Realizing my childhood dream [LGA775 / Nvidia SLI]

[lexluthermiester]

You can call me insane, but fiddling around with all this really got me excited again. And I can´t help but fall in love with this 10 year old platform.

None of us think you're insane. You're enjoying a hobby! Have fun with it!

Bought two more 780i boards (call me insane if you want but I made a decent deal as the one in the picture is a bit dirty and beaten up but still working and the second one coming in will be including a mystery CPU ) and 2x X5470s.

Again, not insane. Those are nice boards and OC champs. Premium stuff. Again have fun with them!

 

[Dinnercore]

I´m on the X5470 from here on out:

It was an easy swap, now that the board accepts 771 CPUs. I´m thanking my past self who bought 6 mod stickers at once.

Well, the physical swap was simple. I went straight into bios and set my voltages manual again, trying to remember where I had them before. These are the voltages I set in bios:
1.343 Vcore / 1.2 FSB aka VTT / 1.9 (!) RAM / 1.3 SPP / 1.5 MCP / 1.2 HT
GTLs +40mV on all.

It booted and I got into windows but it was unstable. Prime95 failed after a while (15 min.) in blend on core 1. I took the time to read into GTL adjustments and how to calculate the estimates from where I can start to test for my individual values.
I figured I was pretty close to the calculation but read that lane 2 and 4 are address lanes that can work with a little less voltage which can increase stability. Since I got a fail on core 1 which is the first die, I raised the first GTLs by one step (+5mV).
Which in total lead me to +45mV / +40mV / +40mV / +35mV.

Small FFTs run stable, or atleast get further then before. At the current state I do not make multiple hour long stability tests, since I need to test so many little steps.
If you read through the thread you already now that something is always up, and it just so happened that on booting the next day I got a BSOD crying that my bios would not support ACPI...
Eh? Thanks, whatever. I rebooted and it started normal. Everything working, windows file integrity test passed.
I decided to run prime95 again and guess what, I got instant errors this time in blend test. I started to question my sanity and re-ran the small FFT test -> that one is still solid, even for 30 minutes. Blend test is using larger sizes to check RAM too, I showed the sticks my angry face through the window and checked my screenshots.

Meanwhile I also installed hwinfo64, thanks to the suggestion of @Edwired . It does report some very low core temps that seem to good to be true, but my SPP temp is showing up there! It sits at 45°C in idle and neither gaming nor artificial load could bother it to go over 50°C. This now leaves the other unlabled mainboard temp sensor to be either VRM or MCP.

Upon investigating my hwmonitor screens I noticed the voltage 'VIN1' changed from 1.488V to 1.528V. I must have set something different then with the X5482 and it hit me that I set RAM higher then before! After adjusting it back down to 1.85V in bios I re-ran prime95 once more and got it now stable in 30 minute small FFTs and blend tests. Also no more BSODs or other fun stuff.

The voltage balance is very critical on this platform and I like it. Much more exciting then todays 'just raise Vcore'. And I learned a lot already.

Now back to a game, I ran some Far Cry 2 benchmarks with the old X5482 at 3.3GHz, close to what the X5470 is doing atm.



I was testing no AA vs 4x AA at 1080p. As you can see those results are within margin of error. The 4x result is fluctuating more, lower min. FPS but also higher max. FPS. I saw the GPU usage scale rather well with the higher AA setting but still only hovering around 40-50% on 4x, the CPU was at 85-95% at all times.
Judging by this I´m either restricted by CPU speed or hitting a bandwith limit somewhere from the SLI-setup. Reviews and benchmarks from the time show a single GTX 295 reaching higher average fps (80) at 1920x1200 with AA. Quad-SLI setups got even higher, but most of them used overclocked i7s at or above 3.6 GHz with the newer platform that had just been released back then.
A graph and an article from tom´s hardware compared the performance of GPUs paired with different CPUs and even at high resolutions a Q9550 @ 3.7GHz was necessary to get a single GTX295 to its full potential. At 2560x1600 this setup even outperformed the i7 920.

Now I hope that my bottleneck is actually the CPU and not with the chipset. The setup is stable now but before I try any OC I want to lower Vcore and try to do the droop mod on that resistor. I don´t trust the pencil method, seems like a quick and dirty trick. Got a 150k Ohm resistor ready to try if that changes anything. At low Vcore first, in case I can overshoot this. But the source stated 100K Ohms, so 150K first should be on the safe side. For a moment I thought about using a potentiometer, but this is a dangerous game knowing myself.

