swimtech

I haven't had much time the past week since my last post in This thread where I was breaking in the machine at stock speeds.

I was able to get some overclocking goodness going though, along with a few other tweaks too.

The following is for bios voltage set a 1.375 which of course drops some under load. Memory is running at 2.1 volts, 4-4-4-12 latencies slightly overclocked to 427Mhz (400 stock) with a 5:6 ratio. The rest is in the screenie below:


Shot at 2008-05-18

The OCCT program doesn't read my 12V line accurately, so I dumped out the reading.

BTW, the ram is mixed. Two Gskill 6400HZ sticks on one channel, two Crucial 6400 Ballistics on the other channel. Playing together so far, so good...

I found the unit was stable at this lower FSB under Prime 95 and OCCT using the 9 CPU multiplier, rather than dropping the multiplier to 8 and running a straight 1:1 ratio at 400FSB with the ram.

I'll post a 3DMark score shortly...

It's really GREAT being a newbie all over again!!!

I'd love to hear your comments, questions, and suggestions!

__________________
It's not so much getting your way that matters or not - what matters is how you go about getting it.

swimtech

...and the 3DMark 06 score:



The fan is variable on the card now, gets loud when it needs to then backs down. The mod works well...

__________________
It's not so much getting your way that matters or not - what matters is how you go about getting it.

PangingJr

i'm just wondering, have you tried setting 400x8 and with memory multi 2.50Ax ?

try with all the voltage setting to Normal, except Vdimm and Vcore settings, and try 4x1GB, Vdimm 2.1V, 5-5-5-15-AUTO..... DDR2-1000MHz,
if you can stable the system with the Vdimm and the memory timings and memory multiplier, then try 5-5-5-15-4-4-(4-8)-(42-54)-4-AUTO-AUTO (and then try Vdimm 2.2V, 4-4-4-12-AUTO.....)

swimtech

Your suggested settings were basically what I was running during the week - but no stress testing. I was at 400x8, memory set at 2.0 A or D (I forget the letter it was now...) to get the ram running at 800Mhz (Bios). I had the CPU voltage manually set at 1.325 and ram voltage manually set at 2.1 (+0.3) with 4-4-4-12 latencies, the rest of the latencies set to Auto ('cause the module pairs specs are just a tiny bit different on those other latencies). The voltage compensation setting was enabled then too, as it is now.

I'm not in front of the unit now BTW...

At those settings I was running the same temps at idle and could pass the 3DMark 06 demo/bench with a score of 12706 (video card was clocked at 830 then...). However, OCCT would reboot or blue screen after less than two minutes. Bumping the voltage up bit by bit, it was finally stable with the CPU voltage set where it is now at 1.375 (Bios).

Then I thought it would be easier on the board to raise multiplier back to 9 and run at a reduced FSB of 356 for the same CPU clock, and set the ram to get as close to 400Mhz as possible. Now I'm thinking of backing down the CPU voltage to see if it'll pass stress testing.

I think I see what you're getting at - should increase the bandwidth and lower working latency overall.

I'll try those suggestions even though I don't think it'll be stable, based upon what I was seeing last week - but I'm definitely interested in seeing if the differing ram modules will "play nice" together at 500Mhz. I did get those D9 single sided DDR2 6400 Ballistix modules that a bunch of folks have had trouble with, but they were sooooo cheap ($31.00 US for the kit after rebate) and a lifetime warranty. Plus, this EX38-DS4 board is supposed to have really good power regulation, and I'm thinking those with trouble are pushing them hard in lesser boards. I guess we'll see together!

__________________
It's not so much getting your way that matters or not - what matters is how you go about getting it.

PangingJr

Vdimm and rest of voltages settings in my EX38-DS4 is quite different from my previous board, which is P35-DS4 rev.1.0,
the ones in P35-DS4 fools me and makes it like a little to no vdrop/vdroop, and the Vdimm overvolt by +.2 - +.4, but it all worked out easier then the voltage setting in EX38-DS4.
i’ll try to elaborate on this a bit further, but it has to be later.

