-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
Alpha Timings and the NF7-S
- Thread starter felinusz
- Start date
dont worry felinusz, my 3d stability is in the dumper too! haha.
i dont do any 3d work, but i was like heck, lemme fire up 3dmark 2001 se and see wha happens. the nerve of it! kickin me back to desktop. who does it think it is!
back to the good ole scratch pad (or shall we say OC from hell! hehe).
i dont do any 3d work, but i was like heck, lemme fire up 3dmark 2001 se and see wha happens. the nerve of it! kickin me back to desktop. who does it think it is!
back to the good ole scratch pad (or shall we say OC from hell! hehe).
- Joined
- Feb 3, 2004
- Location
- Ontario, CAN
Very interesting thread... Thanks for putting your findings up for the rest of us to see.
Like so many others, I have had a hard time getting my NF7S v2 to a higher FSB. For the longest time, the highest I could get was 210 (11-3-3-2) and that was with maxing out the voltage to the memory at 2.9v. The highest I could get with 2.7v was 207 (2.8 didn't make a difference), which was 10 mhz lower than what my OCZ PC3500 EL was suppose to be able to do, so I was starting to wonder if my memory was bad or if I just got a dud of a board.
I started to do some research on this, leading me to try out some of the different moded bioses, but didn't make any difference on my max FSB. Then, I finally did the L12 mod since I have a mobile Barton and it seems to work as it shows up as 166 FSB when I load up the optimized defaults, but it only got me to 213 stable. I use Memtest, then Prime 95 as the first two steps in testing for stabilty. With my system, P95 would either run for hours or fail in the first 1-10 minutes and I find that when it fails, it will generally show signs of instabilty while running other tests, programs etc. I was able to get a little higher with looser timings but the increase of a few mhz on the FSB was not enough to offset the lose of performance with the higher timings.
After reading this thread, in my quest for a higher FSB, I decided to give this a try. First, I tried one of the combinations from the first post that was a little looser than the default ones and @ 214 it failed. So, I decided to just loosen the timings as high as they would go, leaving T(W2P) and T(R2P) at 4, as suggested in a post above and see if this would make a difference and to my supprise it did.
I set the multiplier to 10 @ 1.7 vcore, 2.9 vdimm, 1.6 nb, with memory timings of (11-3-3-2) and alpha timings of (6-6-4-6-3-4-6) and at 214 FSB it ran P95 without an error, so I bumped it up another notch. Now I know that one hour of P95 doesn't meet the strictest standards of stability, but I just want to get a general idea, so I can keep moving it up til it won't go any further... then, I'll drop it a back a little bit and then perform some longer, more thorough stability tests.
This is where it gets kinda interesting... as I got it to pass at least one hour of P95 at 215, 217, 219 and 221 along with no signs of instabilty during benchmarks, normal use and some gaming along the way. This was all with the settings/timings listed above. So, I decided to up the multiplier to 12 @ 1.925v with 221 FSB to do some benchmarking and gaming before I let P95 run all night which ended up being over 8 hours without an error.
I did some more reading and noticed that T(W2R) was one setting that was usually a lot lower than the others, so I dropped it to the default setting of 3, so the alpha timings were (6-6-4-3-3-4-6) and P95 failed right away. I upped T(W2R) to 4 and it ran fine, so on a hunch I changed all the settings back to default except T(W2R) which remained at 4 and everything has run fine since, even after bumping up the FSB to 223. So, after all this time, it was just a matter of making that one change in the alpha settings that seems to have made all the difference in my NF7S being able to run a lot higher FSB... and I'm not even done yet!
So here's some memory benchies at my current speed...
10x223 @ 1.7,2.9,1.6 (11-3-3-2) (2-3-5-4-2-2-3)
Sandra Mem - 3412/3184 @ 96%
Everest: Latency - 76.1ns, Read - 3418, Write - 1306
And I'll get a bit of a boost when I up the multiplier to 12.
I know this post is kinda long but I know that there are a lot of people out there in the same boat as I was and are unaware of the possible benefits of changing the alpha timings with this great little program.
Anyways, I just want to thank those that have done the tinkering and figuring this out and those that have taken the time to share their experiences and let others know about this.
~ Racefan20
Like so many others, I have had a hard time getting my NF7S v2 to a higher FSB. For the longest time, the highest I could get was 210 (11-3-3-2) and that was with maxing out the voltage to the memory at 2.9v. The highest I could get with 2.7v was 207 (2.8 didn't make a difference), which was 10 mhz lower than what my OCZ PC3500 EL was suppose to be able to do, so I was starting to wonder if my memory was bad or if I just got a dud of a board.
