

First, PC-3200 and PC-6400 both refer to DDR RAM, not DDR2 RAM. Assuming you mean "PC2-3200" and "PC2-6400," reported differences range from a few percent to about 20%, depending on the app.

In fact PC3200 might even be faster at 1:1 ratio because you could set the timing to 3.3.3.8 or something.

Not true with real-world hardware.

1:1 is the most effective and most efficient and most money saving to achieve great performance

This is a common misconception.

... because neither RAM or FSB bottleneck each other.

True with respect to sustained throughput (i.e. consecutive memory addresses being accessed), but this ignores the influence of latency, which becomes especially important as memory accesses "jump around" in the address space. For example the MadShrimps test data (http://www.madshrimps.be/?action=getarticle&number=1&artpage=1962&articID=472) shows no big change going from DDR2-533 (1:1 ratio in their setup) to DDR2-667, but a noticeable improvement from continuing on to DDR2-800. Presumably, the slight latency decrease in going to DDR2-667 pretty much offsets losing synchronization between the memory bus and FSB; increasing the memory bus speed even more to DDR2-800 reduces latency further, resulting in performance increases.

BTW, PC3200 is nearly half the price of PC6400.

Current newegg prices are roughly $40 for either, for 2x512MB DDR2.

... The best way to get great performance is buy PC5400 (667Mhz) and downclock it to PC3200 @ lowest timing.

Now this is just silly.



The answer to the OP's question is that you should not run the RAM slower than 1:1 (DDR2-533 in dual-channel mode for 1066MHz data rate FSB CPU). Running one "notch" faster (DDR2-667 in this case) won't boost performance, but running two or more "notches" faster (DDR2-800 or higher in this case), will boost performance a bit.

 

[jackluo923]

667mhz ram are dirty cheap. 25$USD per 1gb stick.while 800mhz 1gb stick is around 35$.

PC-3200 and PC-6400 both refer to DDR RAM, not DDR2 RAM

I did not know that, but now i learned that DDR2 ram is refered to DDR ram.

jackluo923 wrote:
In fact PC3200 might even be faster at 1:1 ratio because you could set the timing to 3.3.3.8 or something.

Not true with real-world hardware.

Which is faster at random/sequential read and write? pc3200@ lowest timing 1:1 ram ratio or pc6400@ high timing?

jackluo923 wrote:
1:1 is the most effective and most efficient and most money saving to achieve great performance

This is a common misconception.

Through my experience, buying 667mhz RAM is cheaper than buying 400,533, 800 or any other speed.

MadShrimps

I'll try that program tomorrow. I didn't know any other program for testing RAM except for mem86.

The best way to get great performance is buy PC5400 (667Mhz) and downclock it to PC3200 @ lowest timing.

After reading a few times over this... it is kind of silly. I could easily up the vdimm a little bit and change the frequency to 800mhz and keep the same timing.

 

[G3rain1]

Thanks for clearing that up.

Now I have another question. Which should be faster, DDR-800 with 4-4-4-12 timings or DDR-1066 with 5-5-5-15 timings?

 

[Mondoman]

The DDR2-1066 (remember, "DDR2" is different from "DDR" RAM) is faster, since latencies are expressed as clock cycles, not absolute times. 5 clocks at DDR2-1066 is shorter than 4 clocks at DDR2-800.

 

[jackluo923]

That clears up everything. I've always thought the latencies are clock cycle of the FSB. What now i know i'm wrong.