Archived from groups: alt.comp.periphs.mainboard.asus (
More info?)
Thanks for reply Paul, I'll try some tweaking later today.
l8r
"Paul" <nospam@needed.com> wrote in message
news:nospam-1908052223320001@192.168.1.178...
> In article <0zuNe.92514$G8.49412@text.news.blueyonder.co.uk>, "Eyghon"
> <eyghon@127.0.0.1> wrote:
>
>> Hi,
>>
>> I have the A8N Premium and originally had 1.5GB RAM, 2x512MB Corsair
>> PC4000
>> in A1 & B1 + 1x512MB Corsair PC3200 in A2. I removed the PC3200 and added
>> 2x512MB Geil PC4000 in A2 + B2. With all 4 slots filled the system
>> wouldn't
>> even post (just a BEEP.......BEEP.......BEEP), however changing the
>> configuration so the Geil is in A1 + B1 and the Corsair in A2 + B2
>> allowed
>> system to post but Windows then either came up with a missing file error
>> and
>> suggested doing a repair install or simply Blue Screened. I've just
>> manually
>> set the RAM to 333 CAS2 and system booted normally again.
>>
>> Main question is am I losing much by running RAM at 333 CAS2 instead of
>> 400
>> CAS2.5 or 3. All other memory settings were left at default so are there
>> any
>> other memory settings I should adjust to tweak performance.
>>
>> System:
>> A64 4000+ (San Diego) (Rev SH-E4 according to CPU-Z)
>> 2xHitachi 250Gb SATA2
>> 1xSeagate 120Gb SATA
>> 4xOptical drives
>> XFX 6800GT
>> 580W Hiper Modular PSU
>> Mobo BIOS 1005 (would updating to 1006\7 fix this)
>>
>> TIA
>
> See slide 24 of this memory tutorial.
>
>
http://corsairmicro.com/corsair/products/tech/memory_basics/153707/index.html
>
> CAS, tRCD, tRP, tRAS are delays, measured in clock cycles.
> When you change from 4000 to 3200, they scale, but must be
> rounded to the next highest half-integer or integer.
>
> Say you are at 3-3-3-8 at PC4000. PC4000 is DDR500, or a clock
> of 250MHz, or 4 nanoseconds. Converting from clock ticks to time,
> that is 12ns-12ns-12ns-32ns. If running the RAM at PC3200 or DDR400,
> the clock is 200MHz or 5 nanoseconds wide. 12ns/5ns = 2.4 clock
> cycles. CAS must be rounded to the next nearest half integer, so
> it becomes 2.5, while tRCD and tRP round to the next nearest
> integer which is 3 clock cycles. 32ns/5ns=6.4 clock cycles, and
> rounded to the next full integer, that is 7. So, your 3-3-3-8
> memory can be run at 2.5-3-3-7 at PC3200 speeds (of course, that
> is an estimate, and the memory has the final say).
>
> To start with, I would simply keep the original timing, until
> the memory is operating properly. If the PC4000 rating is 3-3-3-8,
> keep it at 3-3-3-8 while at PC3200.
>
> The next concept is "Command Rate". It is a delay, but in this
> case, it measures how long the address/command is presented to
> the memory, before the information is strobed into the memory.
> It is a penalty you pay for each address/command sent to the
> memory, and is different than a single delay penalty cost of
> the other parameters.
>
> Command Rate 2T is slower than 1T, but 2T also allows extra time
> for the address to stabilize, so is a good choice for demanding
> memory configurations.
>
> Swapping the slot pairs for the Corsair and the Geil was brilliant,
> as that means you have bypassed whatever problem the BIOS has got.
> If you look at the threads here lately, a number of people have had
> problems with four sticks, and unless I miss my guess, they are
> using Rev.E processors. It is good you found a way to get something
> to work with the four sticks.
>
> Your original configuration, of three sticks of memory, could only
> have worked with a Rev.E processor. A BIOS detecting a previous
> processor would have had to throw away the stick in the A1 or A2
> slots, if you were using three sticks. As it is, a Rev.E processor
> using three sticks, runs in single channel (64 bit) mode, so your
> memory bandwidth is significantly reduced. Moving to two matched
> pairs of DIMMs was the right thing to do. (Using 2x1GB would be
> even better, but would have meant selling the other RAM.)
>
> Using two sticks only, allows room for overclocking. You could have
> a lot of fun with two sticks of PC4000 memory, but four sticks of
> DDR PC4000 memory cannot be pushed that high (perhaps PC3500 on
> a good day - on an Intel chipset). The Athlon64 memory controller
> doesn't have the drive strength to do tricks like that, so with
> four sticks, I would be happy to get stability at PC3200. I don't
> know right off hand, what typical upper speed you can get with
> four sticks. (You'll be able to tell me that, soon...)
>
> There are two potential settings for four sticks. You can use
> DDR333 and a command rate of 1T. You can use DDR400 and a command
> rate of 2T. The DDR400 setting gives better performance, according
> to one poster here who has tried it.
>
> An article on Anandtech warned about the "Auto" setting of the
> BIOS, and how it uses Command Rate 1T when it shouldn't. If you
> want to go to DDR400, make sure to manually set Command Rate 2T,
> ("Enable 2T"), perhaps before you insert the second two sticks.
> If the computer is able to POST right now, you should be able to
> set DDR400 Command Rate 2T, then test for stability by booting
> a test floppy containing memtest86+ from memtest.org. For the
> other timings, you could choose the aggressive numbers, calculated
> as in the math example above (2.5-3-3-7, if the RAM was 3-3-3-8),
> or you could leave the RAM at its original values, like 3-3-3-8.
> The Command Rate setting affects B.W. by 20%, while a small
> difference in CAS might be 3-5%. Dividing those numbers by 3,
> gives an average application level performance penalty of 7%
> for Command Rate, and 2% for a small difference in CAS.
>
> You can use the bandwidth indicator in memtest86 as a barometer
> of how well you are doing on the bandwidth front. You can try
> the different settings in the BIOS, run memtest86+, and get
> a bandwidth measurement at the same time you wait for two full
> error free passes, to prove the new setting is stable.
>
> Give the RAM a bit more Vdimm, like 2.7V, to ensure the memory
> meets the JEDEC 2.6V standard value. A little extra won't do any
> harm, while really large amounts of Vdimm will make the memory
> run real hot. If you download product info for your DIMMs, it
> will usually tell you how much volts the RAM can take.
>
> HTH,
> Paul