PangingJr

shouldn't be a problem to overclock and run your 4x1GB at speed of DDR2-1000MHz CL5 2.0-2.1V on the EX38-DS4,
the memory multi 2.5x on the board should also be okay for the Q6600 and rest of 1066MHz CPUs.

however, since the board provides a Vdimm overvolt by only +0.01V (or even less than that) and the volt drops by as much as -0.04V,
so the Vdimm may works just fine in a Memtesting, but it's not enough volt in Windows and while running your system stress testing.
anyway, i suggest you check your Vdimm voltage reading on your EX38-DS4 when it's under a system's stress testing.

also, after you set the BIOS to 400x8, memory multi 2.5Ax, 5-5-5-15-AUTO, Vdimm and Vcore, i suggest you boot direct to the Memtest first,
and if you get a RAM error you can try the following 4x1GB configuration, or vice versa...

yellow Dimm slot - G.Skill HZ
red Dimm slot - G.Skill HZ

yellow Dimm slot - Crucial Ballistic
red Dimm slot - Crucial Ballistic

----------------------

another thing i find in the EX38-DS4, it's about the Performance Level (which's the Static tRead Phase Adjustment in BIOS setup)
in some certain memory multi, speed, MCH strap the automatically-set PL are either too loose or too tight, and it can cause system stability problems (i've not seen this BIOS setting problem in Gigabyte P35 board for sometime now),
and therefore, you may need to set the PL in BIOS directly if you found it's too tight, and may need to try several value if you found that the PL needs to be tighten down.

however, i find it's best to set the value in BIOS, this due to the system won't boot if you set it too tight, and the board will try to boot itself up with the default CPU's speed and memory setting, and then you can try using another setting again,
but if you try to do this within Windows, then there will be a system lockup and could damage Windows installation.

swimtech

Cool...

You would like to see me mix the memory modules within each channel? That sounds like a hoot! I'll try it! Can't hurt...

Booting to Memtest for overclock/timings check - nice suggestion and thanks - stresses some but not a ton, easy too.

I dumped the EasyTune windows o'c software within days after installing it - decent monitoring but, like you said, it freezes the machine if you don't hit on working settings.

Really nice you and [hobo]eclipse decided to get the same board - I'm really enjoying the company.

I won't be able to get back to the machine until Wednesday evening (maybe)...

About the Static t(Read) adjustment - funny you should mention that. I was hunting for it this past weekend - wanted to fiddle with it 'cause I remembered reading the article at Anand's about how important it was, but couldn't find it to save myself. So, since you spelled it out so well, I just copied-pasted-googled and voila! I found the article!

__________________
It's not so much getting your way that matters or not - what matters is how you go about getting it.

PangingJr

i set tRD value in BIOS since the beginning of using my P35-DS4,
Anandtech likes to play with tRD settings as well, right now, they are trying on a new Asus P45 based, which's still having some of (early) BIOS problems.

anyway, like i mentioned earlier, Gigabyte X38 based do have some problem with the automatically-set tRD value, in some certain overclocks you can't leave it to AUTO, otherwise, you get system stability problem,
if you try to add other voltages like MCH, FSB to fix the problem, then things will get even worse.

PangingJr

here are some of examples... when i left the tRD ( Performance Level ) setting in BIOS to AUTO....

on this (below) example, the automatically-set tRD value ( Performance Level ) was just too aggressive....
tried to add some volts, and try everything, whatever, but system still couldn't be prime stable for very long time,
until i set the value in BIOS to 7 and on the next one i set 8.






and this one was just too loose, at this specific overclock, i can get the tRD value in BIOS down to 5, and still be able to boot to Windows, but i set 6-7.




however, it depends on your overclock, you may not have problem leaving the tRD value to AUTO.

swimtech

At this point, I haven't tried to tweak the Performance Level setting, but applied your suggestions this past week to obtain another stable overclock. The memory latency benchmark in Everest dropped from 81.Xns down to 70.4ns compared to the settings in the previous screenshot.

The following shows stable (OCCT mix 1Hr.) at 1.375v core (bios) 2.1V mem - needed the same voltages as tested before, memory latencies 5-5-5-18-AUTO Running under load...


Shot at 2008-05-23

...and note the DPC latency figures as Prime95 is exited...


Shot at 2008-05-23

__________________
It's not so much getting your way that matters or not - what matters is how you go about getting it.

PangingJr

the temperature looks good.
may be just some adjustments on a bit of things if needed ( like try 5-4-4-15-AUTO with 2.15V, but don't use it if you have to use too much volts on the DIMMs ) and you can just use the setting permanently)
anyway, you may want to save the CMOS setting to BIOS (at CMOS setup utility Main Menu press F11 key) so you can load the setting when you want to use or check it again later.

and btw, Could you check what is the default Memory Frequency/Speed when your CPU is running at its default clocks?
(DDR667 or DDR800? this depends on RAM modules's SPD data and the memory support of the motherboard.)

please follow this steps...



at the MB Intelligent Tweaker(M.I.T.) page, set the "CPU Host Clock Control" to Disabled,
System Memory Multiplier to Auto,
DRAM Timing Selectable to Manual,
set the Standard Timing to 5-5-5-15 and set all the rest of settings below the Advanced Timing section to AUTO,
and then the System Voltage Control to Manual,
DDR2 OverVoltage Control, PCI-E OverVoltage Control, FSB OverVoltage Control, (G)MCH OverVoltage Control and CPU Voltage Control, all to [Normal],
then Save and Exit the CMOS setup... and reboot.

and then enter the CMOS setup right away, check and tell me what is the system default Memory Frequency/Speed?
and also have a look at the line "Static tRead Phase Adjust", on the (SPD) left column, what is the value shown in there?

PangingJr

wow, i think, this already beats my P35-DS4 already.

EX38-DS4 overclocks my E6750 from 333 to 500 FSB with the "Normal" and stock VID voltage on its first boots....

on its second boots.... 533 FSB with 1.325V vcore, MCH/FSB at default volts.

PangingJr

this was the 3rd and 4th reboots after the E6750 installed.

tried using core voltage a bit way below the stock CPU VID (which's 1.275V), set 1.23xV in BIOS and set memory timing and PL in BIOS...

no stress-test in Orthos or Prime yet.

swimtech

OK - got these figures yesterday and wrote them down - and how the heck did you get the screen/print of the BIOS? Anyway, by your instructions:

Memory Frequency mhz (all settings on Auto): 800 800

Static tRead Phase Adjust: 1

That's quite an impressive overclock on such low volts (and goods temps too) for the good 'ol 6750 on this board! I'm betting the extra CPU voltage regulation is helping...

__________________
It's not so much getting your way that matters or not - what matters is how you go about getting it.

PangingJr

this E6750 is pretty easy to overclock, it's one of week 20 with 1.275V VID, it has low FSB wall (540MHz) but this CPU uses quite a low core voltage,
undervolting this CPU on a low FSB overclock is what i've been using it for, and for sometime now.
i had another one but gave it to friend months ago i think i'll get it back and give him an E8200 instead.

this (below) overclock the core voltage was just a bit too low for 500x7,
it can only run a halfway of 2x SuperPi 32, lol, but it can be used for a day or so if i want to use it.

(1.20V BIOS 1.184V before full load)


to pass a strss-test in Orthos and 2x SuperPi 32 runs, this E6750 needs 1.2375V in BIOS (1.216V in Windows and around 1.20V in full load for this EX38-DS4 board), also, PCIe/FSB and MCH = +0.20V/+0.10V and +0.175V for 2x1GB RAM installed (for 4x1GB MCH may need to be upped to 1.50V or +0.025)

and about how the heck did i get the screen/print of the BIOS.... it's taken from the board's user manual (PDF file).

PangingJr

i plan to have a Q6700 on the EX38-DS4 later on as well.

if 3.0GHz is good enough for your daily use, try 333x9 with all the voltage settings to Normal in the BIOS.
and try memory multi 3.0x (DDR2-1000MHz), it worked quite well with the 333 FSB, and it may also work well for you.

5-5-5-15-11-5-10-45-5-4(tRD)-Auto(tRD Phase Adjustment) 2.05V
4-4-4-12-11-5-10-45-5-4-Auto 2.15V
4-4-4-10-8-3-7-30-3-4-Auto 2.2V



i used the boxed Intel cooler and it was in a warm room when running the prime95, so just disregard the CPU's temps.

swimtech

Had a little time to play last evening...

