PI Black 4GB (2x2) OC'ing help

[MarcAntony]

Hi peeps,

I built this PC around 8 months ago and was immediately able to attain an FSB of 1600, link/sync the RAM (Mushkin at the time) to run at 800mhz with 4.5.4.11 timings. All I play is WoW, so with it being such an incredibly CPU intensive game, the CPU frequency is critical. With the 1600FSB and CPU/RAM linked/synced it was humming along and was even able to maintain 40fps in Dalaran (most populated city in the game ? can crunch an i7 without blinking) with 60-80 addons running.

But one of the sticks went bad and so I went on Newegg and saw this set and how it runs at such appealing voltages and ordered the set. I initially linked/sync?ed it at 1600FSB but left the timings at 5.5.5.15 for the first few days to make sure I was stable. I then bumped the voltage to 1.9v and set them at 4.4.4.12 and was rockin for about a week before I started to have problems.

WoW began to app crash citing error #132 (RAM error) quite frequently. I loosened up the timings and became much more stable, but still crashed from time to time. Finally, I backed the FSB down to 1580, so the RAM is now running at 790mhz. At this FSB, I am now using 4.4.4.10 timings and have been stable for a month with this setting.

But getting back to wow being so CPU hungry, I am responsible for Frapsing my guild?s raids and raiding wow with all these addons and Fraps recording is extremely demanding on any PC and mine is struggling to maintain 30fps (I record Fraps Full Size @ 30fps) in the middle of the encounters. In these situations, every single FPS is critical (and not to mention, the movies look better too if they record at a steady 30fps).

So, my question to you guys is, if you were me what would you try to do? Do you believe the RAM is what?s struggling to run at its rated speed or is it the mobo? I?ve never tried giving the RAM more than 1.9v, but I have tremendous cooling in this case, so heat should not be an issue. Also, for a game like wow, will having a higher RAM frequency make for higher fps or having the timings tighter be more beneficial?

I would like to just run everything nice and neat at 1600FSB (L/S) with timings at the stated 4.4.4.10 ideally, but I?m up for suggestions. Thanks for your help!

[genetix]

This memory can run 4-4-3-10 at 900Mhz. and sure it won't be in RAM error at 4-4-4-12 with 1.90 volts at any case or scenario.

* Check that modules are in different controllers (like A1 & B1 or A2 & B2)
* Check that the Voltage really is 1.9v with Windows something like Everest sensor pages.
* Check that Northbridge voltages are good. (not sure what kinda board you're running but P35, X38, X48, P45 1.400v-1.440v NB would be fine)
* Run an MemTest86+, If it errors on 4-4-4-12 would probably be that one of the memory modules is simply bad.


final note: 4-4-4-10 would make no sense.. something like 4-4-3-14 and with 'Different Row' read to write/write to write/read to read -1T from stock, (dunno what board you got, but some boards does Performance Leveling incorrectly. So, check those should be around 7T-9T 'if below modify it'.) would make more ideal, but with 2 sticks should work either way.

[MarcAntony]

Okay, I got some of that, but you went over my head with some as well =p

I don't know why I didn't list my system specs. Could've sworn I did.... Was probably a pretty bad read without knowing the system. Specs below:

Q9550 @ 3.35ish
EVGA 780i SLI (1580 FSB)
4GB PI Black (2x2GB) ** Just ordered another set and will be going to 8GB this weekend **
EVGA GTX275 FTW
x2 WD Raptors (Raid 0)
x1 WD Caviar Black 500GB
x1 WD My Book 1TB
Thermaltake Toughpower 1,000w

Ok, as for nonsensical timings lol, I have no doubts about it. I've self-taught myself everything I know about PCs simply by reading forums and asking questions and have never been able to really grasp or understand RAM as well as I need to. Its definately my weak spot for sure. I do have Everest on the PC and have double checked RAM voltage and it reads 1.9v. I got Everest because HWMonitor reads the vRAM at 1.6v. When I saw that, I was sure something was up and that was the reason for my issues. But someone on EVGA forums said HWMonitor isn’t totally accurate and to get Everest. Which I did, and it reads 1.9v. Nevertheless, figured I’d mention that HWMonitor reads the RAM voltage at 1.6v.

