![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
RAM OverClocking
- Thread starter Karant
- Start date
First - ram in dual channel is faster than single channel at any speeds
Second - well you oc ocz ram the same way you'd oc any other ram, by upping the mhz they run at, but since your ram already runs at higher speeds than your cpu fsb the only thing you can do to oc the ram is to mess with the timmings, if you have an option to do so in the bios, or to oc everything by raising the fsb, but in that case you need to watch the agp/pci ratios, cpu temps and psu voltages
how to test your stuff http://www.ocforums.com/showthread.php?t=418387
Merry Christmas !
D
Second - well you oc ocz ram the same way you'd oc any other ram, by upping the mhz they run at, but since your ram already runs at higher speeds than your cpu fsb the only thing you can do to oc the ram is to mess with the timmings, if you have an option to do so in the bios, or to oc everything by raising the fsb, but in that case you need to watch the agp/pci ratios, cpu temps and psu voltages
how to test your stuff http://www.ocforums.com/showthread.php?t=418387
Merry Christmas !
D
(cf)Eclipse
Freelance Overclocker
- Joined
- Feb 18, 2003
- Messages
- 30,027
try something like this...
first off, get memtest and install it on a floppy or cd
http://www.memtest.org/
next, boot with the memtest floppy/cd in to make sure it boots into that instead of your normal OS. once you have confirmed this, hit escape and then go into the bios.
what we're gonna do is find the tightest timings the ram can run at stock speeds. first, set all the timings to the rated speed of the ram, no auto's should be seen
2nd, make sure the memory ratio is set so that the memory is running at it's rated speed.
from here, we will work on one timing at a time.
start with cas latency (cas or tcl)
1. lower it one step, save bios and restart, boot into memtest
2. in memtest, change to test #5, loop it two or three times
3. if you get no errors, go back to 1, if you do, continue
4. raise timing one step, and go back into memtest, loop #5 4-5 times. you should not get errors
repeat for ras to cas delay (trcd), and ras precharge (trp). just.. leave tras to be cas + trcd + 2
write down the timings when you are done.
now, the where the real fun begins.
1. bump up the fsb or htt 2-3 mhz
2. boot into memtest, do 2 loops of test #5
3. if no errors, go to 1, if errors, continue
4. lower fsb/htt 1mhz
5. go into memtest, and let memtest run. if you get any errors, go back to 4. once you can loop memtest for ~3 hours per gigabyte in your system, write down the mhz obtained along with the timings and voltage it took to get there. i recommend making a chart for this
now we have two options: bump up memory voltage, or raise the timings. either way you go, do follow the above steps. you should always test things systematically.
start out at low voltage and test .1v higher, find max, .1v higher, find max, etc for the voltage. be very careful with voltage, as some can fail prematurely if you give it too much. always be aware of the temps your sticks are running at.
for the timings, you should change one timing at a time until you find which ones make a significant effect on max mhz. here's an example:
let's say i have some memory that's tightest timings at 200mhz is 2-2-2. i find the max at 217mhz with 2.6v.
i try the following combinations and find the max speed at each:
2.5-2-2
2-3-2
2-2-3
and i find that raising the cas latency gives me a very large bump in mhz (250mhz is the new max) while raising either trcd or trp only net me 2-3mhz.
then from 2.5-2-2, i would find the next set of timings that give me the highest bump, i test:
3-2-2
2.5-3-2
2.5-2-3
and this time, i find that the difference is less noticable, but raising trp raises the speed to 259mhz while cas latency and trcd don't get a single mhz
repeat along the lines of the above example until you reach 3-4-4. higher than this is not worthwhile testing.
when you're all done with your testing with both timings and voltage, you'll have a chart that looks somewhat like this:
if at any point along the way you find that you can't get higher mhz, try to figure out what is holding you back. it can be your cpu running at too high mhz, your memory controller maxing out, or your motherboard hitting a limit. just use some common sense and see if you can get around the limitation
hope that helps some. it's kinda the beginning of my memory overclocking guide i've been meaning to write forever now. any contructive feedback would be GREATLY appreciated.
also, this is just the tip of the iceberg for some boards. with the dfi's, you get the aforementioned 4 timings (cas, trcd, trp, tras) along with command rate (cpc), trc, trfc, trrd, twr, twty, trwt, tref, bank interleaving, the drive strength from the memory controller, the max async latency and read preamble, idle cycle limit, and ability to toggle the counter that limit controlls, along with read/write queue bypass, and the bypass max.
once you start messing with all those to fine tune an oc with the primary 3 timings (cas - trcd - trp), things get pretty intense. fortunately, there's a few rules of thumb you can follow
edit for future reference:
common performance ram chips..