Got it to boot into Windows at 3.6Ghz.
FSB 400Mhz x 9 multi
Memory 1:1, 4-4-4-12 timings, all others auto.
Chipset voltage bumped up 0.1v (running 4x1G in all slots)
CPU volts at 1.525 Bios, 1.49 - CPUz. (CPU VID 1.325, L750xxx batch)
Temps were a bit high, 35C idle.
Wouldn't run OCCT mix though.
This might be a tad tough to get stable - I'll relax the timings on the mem. I can still move the CPU voltage up, but not much...

Also want to see what a stable 3.4 looks like - noted Al Vampyre got his there...

I know I can do 3.2Ghz running cool, memory fast, and stress stable for sure though. That was my expectation when I bought these parts - so that's nice.

I'll fiddle more in the next few days, and get some screens.

Did you get to try that 6700 PangingJr?

Glad now I didn't go for that Q9300, it will need a FSB monster of a mainboard to ramp up.

__________________
It's not so much getting your way that matters or not - what matters is how you go about getting it.

PangingJr

i had my friend looking for a good Q6700 but he offered me his Q9300 for free, so i have it installed on EX38-DS4, no overclock on the system, and it's now for my wife's office.
i myself also still having a Q9300 on P35-DS4, and i overclock to 400FSB with memory 2.4x and PL5 for my daily use.
anyway, i think i might keep the asus p5q deluxe with me and get a new wolfdale CPU for my overclocking experiments.

swimtech

The P45 boards look to overclock really well - hopefully it can pump that 9300 to 500mhz. That chip running on a 400-425 Mhz FSB (and 7.5 multi) should be about the equal in performance to what I have (6600 @ 400mhz x 8 multi).

Though for someone using Crossfire on 2 PCIEx8 slots, the graphics hit was about 10% on this or previous gen graphics cards (read an article at Tom's about 4-5 months ago...). I would think the percentage difference would be higher with the latest gen cards.

I know that's not really a concern of yours though.

__________________
It's not so much getting your way that matters or not - what matters is how you go about getting it.

swimtech

I had to work a bit for this one...

These screens represent what I think is about the best this system can do without getting into really stressful voltages (Bios set at 1.475) or temperatures (I'm pretty sure the AC Freezer 7 Pro is close to its limit though with these temps and the voltage applied). Got an OCCT stable 3.4Ghz overclock - that stress test is tough...


Shot at 2008-07-02

I'm pretty proud of this one - as I'm sure many of you were the first time you had to work at one and it turned out well. The CPU voltages are obvious, chipset voltage is elevated a tenth of a volt, and memory timings are loose for this (5-5-5-18), but I can probably fine tune those some more. I can also tweak the multi and the FSB some too, but it won't change real world performance much regardless.

Overall though, I think this pretty decent for a 1.325 VID CPU and mixed ram pairs across all four slots!

Also did a 3DMark 06 test. The card is overclocked to 850Mhz core and 1271Mhz memory - same as the previous posted score.


Shot at 2008-07-02

No, really, the processor is NOT at 3.8+ Ghz. The 3DMark program is reading as though the CPU was using a 9 multi - it's really at 8 for this run. Broke 13K - woot for that! - but the 152 point (1.2%) increase for a 200Mhz (6.25%) increase in CPU speed over the 3.2Ghz figure of 12936 should tell something - gaming really is about the video card.

Edit: It was late last night - 1.325 VID Q6600 processor, not 3.125 as originally posted.

__________________
It's not so much getting your way that matters or not - what matters is how you go about getting it.

swimtech

Wanted to sum up to this point - just finished the DH Photoshop benchmark, available on This page

I went back to reviews done HERE at DH on the QX9770 and QX9650 processors. These are still available for $1470.00 and $1019.00, and are newer 45nm technology quad core processors. The QX9650 costs more than I paid to build my entire case of parts, even if I had bought the video card (again, thanks to DH and ATI - I'm lovin' it!).

This is the reason I'm overclocking, and why I planned the component purchases ahead of time so carefully...


Shot at 2008-07-11

If you compare the benchmark times for the processors in the review versus the time I got for my overclocked Q6600, you'll see it beats even the QX9770 as configured in the test system (which is no slouch in itself). Remember, the score I got is for a stable, stress tested overclock - without pushing any components to their absolute limits - suitable for playing games, editing video, anything really. It's been all about value for me...

If I can do it, you could too.