I started to do some research on this, leading me to try out some of the different moded bioses, but didn't make any difference on my max FSB. Then, I finally did the L12 mod since I have a mobile Barton and it seems to work as it shows up as 166 FSB when I load up the optimized defaults, but it only got me to 213 stable. I use Memtest, then Prime 95 as the first two steps in testing for stabilty. With my system, P95 would either run for hours or fail in the first 1-10 minutes and I find that when it fails, it will generally show signs of instabilty while running other tests, programs etc. I was able to get a little higher with looser timings but the increase of a few mhz on the FSB was not enough to offset the lose of performance with the higher timings.
After reading this thread, in my quest for a higher FSB, I decided to give this a try. First, I tried one of the combinations from the first post that was a little looser than the default ones and @ 214 it failed. So, I decided to just loosen the timings as high as they would go, leaving T(W2P) and T(R2P) at 4, as suggested in a post above and see if this would make a difference and to my supprise it did.
I set the multiplier to 10 @ 1.7 vcore, 2.9 vdimm, 1.6 nb, with memory timings of (11-3-3-2) and alpha timings of (6-6-4-6-3-4-6) and at 214 FSB it ran P95 without an error, so I bumped it up another notch. Now I know that one hour of P95 doesn't meet the strictest standards of stability, but I just want to get a general idea, so I can keep moving it up til it won't go any further... then, I'll drop it a back a little bit and then perform some longer, more thorough stability tests.
This is where it gets kinda interesting... as I got it to pass at least one hour of P95 at 215, 217, 219 and 221 along with no signs of instabilty during benchmarks, normal use and some gaming along the way. This was all with the settings/timings listed above. So, I decided to up the multiplier to 12 @ 1.925v with 221 FSB to do some benchmarking and gaming before I let P95 run all night which ended up being over 8 hours without an error.
I did some more reading and noticed that T(W2R) was one setting that was usually a lot lower than the others, so I dropped it to the default setting of 3, so the alpha timings were (6-6-4-3-3-4-6) and P95 failed right away. I upped T(W2R) to 4 and it ran fine, so on a hunch I changed all the settings back to default except T(W2R) which remained at 4 and everything has run fine since, even after bumping up the FSB to 223. So, after all this time, it was just a matter of making that one change in the alpha settings that seems to have made all the difference in my NF7S being able to run a lot higher FSB... and I'm not even done yet!
So here's some memory benchies at my current speed...
10x223 @ 1.7,2.9,1.6 (11-3-3-2) (2-3-5-4-2-2-3)
Sandra Mem - 3412/3184 @ 96%
Everest: Latency - 76.1ns, Read - 3418, Write - 1306
And I'll get a bit of a boost when I up the multiplier to 12.
I know this post is kinda long but I know that there are a lot of people out there in the same boat as I was and are unaware of the possible benefits of changing the alpha timings with this great little program.
Anyways, I just want to thank those that have done the tinkering and figuring this out and those that have taken the time to share their experiences and let others know about this.
~ Racefan20
- Joined
- Apr 3, 2004
Bottom line:
T(W2R) 3 -> 4.
thanks for the input.
On a side note:you FSB is either due to RAM limitations or a bad L12 mod.
There are different versions of this mod , the XP-M version that worked for me looks like this:
_ . . .
\ . . . .
. . . . .
. . _____
. . |
. . | socket
|
v
RAM
I think it's the 200 FSB one.
T(W2R) 3 -> 4.
thanks for the input.
On a side note:you FSB is either due to RAM limitations or a bad L12 mod.
There are different versions of this mod , the XP-M version that worked for me looks like this:
_ . . .
\ . . . .
. . . . .
. . _____
. . |
. . | socket
|
v
RAM
I think it's the 200 FSB one.
bad L12? never heard of that one b4 (unless of course your desired force doesnt register, then i guess it would be considered bad, hehe).
ah30 to ah32 or ah28 to ah30 will force 166 on an XP-M. you cant force 200 on an XP-M, due to 1st and 3rd L12 being closed already. older bios revisions will detect as 200, newer ones, as 166.
ah30 to ah32 or ah28 to ah30 will force 166 on an XP-M. you cant force 200 on an XP-M, due to 1st and 3rd L12 being closed already. older bios revisions will detect as 200, newer ones, as 166.
- Joined
- Jan 10, 2001
AH30 to AH28 is the only one you should use to force 200 fsb detection on the mobile.Using both connections my mobo was really flakey.I am using the 26 official bios and it does detect it at 200 fsb.I was unsure what was wrong until I found this site which was easy to see the connections for all chips.