Winbond BH: (old + new school)
basically the ram that has turned into a myth of sorts. normally does around 240-270mhz.. not great in itself, but the thing to remember is that it can do it at 2-2-2 timings. the downside is that it typically needs 3.2-3.5v to do that. also, higher latencies do not help much.. forget about cas3, it shouldn't boot. cas2.5 often isn't as stable, or simply does not get many more mhz
UTT-CH: (new ch)
known to be in OCZ VX, twinmosSP or mushkin redline. it's similar to bh-5, in the fact that it likes lots of voltage and generally likes 2-2-2 timings. needs more voltage than bh-5, but does not have some of the same issues with higher timings. also, CH seems to scale with voltage better than BH, but it needs around 3.1v to to get down to 2-2-2 to begin with. being a smaller process size, it's not as safe to pump CH dies as much voltage as BH. i wouldn't go over 3.4v for 24/7
Samsung TCCD/TCC5:
been around for a little while. had.. a cult like following for a while, which i find amusing. anywho, the premise of thise stuff is that you can do really high speeds (often 300mhz+ with good sticks) and doesn't need much voltage. the downside is that it needs high timings to get there. expect to have to drop down to 2.5-4-4 or 2.5-4-3 to get the most out of this ram. also be prepared to do a little bit of testing to find which voltage works best. some sticks actually start decreasing performance if you give them more than 2.7-2.8v, while some can take up to 3v. with newer tccd/tcc5, it seems that more than 2.8v will hurt then after prolonged periods of time, while the older chips can take a bit more, and actually scale really well at cas2 with lots of voltage, but die really fast if you do that.
Micron 5b G:
this stuff is a mix between BH-5 and TCCD. with properly binned sticks, it can do 2-2-2 timings up to a moderate speed (typically 220-225mhz without going nuts with voltage). the sweet spot is usually with 2.5-2-2; 250mhz at 2.7-2.9v and 2.5-2-2 seems typical with 5b G, with more possible with additional voltage. also scales with timings too, though not as well as tccd does, with 3-3-3 being best for hitting highest mhz (around 280-310 usually) 5b G can take 3.1-3.3v pretty reliably when properly cooled, though it will cut into it's lifespan a bit.
if you get it in value ram, simply subtract ~10-40mhz from the above numbers
Micron 5b D
very very similar to 5b G, but being a smaller process (110nm), it doesn't like more than ~2.9-3v in most cases.
first off, get memtest and install it on a floppy or cd
http://www.memtest.org/
next, boot with the memtest floppy/cd in to make sure it boots into that instead of your normal OS. once you have confirmed this, hit escape and then go into the bios.
what we're gonna do is find the tightest timings the ram can run at stock speeds. first, set all the timings to the rated speed of the ram, no auto's should be seen
2nd, make sure the memory ratio is set so that the memory is running at it's rated speed.
from here, we will work on one timing at a time.
start with cas latency (cas or tcl)
1. lower it one step, save bios and restart, boot into memtest
2. in memtest, change to test #5, loop it two or three times
3. if you get no errors, go back to 1, if you do, continue
4. raise timing one step, and go back into memtest, loop #5 4-5 times. you should not get errors
repeat for ras to cas delay (trcd), and ras precharge (trp). just.. leave tras to be cas + trcd + 2
write down the timings when you are done.
now, the where the real fun begins.
1. bump up the fsb or htt 2-3 mhz
2. boot into memtest, do 2 loops of test #5
3. if no errors, go to 1, if errors, continue
4. lower fsb/htt 1mhz
5. go into memtest, and let memtest run. if you get any errors, go back to 4. once you can loop memtest for ~3 hours per gigabyte in your system, write down the mhz obtained along with the timings and voltage it took to get there. i recommend making a chart for this
now we have two options: bump up memory voltage, or raise the timings. either way you go, do follow the above steps. you should always test things systematically.
start out at low voltage and test .1v higher, find max, .1v higher, find max, etc for the voltage. be very careful with voltage, as some can fail prematurely if you give it too much. always be aware of the temps your sticks are running at.
for the timings, you should change one timing at a time until you find which ones make a significant effect on max mhz. here's an example:
let's say i have some memory that's tightest timings at 200mhz is 2-2-2. i find the max at 217mhz with 2.6v.
i try the following combinations and find the max speed at each:
2.5-2-2
2-3-2
2-2-3
and i find that raising the cas latency gives me a very large bump in mhz (250mhz is the new max) while raising either trcd or trp only net me 2-3mhz.