I remember these boards, they did royally suck from my experience.. That north bridge temp is low compared to what mine was, about 60C if I recall, will need to do some digging for some pics etc Not quite the same with the GTX 295, but had dual 8800 GT's which weren't bad at all. Having gone nuts and bought a brand new QX9650 at the time, overclocking on that board was a nightmare.. Memory holes and temps on it were awful.
I'll see if I can grab some pictures of the setup, I believe it was when I had just started water cooling and a bit of benching as well... Was fun but that board was damn frustrating lol

I have nothing to compare them against, this is my first time on 775. I did notice that they require patience and careful balancing but seem to behave well once you hit the right numbers. I changed the TIM between all chips and coolers on the board, this might have helped with the temps. On top I use the 'optional' fan on the heatsink and I have no other heatsources thanks to watercooling.
I would love to see pic of that, I had to settle with a 8600GTS back then. My first PC :,)

Again, not insane. Those are nice boards and OC champs. Premium stuff. Again have fun with them!

I love the boards, and they might not compete with sth. like a maximus formula in terms of FSB clocks but so far mine takes the high Vcore / power draw very well. Once it is balanced out it feels solid. Bios has all I need and is replaceable if something happens.

 

[phill]

I've had a few 775 boards and with the memory holes and the overclocking issues I remember they where a pain!! Here's a few pics of one of my first water cooled rigs



Ah the days where I didn't really know too much about stuff but would do something daft anyways... Funny, things not really changed much!!



I loved that screen.....

 

[Edwired]

That a lot of information as im trying to soak it in. Yeah that one i posted up on xeon owner club is stable for cpu only not gaming. Im at the moment testing the overrall system stability as i use project cars game to make a load on cpu, ram, northbridge and graphic card that way i can tweak one setting at a time . The trick i discovered for gtl settings is to do vtt x gtl = math that way you know if you are getting the correct gtl voltage.

Also another thing for bluescreen you need to check it with bluescreen viewer that way you can get the error code and look online what that cause by.

Heres the link for information on 775 error codes
https://www.overclock.net/forum/297...ing-must-have-info.html#/topics/940091?page=1

 

[Dinnercore]

@Edwired Thank you, I know about the bluescreen viewer and how to use it. It´s just that im confident that any bluescreeens immediatly after my hardware and bios setting changes are likely a result of unstable / wrong settings. And if my memory bus is unstable and corrupts my VGA driver while booting then bluescreen viewer would lead me to believe that my drivers are broken while instead I have a hardware level problem.


I now spend many hours with my bios and the methods of how to flash it to a more recent version. I was still on version P06:

Latest is P10 from EVGA, but XFX ones are little more hard to find. They should theoretically work if you cross over BUT I got an XFX board so I stick with XFX bios. The latest version from XFX was P09, the only difference in P10 is increased CPU compatibilty. Since I have no issues so far I decided to leave that out.

What got me interested in updating the bios were changelog notes like improved WHQL compatibilty, improved GPU compatibility, corrected SLI memory calculation, improved RAM compatibility when using 4x 2GB.
After some digging online I found the XFX-Bios archive and had to piece together how to actually get the flasher to work from USB. I have no DVD-drive nor floppy. I got the .iso and had the intuitive reaction to extract the files from it, made a bootable freeDOS stick with rufus, copied files over and booted from that.

First time I did it like that too, my AMD X370 board had an ez-flash utility so I could skip the freeDOS part and was a little unsure if I would remember how to navigate in DOS. Well my dad had an old PC he let me play on when I was 5 years old and that was using DOS, so ofc it was no trouble. I don´t need to look things up, I´m a big boy!

Thanks to XFX a handy autoexec was already set up and I got straight to this little horror movie:

Now for the sake of my own sanity, you do see the p09.bin at the end of the file name to program?!
After this screen came one erasing step, then one programming step, another erase and another program. It played a sweet tune in some high key major on my speaker and I somehow got the image in my head of a dog proudly wagging its tail because it knows it is a good boy.
It wanted to reboot and halted on the post screen like after a bios reset. I decided to be on the safe side and powered off, switched PSU off, took battery out and cleared CMOS for a good 10 minutes. The whole process included another 20 minutes of me having to remove the PSU and turn it upside down because in order to get to the battery I had to take out the lower GPU, which then spit one of its slot screws right into the PSU-fan...

