Basic RAM principles:
1) General - Random access memory relies on an electrical charge, which is constantly refreshed, to hold data. Without power, memory is lost. It does not retain data without power.
2) Definition - DDR RAM is Dual Data Rate RAM. If your mobo uses DDR RAM, your BIOS may list RAM frequency (half of RAM's rated DDR speed) or actual DDR speed. The PC equivalent is 8 times the DDR speed or 16 times the actual frequency, e.g. DDR400 runs at 200 MHz and is rated as PC3200 (also DDR2-400 = 400 MHz = PC2-3200). DDR2 works on the same principles as DDR, but it has a 240 pin interface compared to a 184 pin interface for DDR. Unless otherwise states below, info for DDR applies to both DDR and DDR2.
3) RAM Timings - Higher numbers (5-5-5-15 and higher are possible) equate to looser timings. Lower number (2-2-2-5 being the typical lowest) equate to tighter timings. Ultimately, a timing is an interval based on clock cycles. If a particular event that coincides with a particular timing is occurring every two clock cycles, then it is running tighter than if it were occurring every 4 clock cycles. The occurrence is effectively happening more quickly (in less clock cycles) with a "tighter" timing.
4) Dividers - 1:1 CPU:RAM synchronicity generally offers best performance on Intel systems. Loosening timings to achieve higher FSB will most often offer better performance than tight timings with limited FSB. With this principle in mind when overclocking, achieving the highest CPU FSB even at the expense of very loose RAM timings is the ultimate goal.
5) Timings vs. Performance - At any given clock speed, tighter timings will offer better performance than looser timings.
6) Dividers vs. Performance - Running RAM faster than the CPU may offer better performance than 1:1. This is particularly true on DDR2 as some DIMMs can run at DDR2-1200 (600 MHz) or faster. If you have found your CPU's maximum frequency and have the headroom to run a divider that will run the RAM faster than the CPU (i.e. your RAM is not yet at its maximum frequency), then go for it. Limiting your CPU overclock to run the RAM faster than the CPU, however, will reduce performance. Again, go for the highest CPU FSB possible first and foremost.
7) RAM speed vs. Motherboard Specifications - RAM's rated speed is not its required speed. Even though RAM may be rated to run up to DDR500, for example, it will still run at 400 MHz if that is the corresponding bus speed of the system. Therefore, if your motherboard lists DDR400 or DDR2-800 as the maximum RAM speed, it is very rare that it would not safely run RAM rated at a higher speed.
A few examples: if you have an 800 mhz FSB CPU, and you install DDR500, it will still only be running at DDR400, at stock bus speeds, until you up the mobo's FSB. If you are running a 533 mhz FSB CPU in conjunction with PC3200, the RAM will still be running at 266 mhz in 1:1 at stock bus speeds. The benefit of having faster-rated RAM is to allow for higher overclocking.
8) Your mobo will support any speed of RAM above what it is rated for. You can always run faster-rated RAM at slower speeds. The highest official JEDEC DDR RAM speed is PC3200, and all s478 motherboards offer support up to this speed (i.e. w/ an 800 mhz FSB CPU, your highest available divider will be 1:1, which supports DDR400). If you look a little futher into your manufacturer's compatibility list (if they've actually taken the time to test compatibility), you will most likely find that the board does officially support any given higher-rated RAM.
For example, if you can achieve a FSB of 270 with your CPU, then your mobo will support PC4400. If you can achieve a FSB of 240, then your mobo will support DDR500. Again, you can install any faster-rated RAM in your motherboard, but it will only be running at the twice the speed of the system bus.
9) If your mobo supports dual channel memory, then take advantage of it. Dual channel memory does not necessarily need to be purchased in a dual channel "kit", but you are best to install two identical matching sticks when running dual channel. If running dual channel with two speeds of RAM, your mobo will only operate in dual channel mode at the speed of the slowest stick. Dual channel memory greatly increases memory throughput (ideally it is supposed to double memory bandwidth, but in reality, this is not true).
10) Running RAM slower than the CPU (also referred to as "using a divider") hinders performance.
With these things in mind, as they pertain to overclocking, the following is recommended:
1) Find your highest possible CPU overclock, even if this means running a divider.
2) Leave timings as loose as possible (3-4-4-8) in 1:1 to establish the highest possible speed that your RAM will achieve.
3) If you are using a divider, or you have established that your CPU will not overclock any further, or you have decided that you only want to run your CPU at clock speeds slower than the RAM's max, then start tightening timings to achieve more performance from your RAM.
4) If you are looking for maximum performance, and are forced to use a divider to achieve maximum CPU overclock, test your highest 1:1 (assumedly with loose timings) against your highest CPU overclock (with a divider and tightest possible timings). Establish for yourself which offers best performance in your system.
5) If you are using a divider, you will greatly improve performance by installing RAM that corresponds with your CPU's overclock (i.e. allow stable 1:1 operation). Along these same lines, if you are in the market for RAM, and want the best performance out of your system, establish your CPU's highest overclock before making your purchase.
Synthetic benchmarks effective in establishing performance and testing stability:
1) Prime 95 (separate CPU and RAM stress tests, along with an overall system stability tester)
2) Memtest86 (tests some CPU, mostly RAM)
3) SiSoftware Sandra Memory Bandwidth Benchmark Test (establishes memory throughput)
4) 3dMark01 (random game tests based on CPU, RAM an video card performance equally)
*For specific examples, see charts at post #12 of this thread
hmmm.... i think you should specify that memtest also tests for vcore/fsb stability, specifically on tests 1-4. errors in these tests are usually fsb or vcore related, like too high for the former or too low for the latter.
tests 5 and above are geared towards stability of the memory (timings, vdimm).
people often mistake memtest as a "purely" memory related diagnostic. thats why there are many people complaining about pc3500s or pc4000s getting errors on memtest on oc'ed systems, even if they're running below or in spec.
hope this helps
oh, and i learned this the hard way =)
searchin the whole net for the memtest functions made me dizzy.
NoAffinity also helped me diagnosing the problem I had
Thanks for clarification, macg, I adjusted it. I've never really used memtest...and I did think it was only a memory tester.
i decided to see if there was any truth to the articles i read, mostly on other sites that i had to use a translator on.
i ran my my 2.4a @ 3.6 which was unstable, used memtest and looped tests 1-4. all of them got loads of errors, looped 5 and 8 but got only 2 errors.
ran it again @ 3.4 but with a much lower vcore, still got errors on tests 1-4, nothing on test 5... upped the vcore a notch higher and tests 1-4 were error free... i didnt touch mem timings and vdimm at all..
verified this with a friend from another forum and he got the same results too..
i've read a LOT of posts regarding RMAs of high quality modules mainly because of Memtest errors, and on tests 1-4. guess this should clarify things and help people.
oh and btw, great guide you put together!
I vote sticky.
|All times are GMT -7. The time now is 11:54 PM.|
Powered by vBulletin® Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.
User Alert System provided by Advanced User Tagging (Pro) - vBulletin Mods & Addons Copyright © 2020 DragonByte Technologies Ltd.
vBulletin Security provided by vBSecurity (Pro) - vBulletin Mods & Addons Copyright © 2020 DragonByte Technologies Ltd.
vBulletin Optimisation provided by vB Optimise (Pro) - vBulletin Mods & Addons Copyright © 2020 DragonByte Technologies Ltd.