then from 2.5-2-2, i would find the next set of timings that give me the highest bump, i test:
3-2-2
2.5-3-2
2.5-2-3
and this time, i find that the difference is less noticable, but raising trp raises the speed to 259mhz while cas latency and trcd don't get a single mhz
repeat along the lines of the above example until you reach 3-4-4. higher than this is not worthwhile testing.
when you're all done with your testing with both timings and voltage, you'll have a chart that looks somewhat like this:
Code:
2-2-2 2.5-2-2 2.5-2-3 2.5-3-3 3-3-3
2.6 216.46 250.51 258.84 268.68 295.17
2.7 219.48 253.54 263.38 273.22 299.71
2.8 224.02 258.08 266.41 277.00 302.73
2.9 226.29 260.35 269.43 279.27 304.25
3 229.32 262.62 273.22 283.81 305.76
3.1 231.59 264.89 275.49 287.60 305.00if at any point along the way you find that you can't get higher mhz, try to figure out what is holding you back. it can be your cpu running at too high mhz, your memory controller maxing out, or your motherboard hitting a limit. just use some common sense and see if you can get around the limitation
hope that helps some. it's kinda the beginning of my memory overclocking guide i've been meaning to write forever now. any contructive feedback would be GREATLY appreciated.
also, this is just the tip of the iceberg for some boards. with the dfi's, you get the aforementioned 4 timings (cas, trcd, trp, tras) along with command rate (cpc), trc, trfc, trrd, twr, twty, trwt, tref, bank interleaving, the drive strength from the memory controller, the max async latency and read preamble, idle cycle limit, and ability to toggle the counter that limit controlls, along with read/write queue bypass, and the bypass max.
once you start messing with all those to fine tune an oc with the primary 3 timings (cas - trcd - trp), things get pretty intense. fortunately, there's a few rules of thumb you can follow
edit for future reference:
common performance ram chips..
Winbond BH: (old + new school)
basically the ram that has turned into a myth of sorts. normally does around 240-270mhz.. not great in itself, but the thing to remember is that it can do it at 2-2-2 timings. the downside is that it typically needs 3.2-3.5v to do that. also, higher latencies do not help much.. forget about cas3, it shouldn't boot. cas2.5 often isn't as stable, or simply does not get many more mhz
UTT-CH: (new ch)
known to be in OCZ VX, twinmosSP or mushkin redline. it's similar to bh-5, in the fact that it likes lots of voltage and generally likes 2-2-2 timings. needs more voltage than bh-5, but does not have some of the same issues with higher timings. also, CH seems to scale with voltage better than BH, but it needs around 3.1v to to get down to 2-2-2 to begin with. being a smaller process size, it's not as safe to pump CH dies as much voltage as BH. i wouldn't go over 3.4v for 24/7
Samsung TCCD/TCC5:
been around for a little while. had.. a cult like following for a while, which i find amusing. anywho, the premise of thise stuff is that you can do really high speeds (often 300mhz+ with good sticks) and doesn't need much voltage. the downside is that it needs high timings to get there. expect to have to drop down to 2.5-4-4 or 2.5-4-3 to get the most out of this ram. also be prepared to do a little bit of testing to find which voltage works best. some sticks actually start decreasing performance if you give them more than 2.7-2.8v, while some can take up to 3v. with newer tccd/tcc5, it seems that more than 2.8v will hurt then after prolonged periods of time, while the older chips can take a bit more, and actually scale really well at cas2 with lots of voltage, but die really fast if you do that.
Micron 5b G:
this stuff is a mix between BH-5 and TCCD. with properly binned sticks, it can do 2-2-2 timings up to a moderate speed (typically 220-225mhz without going nuts with voltage). the sweet spot is usually with 2.5-2-2; 250mhz at 2.7-2.9v and 2.5-2-2 seems typical with 5b G, with more possible with additional voltage. also scales with timings too, though not as well as tccd does, with 3-3-3 being best for hitting highest mhz (around 280-310 usually) 5b G can take 3.1-3.3v pretty reliably when properly cooled, though it will cut into it's lifespan a bit.
if you get it in value ram, simply subtract ~10-40mhz from the above numbers
Micron 5b D
very very similar to 5b G, but being a smaller process (110nm), it doesn't like more than ~2.9-3v in most cases.
(cf)Eclipse
Freelance Overclocker
- Joined
- Feb 18, 2003
- Messages
- 30,027
hell yeah. it's gonna take a while till it rivals what my a64 oc guide looks like thoughHDBox2d1 said:"Now now, the where the real fun begins."
haha
thats alot of writing
thanks for the pimpage sky <3
(cf)Eclipse
Freelance Overclocker
- Joined
- Feb 18, 2003
- Messages
- 30,027
nah, it needs a bit of work still, and i want it to go in the memory forumMrSlacker said: