Thread: Tightening the timings

Thats right, I am at it again..........

So, I figured out after a lot of tweaking that my cpu (1055T) has a bad/flaky memory multiplier. So while I am overclocked to 250 FSB my ram, which should run at 1600 mhz just fine, won't run at that memory multiplier. So I am forced to put it down to 1333 mhz.


So until I get better cooling for my cpu and can achieve 300 FSB, I am looking at tightening my timings as the only option left open to me.


With that out of the way, I am looking for advanced timing explanations or guidelines, even formulas would be helpful.


So far I found this site: Tweakers.fr


Included here:

Code:

-RAS# Precharge (tRP). 
Controls the number of clocks that are inserted between a row precharge
command and an activate command to the same rank.


-Activate to Activate delay (tRRD). 
Number of clocks between two row activate in different banks of the same rank.


-RAS# to CAS# Delay (tRCD).
Controls the number of clocks inserted between a row activate command
and a read or write command to that row. Last Intel chipset (965 and P35)
allow to change RAS# to CAS# Read Delay and RAS# to CAS# Write Delay separately


-Activate to Precharge delay (tRAS). 
Number of clocks taken between a bank active command and issuing the
precharge command. Usually, tRAS=tCL + tRCD + 2.


-Write to Precharge Delay / Write Recovery Time (tWR). 
-Write Recovery time is an internal dram timing, values are usually 3 to 10.
It specifies the amount of delay (in clock cycles) that must elapse after the
completion of a valid write operation, before an active bank can be precharged.
-Write to Precharge is a command delay, and is calculed as:
Write to Precharge = tCL - 1 +BL/2 + tWR.
BL(Burst Lenght) practically always 8.


-Write to Read command Delay / Write to Read Delay (tWTR). 
-Write to Read delay is an internal dram timing, values are usually 2 to 8.
Specifie the number of clock between the last valid write operation and the next
read command to the same internal bank
-Write to Read command is a command delay, and is calculed as:
Write to Read = tCL - 1 +BL/2 + tWTR.
BL(Burst Lenght) practically always 8.


-Read to Precharge delay (tRTP). 
Number of clocks that are inserted between a read command to a row
pre-charge command to the same rank.


-Row Cycle Time (tRC). 
Determines the minimum number of clock cycles a memory row takes to
complete a full cycle, from row activation up to the precharging of
the active row. For optimal performance, use the lowest value you can,
according to the tRC = tRAS + tRP formula. For example:
if your memory module's tRAS is 7 clock cycles and its tRP is 4 clock cycles,
then the row cycle time or tRC should be 11 clock cycles.


-Cas# Latency (tCL).
Number of clocks that elapses between the memory controller telling
the memory module to access a particular column in the current row,
and the data from that column being read from the module's output pins.


-Command Rate / Command per Clock (1T/2T). 
Delay between when a memory chip is selected and when the first active
command can be issued. The factors that determine whether a memory
subsystem can tolerate a 1T command rate are many, including the number
of memory banks, the number of DIMMs present, and the quality of the DIMMs.








-Refresh to Activate Delay / Refresh Cycle Time (tRFC). 
Determines the number of clock measured from a Refresh command (REF)
until the first Activate command (ACT) to the same rank


-Refresh Mode Select (RMS) / Refresh Period (tREF). 
Determines at what rate refreshes will be executed. Contrary to other timings,
higher value is better for performance.


-Performance Level / Read Delay (tRD).
tRD is the number of memory clocks from DRAM Chip Select# assert
to Host Data Ready# assertion on the FSB.
Hight influence on performance and stability.


-Read to Write delay (tRTW). 
Number of clocks that are inserted between a read command to a write
command to the same rank.


-Round Trip Latency. 
Number of Uncore clocks that are inserted for Read data after
a Read Cas# is send to a DIMM.


-Four Activate Window (tFAW). 
Specifies the time window in wich four activates are allowed the same rank.


-Precharge to Precharge delay (tPTP). 
Number of clocks that are inserted between two Precharge command in
different banks of the same rank.


-Write-Read Command Spacing (tWR-RD). 
This field determines the number of turn-around clocks on the data bus needs
to be inserted between write command and a subsequent read command on Different Rank.


-Read-Write Command Spacing (tRD-WR). 
This field determines the number of turn-around clocks on the data bus needs
to be inserted between read command and a subsequent write command on Different Rank.