As mentioned above, I use the EVGA 780i board with the latest drivers. There is a comprehensive OC’ing guide for the board which I followed to attain the clocks I’m using. There is also another guide on their forums pertaining to GTLVREF and that with proper GTLVREF settings, you can easily attain a stable 1600FSB (L/S) at 1.2v to the NB. I followed that guide and so that is what my NB is set to. Knowing that most people would use 1.3v to attain 1600FSB, I did try to use 1.3v on the NB for a few days and while I did have slightly better stability, the app crash did persist. Since then, I’ve just been running everything like it is too busy to delve into finding a solution to the problem.

“final note: 4-4-4-10 would make no sense.. something like 4-4-3-14 and with 'Different Row' read to write/write to write/read to read -1T from stock, (dunno what board you got, but some boards does Performance Leveling incorrectly. So, check those should be around 7T-9T 'if below modify it'.) would make more ideal, but with 2 sticks should work either way.”

That is the portion of your reply that went over my head. Sorry, but when it comes to RAM, I’m still in a learning phase. And thank you so much for your help.

[genetix]

4GB PI Black (2x2GB) ** Just ordered another set and will be going to 8GB this weekend **

When you do this consider also that full controller will NOT handle same latencies. The memory won't be going to spec as only 2 sticks are guaranteed. Although I run 8GB on these timings.

Ok, as for nonsensical timings lol, I have no doubts about it. I've self-taught myself everything I know about PCs simply by reading forums and asking questions and have never been able to really grasp or understand RAM as well as I need to. Its definately my weak spot for sure. I do have Everest on the PC and have double checked RAM voltage and it reads 1.9v. I got Everest because HWMonitor reads the vRAM at 1.6v. When I saw that, I was sure something was up and that was the reason for my issues. But someone on EVGA forums said HWMonitor isn’t totally accurate and to get Everest. Which I did, and it reads 1.9v. Nevertheless, figured I’d mention that HWMonitor reads the RAM voltage at 1.6v.

Dunno where the limit goes to what is too much. but every DDR2 stick usual voltages are around 2.100v. Now I know this memory spins on even 1.8v to 900Mhz, but like to keep it around 2.000-2.100v myself just for the kick. I do not suggest the change, but just saying this for what I use. It ain't gonna heat up even on 2.3v with these cooler pins and even stock box cooling solution.

btw, everest uses HWmonitor as base of information. however I don't think the board is even capable of 1.6v, lol.

As mentioned above, I use the EVGA 780i board with the latest drivers. There is a comprehensive OC’ing guide for the board which I followed to attain the clocks I’m using. There is also another guide on their forums pertaining to GTLVREF and that with proper GTLVREF settings, you can easily attain a stable 1600FSB (L/S) at 1.2v to the NB. I followed that guide and so that is what my NB is set to. Knowing that most people would use 1.3v to attain 1600FSB, I did try to use 1.3v on the NB for a few days and while I did have slightly better stability, the app crash did persist. Since then, I’ve just been running everything like it is too busy to delve into finding a solution to the problem.

Don't really care what some guide says. Although, I do understand nforce 780i chipset is a bit different than general intel chipsets, however, if it ain't stable it ain't stable and 1.2-1.3 is pretty far from high even for 780i (People out there are overclocking this board with SPP(northbridge) around 1.5v and FSB's around 1.4v which is a bit much but just for an example). I know that at 400Mhz FSB board is gonna need some volts and controller is gonna need some volts. That's a fact. Only true testing you got are the MemTest86+ to test if the memory is does error itself and on windows LinPack or Prime95 to see if the memory errors when all equipment are in use.

“final note: 4-4-4-10 would make no sense.. something like 4-4-3-14 and with 'Different Row' read to write/write to write/read to read -1T from stock, (dunno what board you got, but some boards does Performance Leveling incorrectly. So, check those should be around 7T-9T 'if below modify it'.) would make more ideal, but with 2 sticks should work either way.”

That is the portion of your reply that went over my head. Sorry, but when it comes to RAM, I’m still in a learning phase. And thank you so much for your help.

I just ment that. tCL=4 tRCD=4 tRP=3 tRAS=14 would make more sense than 4-4-4-10. and tRD as in perfomance level should be around 8, if your BIOS has such and on top of that on cosmetic tweaking getting everything out from RAM smaller values (if there is such) tRTR, tWTW and tWTR could even be lowered by 1T to different memory row/slots, but this isn't too important as said just cosmetic changes..

[MarcAntony]

1) Yeah, I know you can't, in theory, get as aggressive with 8GB as you can 4GB. This is why I've stuck it out with 4GB as long as I have. But I have dual monitor set-up and like to have browser, wmp, etc open while I'm gaming (unless I'm Frapsing) and I'm banging my head against the 4GB max.

2) Really lol? Thats weird. But as for Voltages, that's good to know. Since getting started in OC'ing, I have always err'ed to the side of caution (as I'm sure you can tell) when it comes to voltages. I'm certain that my system and components can take more. I just have to all but certain its ok before I begrudgingly up the voltages =p

3) Right now, with my current state of things (1580 FSB & L/S), I'm able to pass Linpack, OCCT, etc just fine. But if I go to 1600FSB, I can't pass Linpack. I don't know if its purely the fact that I need more power to the RAM, SPP, and FSB or that maybe my board has a black hole at 1600. I've read where some boards will have black holes. Only problem with me is I'm not daring enough to test much higher than these speeds to find out if that's the case here.

4) Ok, I think I gotcha this time. I did change the timings to 4.4.3.14 as you recommended and gamed last night at those timings for 4+ hours and she ran like a champ. But this is still at just 1580 of course.

As for the other settings you're talking about, are you recommending I lower them each by a value of 1? I've always left those on Auto, but I can certainly go in and change the values.

[genetix]

1) Yeah, I know you can't, in theory, get as aggressive with 8GB as you can 4GB. This is why I've stuck it out with 4GB as long as I have. But I have dual monitor set-up and like to have browser, wmp, etc open while I'm gaming (unless I'm Frapsing) and I'm banging my head against the 4GB max.

Yeah, some apps takes alot of memory specially graphics and 8 GBs with these latencies ain't too bad, but I would still reconsider of buying for example Windows 7 won't use memory like Vista does when you got alot of similar apps on screen, but of course it might be still too much dunno what you are running.

2) Really lol? Thats weird. But as for Voltages, that's good to know. Since getting started in OC'ing, I have always err'ed to the side of caution (as I'm sure you can tell) when it comes to voltages. I'm certain that my system and components can take more. I just have to all but certain its ok before I begrudgingly up the voltages =p

Well, todays hardware is like made for 35% OC.. Caution on 50% What goes to voltages yeah, there's some really bad ones to OC when temperature to consider, but good monitoring wins this one.

3) Right now, with my current state of things (1580 FSB & L/S), I'm able to pass Linpack, OCCT, etc just fine. But if I go to 1600FSB, I can't pass Linpack. I don't know if its purely the fact that I need more power to the RAM, SPP, and FSB or that maybe my board has a black hole at 1600. I've read where some boards will have black holes. Only problem with me is I'm not daring enough to test much higher than these speeds to find out if that's the case here.

Well, I wanted to see how this memory stands on these conditions. I actually pumped the FSB to 1800 to create same scenario as your 1600Mhz would be. You sure that the CPU & NB is stable and it would in-any-case be the memory. anyway, I actually dropped an quick test on these on values with 8 sticks on board I am/was recommending to you picture speaks thousand words as they say:



(This is absolute stock settings on my board on latencies except 4 first CLs. Voltages as I use them 24/7 These voltages can do hell alot more than 900Mhz, tho. btw, did you know some boards drops something crazy like full voltages automatically while overclocking now that's crazy. O an in case you wonder on why 'custom' 2048MB well new IntelBurnInTest 2.3 has bug on 'High' and 2048MB is amount to go through 2 blocks simultanously so that's Channel A1&A2 providing hardest stress on memory controller there is (well comparable to Maximum).)

