*NOTE* This has originated from www.extremeoverclocking.com and their forums. I take no credit in the making of this guide as I am not the aurthor. I am simply reposting this for others to view.
Memory Timings is a Very confusing subject for most folks. I admit that I do not know everything there is to know, and this guide will continuously evolve as I learn more and with input from everyone here.
I’m most comfortable on DFI motherboards as that’s all I run so some terms may be DFI Specific, but I will attempt to include DFI and Industry standard abbreviations/Translations where I can.
Much of this info has come from my own personal experience, but a lot of the info is compiled from various websites and articles which will be referenced at the end of the guide.
I’ve spent a LOT of time tweaking my ram lately to get the absolute best performance from my system. A64 Tweaker is a great tool for playing with these settings without countless lockups, no-posts, and reboots. I figured I’d share some of my findings and results and the procedure I use to tweak your ram to its highest performance.
TOOLS You’ll Need.
A64Tweaker: I use .60Beta but many people have reported issues with .60. Some versions offer more options than others and I like the mix in .60. Use whatever version you’re comfortable with.
SuperPi Mod 1.4 The Mod version is critical because we can be talking .050 differences here in some cases.
MEMTEST86 Bootable CD Image linked. Lots of other versions available if you look. The Standard in putting your Memory through its paces.
Sysoft Sandra Benchmark
The Big 5
Most motherboard had manual adjustments for “The Big 5” memory timings. These are
CAS Latency (tCL)
RAS-to-CAS Delay (tRCD)
RAS Precharge (Trp) (AKA: Row Precharge)
Act-to-Precharge Delay (tRAS) (AKA: Min Ras# Active Time)
Command-Per-Clock (CPC) (AKA: Command Rate)
The first 4 are typically publicly specified by Ram manufactures in the order listed. For Example,:
OCZ Gold BH-5 is Specified at PC3200 with Timings of 2-2-2-8. tCL, tRCD, and tRP are all “2” and Tras is “8”
OCZ Value (VX) is specified at 2.5-3-3-7. tCL-2.5, Trcd-3, Trp-3, Tras-7 (But we all should know by now those are just a starting point)
CPC will be covered in depth in the next section.
What do the “big 5” settings do and what are their effects on Speed/Bandwidth?
tCL: CAS controls the amount of time between receiving a command and acting on that command.
Typical Settings: 2,2.5,3
Effects: Large Effect on Bandwidth and Stability
NOTE: BH5/BH6 typically will NOT run at anything higher than 2
tRCD: tRCD is the cycle time between the first stage in memory access, the row strobe, and the second stage.
Typical Settings: 2,3,4,5
Effects: Medium Effect on Bandwidth and Large effect on Stability
tRP: tRP is the amount of time it takes for memory to terminate the access in one row and begin another.
Typical Settings: 2,3,4,5
Effects: Medium Effect on Bandwidth, Large effect on Stability
tRAS: tRAS is the time between receiving a request for data electronically on the pins of a memory module and then initiating RAS to start the actual retrieval of data.
Typical Settings: 5,6,7,8,9,10
Effects: Minimal Effect on Bandwidth, Medium effect on Stability
CPC: This is the delay between when a IC is selected and the time commands can be issued to the IC. The more chips on a module (Single vs. double sided) the more difficult it is for the memory controller to do this in 1 command clock. Most quality modules with an A64 Processor can do 1T rates on 256 and 512mb modules. 2T can increase your overclock but at a substantial loss in bandwidth.
Typical Settings: 1T,2T
Effects: Large effect on Bandwidth, Medium effect on Overclocking
Extended memory settings
These settings are found on DFI Nforce3/4 boards and many of them can be found in bios’ from other manufacturers as well. Exact definitions are rather hard to find, but I’ve found lots of references to these timings as well as done extensive testing with A64 Tweaker to be able to report these findings.
Row Cycle Time(tRC)
Typical Settings: 7-14
Effects: Large Influence on Bandwidth/Stability.
Row Refresh cycle time (tRFC)
Typical Settings:11-17
This timing is usually always set to 2-4 clocks higher that the tRC.
Effects: Large Influence on Bandwidth/Stability.
Row to Row Delay(also called Ras to Ras delay)(tRRD)
Typical Setting:0-4
Effects: Slight Influence on Bandwidth/Stability.
Write Recovery Time (tWR)
Typical Settings: 2, 3
Effect: Minimal Influence on Bandwidth/Stability.
Write to read Delay (tWTR)=
Typical Setting: 1, 2.
Effect: Minimal Influence on Bandwidth/ Stability.
Read to Write delay (tRTW)
Typical Settings1-4
Effects: Minimal Influence on Bandwidth/Stability.
Write CAS# Latency (tWCL)
Typical Setting:AUTO/1
Note from RGone of DFI: Note from RGone of DFI-Street.com Posts with AUTO,1 or 5 in setting works on my board with “any” brand or size and speed of memory!
Effect: Large Influence on Stability.