I am going to try raising my t(w2r) to four as I have hit my stable wall at 217.Anything higher and it fails P95 but is memtest stable to 223.Great thread.
I am going to try raising my t(w2r) to four as I have hit my stable wall at 217.Anything higher and it fails P95 but is memtest stable to 223.Great thread.
- Thread Starter
- #28
I am very glad to hear that this helped you out
.That is quite a signifigant gain that you achieved with your board, and totally free as well - more power to you!
Also, WELCOME TO THE FORUMS!, enjoy your stay!
I have found, with the 5 different socket A processors that I have run with the L12 mod in my old NF7-S, that using a copper wire to do a 'socket pin mod' often has no effect, a "bad L12 mod" - even though the wire was in there
. I pretty much started doing all of my processor "pin mods" with conductive ink, painted on the bottomside of the processor, physically bridging the pins.
Runner30 - If I understand the ASCII art correctly, that's the same L12 mod that gave me the best results with my particular motherboard. The other pin connection you can use didn't do anything (except detect the processor's stock FSB as 166 MHz) for me.
hey hey hey.. long time no chat guys..
remember the same bh5 that was in the nf7 that would barely do ~239-240, 2-2-2-11/5, cmd rate 2T? its doing ~250-251, 1.5-2-3/2cmd rate 1T in the dfi nf4.
that goes to show ya what a difference the memory controller can make surprisingly, cl2 is a bit more unstable then cl1.5 is for my cpu's mem controller. go figure, hehe.
remember the same bh5 that was in the nf7 that would barely do ~239-240, 2-2-2-11/5, cmd rate 2T? its doing ~250-251, 1.5-2-3/2cmd rate 1T in the dfi nf4.
that goes to show ya what a difference the memory controller can make
surprisingly, cl2 is a bit more unstable then cl1.5 is for my cpu's mem controller. go figure, hehe.
Last edited:
- Joined
- Aug 15, 2001
- Location
- Toronto,Canada(I can see you....)
pershoot said:hey hey hey.. long time no chat guys..
remember the same bh5 that was in the nf7 that would barely do ~239-240, 2-2-2-11/5, cmd rate 2T? its doing ~250-251, 1.5-2-3/2cmd rate 1T in the dfi nf4.
that goes to show ya what a difference the memory controller can make
Nice job there Pershoot.
definitely not. 1.5(CL)-2(tRCD)-3(tRP) is faster then 2(CL)-2(tRCD)-2(tRP). the ultra low latency from 1.5 alone is a show stopper. the memory can do 1.5-2-2 @ ~ 250 (memtest ok) but the memory controller has an issue with tRP of 2 this high up.
im still workin on 100% stability, but im gettin there (so many settings and they all have a relationship on one another.. so will take some time to get it just right..).
im still workin on 100% stability, but im gettin there (so many settings and they all have a relationship on one another.. so will take some time to get it just right..).
- Joined
- Apr 3, 2004
I'm still not sure 1.5-2-3 is better than 2-2-2 at same speed.
Sandra benchmark isn't affected much by timings, so it's pointless to check.
However if you run memtest, can you compare the "ram speed" numbers?
I think they're using some sort of random seek algorythm there.
Sandra benchmark isn't affected much by timings, so it's pointless to check.
However if you run memtest, can you compare the "ram speed" numbers?
I think they're using some sort of random seek algorythm there.
in terms of memtest reported throughput, you are correct. tRP of 3 chops off approx ~85-90mb/s versus tRP of 2. 2-2-2 and 1.5-2-2 nets the same throughput in terms of memtest report.
however, we are not only measuring throughput here, but also measuring speed of access. cl1.5 is of lower latency then cl2, hence faster in that realm within reason (i.e. having 4-4 with cl1.5 is obviously going to be slower then 2-2-2). that and cl2 is a bit more unstable for me, ha.
im still testing, and trying to 100% it. i will try to tRP 2 it as well, but i have a feeling my memory controller wont be able to cope. it may be able to cope if i lower my resulting clock (which id rather not do). so a give and take on both sides, ya know.. the art of oc'ing, hehe.
however, we are not only measuring throughput here, but also measuring speed of access. cl1.5 is of lower latency then cl2, hence faster in that realm within reason (i.e. having 4-4 with cl1.5 is obviously going to be slower then 2-2-2). that and cl2 is a bit more unstable for me, ha.
im still testing, and trying to 100% it. i will try to tRP 2 it as well, but i have a feeling my memory controller wont be able to cope. it may be able to cope if i lower my resulting clock (which id rather not do). so a give and take on both sides, ya know.. the art of oc'ing, hehe.