4) Ok, I think I gotcha this time. I did change the timings to 4.4.3.14 as you recommended and gamed last night at those timings for 4+ hours and she ran like a champ. But this is still at just 1580 of course.

As for the other settings you're talking about, are you recommending I lower them each by a value of 1? I've always left those on Auto, but I can certainly go in and change the values.

Best idea would be to leave anything that says 'same row' or 'same rank/bank' alone also same to tRRD, TRFC, tRTP. Rest of em yeah. Think of it as 'generic tCL values & plan B when everything fails and you have to modify the general CLs then the smaller values comes to count or in case you get stable compination and then wanna get some more out of board.

[MarcAntony]

How much voltage to you give to the FSB to get that OC? And how much to the SPP?

And btw, I did confuse my acronyms earlier when I was talking about setting the GTLVREF lanes. I was meaning to say that you can get away with a 1.2 FSB voltage at 1600FSB IF you can accurately set the GTLVREF. I say this because that seemed to be the major source of my problem. I decided to set the GTLVREF lanes back to auto and up the FSB voltage to 1.3v.

After doing so, I am able to attain the 1600fsb, RAM linked/synced, with timings of 4.4.3.14. Things are running real smooth. Thing is, I got greedy (lolol) and tried to maintain linked/synced with a 1666fsb and the PC locked up 10 minutes into OCCT.

Now my heart is set on 1666 =p. I only tried once, but something is telling me that I may have not been giving the CPU enough on the Vcore. I have it set to 1.3675 in BIOS, which is actually just 1.31 in Windows due to Vdroop. I'm considering performing the Vdroop pencil mod on the mobo and trying again this weekend.

Do you have any recommendations for me? You've now proven to me that the RAM is not holding me back from going above 3.5ghz (l/s) and with wow being my game, that would prove to be quite beneficial. Idunno, maybe I should be happy with it hehe. Its running wow very smoothly now. Its just Frapsing where something is bottlenecking me and my framerates drop down to 15-18.

Considering my framerates are as high as 400fps in certain areas/situations in wow, I just feel there is a way to maintain a solid 30fps in a raid setting, while frapsing.

[genetix]

Lets rewrite a bit... Wrote a damn novel.

400FSB will need around 1.40v exact (Actual voltage) on SSP. FSB VTT/FSB will need around 1.188v stable on stress. vCore depends on 3200Mhz (400FSB-1600Mhz) would need only around 1.200v (actual voltage), ~3,8Ghz would be around 1.324v (Idle). Always calculate the FSB VTT to vCore it needs to sync. vCore under LinPack stress = x.xxxv(actual volts) + ~0.032v = FSB VTT is calculation I use.

You should aim to 3,8Ghz = 8.5x450FSB(1800Mhz) = 900Mhz memory.

This would look something like this to be fully stable at those speeds:

(SPP would in this case be around 1.440v-1.520v depending how PSU rails stay stable, FSB VTT you can read from above same calculation, vCore however you will need that 1.36v on 3.8Ghz probably but test it with the 1.324v before this since +0.5 Multipliers doesn't count same way as raising FSB to get core stable)



(That is Q9450 at 3632Mhz-454FSB(1816Mhz) with 8GB of PI-B spinning on 909Mhz(BIOS). North bridge sensor I got attached here separately shows 37 degrees, but I got extra small fan (as in picture ~4000RPM) attached on top of NB. With out the fan it'll stay around 40 degrees on room temps around 28c depending of course what kinda box you have build.)

Also you mentioned OCCT, if I were you I wouldn't count on that too much. OCCT is something really bad for testing stability as whole software stability is in questionable. Use Prime95, IntelBurnInTest (Linpack), MemTest86+ before starting to know that RAM is stable and on OS even if Linpack actually success you you 2 different style of memory read/writes to verify 'Write NT' which is NT based writing through and with Caches and 'Write' which is old style writing with memory only, Read thought SSEs and Read with Paging included.

