Why When Overclocking Does One Lower Memory MHz?

[Livin_Lenend]

Why When Overclocking Does One Lower Memory MHz?

I am building a amd 64 x2 3800 and i plan on trying to overclocking it to 2.5 so i am doing research to see good settings to test

For Example:
1-Vcore to 1.425V
2-HTT to 3x
3-Memory supply Voltage to 2.9V
4-Config Memory to Run 333MHz instead of 400MHz
5-Memory settings to (2-3-2-5)
6-Raise Processor from 2.0 to 2.5 at 10X

thanks for your help, remember, first time building computer, then first time overclocking.

thanks

mike

[kenc51]

because by default the ram communicates with the cpu @ 200MHZ (400MHz DDR)
Your cpu has a multiplier of 10x

SO 200MHz * 10 = 2GHz (speed of cpu)

For you to get 2.5GHz (or 2500MHz) you need to get to 250MHz * 10 = 2.5GHz

To do this you would either need to OC the ram to 250MHz (ddr500) or use a memory divider.....
With a memory Divider you can OC the CPU without needing to OC the ram....Normally if you OC the CPU you shouldn't use a divider (333Mhz setting) but with AMD64 there is no decrease in performance (@ least very very little)

What RAM have you got?

[FZ1]

Note that sometimes tighter memory timings make up for lack of bandwidth - i.e. running 2-2-2-5 @ 400Mhz may be better than 3-4-4-7 @ 450Mhz. When you get things going, do some benchmarking to see the better settings.

[Livin_Lenend]

i have OCZ 2GB (2 x 1GB) 184-Pin DDR SDRAM Unbuffered DDR 400 (PC 3200) Dual Channel Platinum System Memory - Retail

DDR 400 (PC 3200)
Timing 2-3-2-5
Unbuffered

[TheGeekMistress]

theres a very good reason that memory bandwidth scores of over 3200mb/s is useless to the 754 core and 4000mb/s for the 939 core.

the new design of the A64 basically brought the FSB "on die" like they did for the L2 cache in the past. the "on die" memory controller has a locked internal FSB of either 800mhz for 754's or 1000mhz for 939s. this locked internal FSB sends data from the integrated memory controller to the core thru the 800/1000mhz internal FSB.

so let's take the 939 socket cpu and nf4 mobo. the 939 core can only make use of the internal FSB link of 1000mhz. so this is why dual channel memory is important to the 939 platform as it impliments the dual core. it needs to run at JEDEC standards of DDR400. DDR400 or PC3200 only gives the memory controller about 3200mb/s the internal memory bus of 1000mhz requires 4000mb/s (if it were the 754 800mhz internal FSB it would only require 3200mb/s). so when running dual channel DDR400 the system memory bandwidth exceeds the core's internal 1000mhz FSB bandwidth and therefore makes no use of the extra system memory bandwidth under overclocked speeds.

so with the 939 core with it's internal core to memory controller fixed at 1000mhz FSB, the dual channel DDR400 supplies it's needs just fine. now, the reason for overclocking the system memory bus is only to lower the latency between the on die memory controller and the system ram for "bussiness" desktop memory resident apps, like Windows Shell Explorer, IE, Adobe Photoshop, etc. you can see the latency improvements for the internal memory controller to system ram in benchmarking programs such as ScienceMark 2.0, WinBench, Content Creation, SuperPI. the only way to lower the L1 and L2 latency is to overclock your cpu, and in doing this overclocking to the cpu, you also overclock the memory controller frequency as it runs the same speed as the core.

TGM