DQS skew Control
Settings: Auto, Increase Skew, Decrease Skew.
Effects: Increase for performance, and Decrease for Stability,
Minimal Influence on Bandwidth/Stability.
DQS Skew Value*= 0-255 in 1.0 increments. This is the value that is Increased or Decrease when you set the DQS skew control. I typically run 125 Increase on my BH-5
DRAM Drive Strength
Typical Settings: Level 1-4
Effects: Level 1 should be run with CPC Enables, I had instability with anything else Some People others have had success w/ using level 2-4 if CPC is disabled(2T).
Large Influence on Stability.
Max Async. Latency
Typical Setting: 5.0-10.0 (7.0 netted the best results on my BH-5)
Effects: Minimal Influence on Bandwidth/Stability.
***8-9 may help increase stability when using HTT speeds of 300+
Read Preamble time
Typical Setting: 4.0-7.0 (5.5 netted the best results on my BH-5)
Effects: Minimal Influence on Bandwidth/Stability.
***5.5-7 may help increase stability when using HTT speeds of 300+
Idle Cycle Limit 000-256 in doubled increments.
Minimal Influence on Bandwidth/Stability.
Dynamic Counter
Typical Setting: Auto, Enable, Disable. Enable for Slight Performance increase, Disable for slight Stability increase
Minimal Influence on Bandwidth/Stability.
Refresh Period (tREF)
Typical Setting: AUTO/0032-4708 in variable increments.
Typical translations: Some motherboards have more options than listed here, but I have not found what they us values are for the addition settings so they are not included.
---------------------
0388= 100mhz(15.6us)
0516= 133mhz(15.6us)
0648= 166mhz(15.6us)
0780= 200mhz(15.6us)
---------------------
0908= 100mhz(7.8us)
1032= 133mhz(7.8us)
1168= 166mhz(7.8us)
0016= 200mhz(7.8us)
---------------------
1536= 100mhz(3.9us)
2048= 133mhz(3.9us)
2560= 166mhz(3.9us)
3072= 200mhz(3.9us)
---------------------
3684= 100mhz(1.95us)
4196= 133mhz(1.95us)
4708= 166mhz(1.95us)
0128= 200mhz(1.95us)
I’ve found 3.9us to be the best setting for my BH-5, 15.6 results in slightly lower bandwidth, but the same stability it seems.
Effects: Minimal effect of Bandwidth, Minimal to Extreme effect of stability
**Make SURE your tREF setting is set to a value related to your memory Divider! Extreme system instability will result if this is not matched up!
When using the DFI Specific dividers I’ve left this on “AUTO” , Once I finish deciphering the settings, I will add to the list.
R/W Queue Bypass
Typical Setting: 8x, 16x.
Minimal Influence on Bandwidth/Stability (Lower slightly increases stability).
Bypass Max
Typical Setting: 4x-7x
5x-7x for max performance,0x-4x has shown to increase overclock slightly
Minimal Influence on Bandwidth/Stability.
32 byte Granulation
Typical Setting: Disable (8burst), enable (4burst).
Effect: Try Disable (8burst) for more bandwidth. Try enabling 4 burst for more stability
Minimal Influence on Bandwidth.
All Of this is great information right? Now what to do with this info is next….
Time for some Tweaking!
First Step First……
CORE SETTINGS.
Every manufacturer lists the 4 primary ram settings they feel the ram can do at its rated speed. This is CAS, TRRD, TRP and TRAS (See my somewhat complete guide to memory settings first if your not up to speed on what I’ve said so far)
Set your bios for these settings, set your Idle Cycle Limit to 256, (Dynamic Idle Cycle Limit should be on Dynamic as well) and EVERYTHING ELSE ON AUTO. Boot your system and open up A64 Tweaker. Write down all these settings (I usually do a Print screen and print a hardcopy…. Messy handwriting…)
Now that you know all of your base info from the SPD Settings you can reboot and find your max memory speed. Adjust the 4 primary settings as you need to get your max overclock.
Once your gotten your ram up to speed its time to start tweaking. I’m assuming at this point that you are Memtest stable (At least 10 passes of Test 5 and Test 8) and your rig is Prime Stable as well. This procedure is NOT about getting the highest overclock out of your ram. It’s about getting the most performance out of the clock speed you can easily hit! If you have to divide your memory some more and end up 20 MHz shy of what you’ve done in the past, these tweaks can make up for that difference in MHz in most cases!
Where to Start:
Fire up A64Tweaker and hit REFRESH (I usually hit it twice as I’ve seen it pull some incorrect info a couple times)
Run a Reference 2 Mil Super Pi (Or higher if you have the time, but I’ve found 2 mil is a good base)
Run a Reference Sandra Memory Bandwidth
Run SuperPi to 1 Million Digits (I’ve found anything lower doesn’t show enough difference sometimes) THREE TIMES. This is important as 1 million times tend to vary .100 seconds and we could be dealing with gains in that range in most cases. Average it out and record that score.