Even while you would get en basic Everest Write NTs to punch you an decent speeds that's not all of it as your softwares/games might actually restrict that to with paging read + old style writing.

Edit #01:

Anyway, you can post me an similar screenshot (Everest v4.60 you can get for free on net). I can check against actual data what's going on. Needs an ~5 seconds full stress to get voltage readings on screen and you to copy me that screen & Idle screen would be needed.

[MarcAntony]

Fwiw, right now, my voltages are as follows:

FSB: 1.3v (I'm too much of a wimp to go to 1.4v - not gonna happen =)
SPP: 1.4v
MCP:1.525v
HT: 1.25v
RAM: 2.00v

GTLVREF Lanes: Auto

All timings in the "lower section" below the main 5, are all set to Auto. I can take a screenshot to get their exact values if you'd like to see what the board is setting them to.

Lemme know if there's anything else noteworthy. Also, under load, my NB temps get as high as 52c and the MCP gets as 48c. At idle, they are 50c and 45c, respectively.

Are those low, average, or high in your opinion? That is something I've really been wondering about is the temps of other people's NB and SB temps

[MarcAntony]

I posted that before I saw your reply.

I can only change my VTT in incriments of .1. Its either 1.2 (grey text), 1.3 (green text), 1.4 (green text), or 1.5 (red text).

I use a Q9550, Yorkfield CPU. That's why 1600fsb is such a popular config for me. Going to 1600fsb sets the CPU at 3.40 and the RAM at 800mhz linked/synced. But I'm feigning for a lil more =p.

But I do NOT want to break any eggs to make this omelette ; ) I'm not one of those guys who can fry a CPU or Mobo and just install another like it was nothing. I'd be a bad thing if I did that.

[genetix]

check my last edit and post as I said you are not gonna need FSB over 1.26v(1.2v/1.3v)....

Edit:

NB should never get over 45 degrees even on stock cooling and full stress.

Edit #02:

400Mhz aka. 1600Mhz FSBs are bad because you will get the extra memory stress and you cannot clock your memory accordingly 800Mhz on DDR2 simply ain't good on 1:1 on this case you could clock your DDR2 better at 4:3 or 5:6 ratios because those will automate tRD (Read Delay/what read speed is all about) to +1T with Static Reading of memory enabled. on FSB 1.3v would be correct(use 1.2v on 400FSB), but you will need to clock your vCore accordingly to that. Check what FSB VTT is at OS and see as I said above calculations to match. It cannot be too low and 1.36v on vCore for 1.30v FSB VTT is too high (unless it has an droop on there somewhere.).

Just copy me the screen from OS as I did from Everest. we'll see what it should be and what is wrong.

Edit #3:

and that MCP is way too high.. work that on later when I get some accurate readings other that what you tell me what your BIOS is reading. Please post the 2 (stress/idle) screens from 'Voltage/Sensor'-pages. So we can advance.

[MarcAntony]

Been outta town, but am back now.

I'll get those screenshots up when I get home tonight.

[MarcAntony]

http://img43.imageshack.us/img43/126/voltages.jpg

http://img43.imageshack.us/img43/6041/timings.jpg

http://img43.imageshack.us/i/ramtest.jpg/


Herrrrre we go. Sorry that took so long.

[GSKILL TECH]

Looks good. Running stable now? Read test is pretty good for DDR2 800.

GSKILL SUPPORT

[MarcAntony]

Really? Hehe, just kind of assumed my configuration wouldn't be optimal since I barely know what I'm doing. And yes, I am perfectly stable at this configuration. I did ask some of the people on EVGA forums about my SPP and MCP temps and they said those temps were actually very good. I guess it really varies from different kinds of boards.

I read about how the 780i SPP runs hot and there's a couple huge guides on replacing the stock cooling with a new heatsink and fan.

But if I'd be hard pressed to get much better performance, I'm wondering if I should just stay 1:1 at 1600fsb for now and shift my focus to replacing the NB cooling before attempting a higher fsb.


edit: "temps were actually very good" meant "actually very good for a 780i" =p