-Write-Write Command Spacing (tWR-WR).
This field controls the turnaround time on the DQ bus for WR-WR sequence to
different ranks in one channel.


-Force Auto Precharge.
When enabled, force auto Precharging with every read or write command.
This may be preferred in situation where powers savings is favored over performance.


-Maximum Asynchronous Latency.
Specify the maximum round trip latency in the system from the processeur to
the DRAM devices and back.


-Maximum Read Latency.
Specify the maximum round trip latency in the system from the processeur to
the DRAM devices and back. This time is specified in NorthBridge clock and
includes the asynchronous and synchronous latencies.


-Read/Write Queue Bypass
Specify the number of times that the oldest operation in the DCI read/Write
queue may be bypassed .


-Queue Bypass Max
Specify the maximum of times that the oldest memory-access request in
the DRAM controller queue may be bypassed .


-DRAM Idle timer. 
Determine the number of clocks the DRAM Controller will remain in the idle
state before it begins precharging all pages.

Now, I have my ram stable at the following settings currently:

Code:

Trp     Trrd    Trcd    Tras    Twr     Twtr    Trtp    Trc     Tcl     CRate   Volts
5       4       7       17      8       Auto    4       22      5       2T      1.65

The next thing I am going to try lowering is the Twtr, because auto doesn't suit me...


If anyone has info on RAM Drive Strength or the timings that I haven't listed here it would be much appreciated!!!!!!

Just tried setting the Twtr to 4 instead of auto. Even though 4 is supposedly what auto places it at.... it didn't like it. Perhaps there is a linked setting somewhere that I have to change with it.


Then I tried lowering Trcd to 6 instead of 7. This was also unstable.


I looked over the general formulas I have and I have 2 conflicting formulas for the Tras.
tRAS = tCL + tRCD + 2
and
tRAS = tCL + tRCD + tRP


I had been following the second and decided to give the first a try. tCL+tRCD+2=14 for the current setup.... well I couldn't set that as 15 is the lowest the Crosshair IV will go. So I set 15 and it passed hyperPi 32M.... so I think its stable.


Table:

Code:

Pass/Fail    Trp     Trrd    Trcd    Tras    Twr     Twtr    Trtp    Trc     Tcl     CRate   Volts
Pass-Stable  5       4       7       17      8       Auto    4       22      5       2T      1.65
Fail         5       4       7       17      8       4       4       22      5       2T      1.65
Fail         5       4       6       17      8       Auto    4       22      5       2T      1.65
Pass-Stable  5       4       7       15      8       Auto    4       22      5       2T      1.65

So I have my ram stable at: 1333 5-7-5-17-22-2T (Processor at 3.5 ghz)

After a slight bump in voltage and FSB I now have the following stable settings:
~1066 5-7-5-17-22-2T (Processor at 4 ghz)
or
~1600 8-8-8-22-31-2T (Processor at 4 ghz)

Those are the two memory multipliers available to me.... perhaps I can get the timings lower for each, but my question is; Which one should I spend my time working on?

I have read that overall when you come down to it the difference isn't that great between the ram mhz ratings. So will the significantly better timings be better even if its technically going so much slower.


I forget the actual mhz values, but my cpu settings are currently:
FSB: 285
CPU Multi: 14.0

Why not spend 30-40$ on a quality cooler, that will solve all your problems!

I do have a quality cooler, the Scthe Mugen 2 rev b.

While I am playing a close game with heat, I have it generally under control. My processor idles at 23c and at load goes to 56c (prime95 small fft).

For 4ghz stability I have a 1.45 vCore and a 1.221875 CPU/NB (volts). The ram is only at 1.65 volts (standard 1.5), and for now I would like to keep it there.


I am doing a blend test on prime95 currently.... I might stop it to play some l4d2 later tonight.... we shall see.

Here is an update:



apologize for the overclock.net in the SS..... I was too lazy to re-take.

Another update...

This is with a ton more tweaks. Most improvement was gained by maxing out my CPU/NB multiplier. I also got a few hundred MB/s by adjust various sub-timings, the biggest gain from tWR and tRTW both of which went from 8 to 4 clocks.



Over at the Overclock.net MaxxMEM Rankings I am climbing higher.....
AMD MaxxMEM Results/Rankings - Overclock.net - Overclocking.net

Man, this is pretty impressive. All above my head but still pretty damn cool.