PROCEDURE: After each Adjustment, run SuperPi 1mil three times and average out the time.
I always start with Max Async Latency and Read Preamble. You will most likely lock up your rig with these settings so it’s best to get em out of the way first.
Drop Max Async Latency 1ns and re-run SuperPi 1 mil three times and repeat until you start to see a slower average speed or your system locks up. (Generally, you will lock the system up before you see a decrease in performance.)
Move to Read Preamble and drop the time .5ns at a time. You will find there is a point of degraded performance here.
Move to Row Cycle Time and Row Refresh Cycle Time Again, this is a place where you will most likely lock up your Rig. (You did write down your Async and Read Preamble finding yes?) These settings are closely tied together so I recommend adjusting them together. I’ve found that TRC+4clocks have netted the best results. If Row Refresh NOT 4 higher apart to start, change that first and run you’re Superpi’s. Now drop them together 1 clock at a time until the system Locks.
Now do TWTR and TRTW. I’ve found that these setting like to be the SAME. I’ve gotten better times with them both at 2 instead of 1-2 in most cases.
TWCL is next.
Followed by TWR
Then TRRD
At this point you should be .250-.3 faster on your SuperPi Times…. Pretty cool eh?
The last thing I’m going to go over in depth right now is Idle Cycle Limit and Dynamic Idle Cycle Counter. I’m still working on TRef and some of the other settings for optimal performance and hope to add a big tRef section soon as this field tend to confuse most people (Including me)
Drop your Idle Cycle limit one notch and run your SuperPi’s. Repeat this until you see degradation in performance. If you are able to get your Idle Cycle Limit to 8 or below then you can DISABLE the Dynamic Idle Cycle Counter. Once you’re this quick, the dynamic adjustment the System makes can hurt performance.
At this point, you should be .3-.5 faster on your SuperPi Times. If not, then you have some ram that has Awesome SPD settings. The likes that I’ve never seen…..
Finally Run a Sandra Memory Bandwidth Mark and check out the increase in Bandwidth! This seems to vary a lot but a 200-250 gain is not uncommon!
Now reboot into Bios and enter in all the settings you’ve written down, Reboot, and Make sure you’re still good on Memtest tests 5 and 8. If it doesn’t pass, I’ve found that adjusting READ Preamble up .5 and upping the Idle Cycle up one notch are your best bets at clearing up the errors. (Or a small voltage bump if you’re running VX or BH-5)
Please post your results after your tweak your settings! I’d like to see that this procedure works as well for others as it has for me! I ask that you post the following.
Your Original A64Tweaker Window
Final A64 Tweaker Window
Original Reference Memory Bandwidth
Reference SuperPi Time (Your 2mil or up time)
Bandwidth Results after tweaking
And finally your final SuperPi reference speed.
(Screenshots aren’t necessary for the SuperPi and Sandra’s… we’ll take your word for it.)
My results on my G.Skill HX
Reference SuperPi (2mil) 1.10.748
Reference Memory Band 7011/6996
Final SuperPi (2mil) 1.8.253
Final Sandra 7229/ 7156
I most likely will be integrating this guide with my “Somewhat complete guide to memory settings” sticky I feel the information here expands on that guide perfectly.
Reference SuperPi Time (Your 2mil or up time)
Bandwidth Results after tweaking
And finally your final SuperPi reference speed.
(Screenshots aren’t necessary for the SuperPi and Sandra’s… we’ll take your word for it.)
Memory Dividers
This has been covered multiple times, but needs to be said again here.
From: “The Somewhat Complete AMD 64 Overclocking Guide:
There is ALWAYS a memory divider.
Setting the Memory to 1:1 means that the HTT bus is multiplied by the CPU Multiplier and then Divided by the CPU Multiplier to set the Memory speed. This means that it is OK to run your Memory at its peak efficiency and still go higher with your HTT bus if your CPU can take it. Take note that I said its "OK" not advisable. There are still sufficient tests out there showing that running a 1:1 ratio will garner you the best overall performance.
Since A64's use an On chip memory controller, the Ratio must be calculated a bit differently than old. 5:6 is NOT always 5:6. For a Chart of Memory divider effect, see the end of the Overclocking guide.
ALSO. Please make sure your read Transam’s Memory Divider training guide! There is also a small program that will calculate your speeds at the end of his excellent guide!
http://forums.extremeoverclocking.co...d.php?t=149717
References
As said before much of the info in this guide has been complied from sources around the internet.
Justin Allman’s A64 Overclocking guide on auphanonline.com
http://www.auphanonline.com/articles...id=1518&page=1
Jess, RGone, AngryGames, Johnrr6 from DFI-Street
http://www.dfi-street.com/forum/showthread.php?t=11397
Repost of jess's info on VR-Zone
http://forums.vr-zone.com.sg/showpos...9&postcount=22
PCStats Memory Bandwidth vs. Latency Timings article
http://www.pcstats.com/articleview.c...eid=873&page=1