__________________
It's not so much getting your way that matters or not - what matters is how you go about getting it.

swimtech

Went back to 3.2 to tweak out a 24/7 setting. I wasn't comfortable running at 3.4 with the voltages and power usage I was getting with those settings, so this is what I've settled on after working some this morning...


Shot at 2008-07-13

Note that the power usage in Everest shows the full load characteristic of the CPU now well below the full rated TDP of 130 watts (it had gone as high as 128 watts in the 3.4Ghz overclock) and that the load CPU voltage is just above my CPU's VID of 1.325 (can we call that the "kick start voltage"?), yet below the rated maximum of 1.350 (from the Intel box the processor came in). The CPU voltage is set at 1.3815 in the bios, and shows at 1.36 at idle in Windows. I did have the Bios set to 1.375 in my previous OCCT stable overclock at this frequency, but I had a blue screen later while playing Civ4 for several hours - bumping the voltage to 1.3815 in the bios fixed that. Full load power usage of the system (less the monitor and speaker system) is 230 watts, 130 at idle. At the 3.4GHZ settings, idle usage was at 140 watts.

The cooling is well within tolerances, maxing out at 56C (RealTemp figures) instead of 63C.

I'm tickled with the report from the DH Photoshop benchmark at this 3.2Ghz frequency too. It did fall from the 3.4Ghz score of 84.4 back to 88.8 - yet is still ahead of the QX9650 (94.6) and close to the QX9770 (85.5) - still got me a $1000+ processor! The percentage decrease in clock speed (5.9%) is greater than the decrease in benchmark time (5%), so it looks like I was able to hold onto a bit more of the performance than I thought I would.

The memory configuration I'm using is holding me back a tad though, I think. I tried to improve the main settings from my previous stable 3.2Ghz overclock to no avail, but I was able to improve the Performance Setting from 9 down to 7 (thanks for that suggestion PangingJr!) - made a nice difference in the memory subsytem performance - and is probably the reason I held onto some of the DH Photoshop benchmark performance. The board is doing a good job of keeping the two disparate sets of memory talking in unison though. Everest latency measurement is at 63.0 or so (it can vary a bit between runs), Read is at 8197MB/s, Write is at 7234MB/s, and Copy is at 7311Mb/s.

Hopefully this discussion and pictures so far can help others find their "sweet spot" too.

__________________
It's not so much getting your way that matters or not - what matters is how you go about getting it.

swimtech

No screenies this time, but I did get a "usable" 3.6Ghz overclock last evening.

Bios CPU voltage set to (gulp) 1.525 volts, load CPU volts came down to 1.46 or so. Memory running at 5-5-5-18, Performance Level 7, 1066Mhz (3:4 ratio), Mem bios voltage at 2.2 volts, MCH +0.1 volts, all other settings auto.

I was able to get a nice score on the DH Photoshop Bench HERE.

Also tried to get it stress stable by upping the voltage two more steps, but didn't have the guts to push more CPU juice to it. Core 3 would fail Prime95 (64bit) after about two minutes or so. I may revisit this, but I'll have to look at changing other voltages/skew/PLL before I contemplate going higher than 1.55 volts - that's getting excessive for me - and maybe rigging a little fan on the Northbridge. Surprisingly, my CPU temps on that Prime95 test didn't go above 57C (RealTemp 2.6) in a 23C ambient room - might have a bit more cooling headroom than I thought - but I've noticed Prime95 doesn't seem to stress the CPU/Mem as hard right off the bat as OCCT seems to.

__________________
It's not so much getting your way that matters or not - what matters is how you go about getting it.

PangingJr

right now i don't seem to need CPU PLL voltage to help on a high FSB overclock, be careful when using PLL voltage setting, i've seen so many times that overclockers have mentioned about using high PLL voltage can kill a processor faster than using high CPU core voltage, however, i believe every overclocking can get different results depending upon so many factors.

here's my 475x7.5 CMOS setup...



i cannot touch CPU/NB Clock Skew settings at all, the overclock either hang at BIOS or couldn't get into Windows,
however, i sometime able to play with DRAM Clock Skew settings.

CPU GTL voltage ref's settings is just a must for this particular Q9300 and this motherbaord.
for 475 FSB the settings must be .0670 on both, otherwise, no prime95 stable.
couldn't leave it at AUTO, couldn't use any values below or above the .0670.