Back up I went into bios, set all settings the way they had been and got into windows. I started CPU-Z, excited to see my new P09 bios version but:

Eh? P06 still?... msconfig32 told me the exact same thing. BUT I compared the date and saw that this version is now a whole year more recent then the version before. What. I was confused and disappointed and frustrated after what seems like another hour wasted.
I googled the exact issue and found a user in EVGA forums to have the same trouble, he too had the P06 before and even after 4 tries he got the same result. Someone there told him to check the POST screen too. He did not answer back so this is what I will be doing now.

I will edit a picture of it in this post later.
EDIT:

So this is saying P09. I would like to trust my bios knowing what version it is more than any software.

On another note, I tried my 150K resistor:

I was a bit hesitant to use a carbon film resistor instead of an MF due to noise they can produce but soon threw these worries over board. This is more a thing to consider in amplified circuits for hi-fi audio. And the pencil method is nothing other than applying a carbon film...

I carefully placed it on top of the marked SMD while the PC was running and saw these changes in voltage:

Seems like a safe thing to do so I will now solder it and test under load of the droop is less or if I just fixed the difference between requested and delivered voltage in idle.

But this whole bios thing is really bugging me. Old version number from flashing the new version.bin file and getting a different date... Did I create a mutant bios, a mix between the two?

 

[Edwired]

Jeez erm do google about the bios update update as sometimes the bios said it erasing the old bios in order to write new codes but it could be the way the bios flasher works sometime you have to add switches in the code to force flash the new bios

Here the link for the bios update im sure it can be fixed
https://www.overclock.net/forum/167...ios-p09-no-download-link.html#/topics/1457704

 

[Dinnercore]

I checked the fan settings in my bios and tried out if the thing mentioned in changelog (fan speed fixed after resuming from sleep state) for version P09 is working. Yes it is, and my post screen not only mentions the different version number but has a slightly changed font size too. Going on from this I will not bother with reflashing a dozen times and assume that I´m on P09.
Board is working, CPU-Z bench consistently gained 5 points single thread performance after the flash. Seems good to me.

Next I fought with the Vdroop mod. Even after mastering my soldering skills on a practice pcb I was not confident enough that I could solder these tiny contact points, not with other components so close by. Since there is not much difference between a carbon film in a case or just open on top of the smd I just went for the pencil method.

I lowered my voltage in bios to 1.325V and now get these numbers under load:

And idle:

Solid result. Before that I had to set 1.343V in bios and got a drop down to 1.256V under load. The rise in temperature after this mod confirms it is working as intended. Now I can set a much lower voltage resulting in happy idle temps.

To help temperatures further I then went ahead and rigged another fan (60mm) to blow some air into the fins of the VRM heatsink:

Since the only fan for the heatsink was on top of the NB I would guess that the VRM is the hottest part, only a passive heatsink that has to channel that heat all the way down to the fan with just a single heatpipe. Now it got some more airflow and the vague 'mainboard' temp dropped by 2°C during small FFTs with the case closed, as with the screenshots before the case was open!

At last I feel prepared for higher clocks.

 

[Edwired]

Cool i say pretty cool as mine is air cooled i did get to boot with gskill ddr2 to run at 1200mhz while stock was 1066mhz for a few minutes before crashing in intel burn test so i know i have the ram capable of running near 1200mhz

 

[Dinnercore]

The day went well, I did tiny steps up and ran 2x 15 min. tests in prime95 each time. First one blend then small FFT torture.
This also was the first time that I OC a CPU and continuously lower the Vcore. I noticed an increase in MCP (or whatever TMPIN0 is) temperature by ~1-2 °C for every 50 MHz I went up and decreased Vcore in an attempt to combat that additional heat.

I had several points where one or two cores threw errors in prime and each time I just had to bump the GTL values up by +5mV to get it stable again. I´m now on 1.29V CPU in bios (1.272 in windows). Still on the same 1.2V FSB / VTT, 1.85V memory etc.
GTLs are now +60mV / +50mV / +60mV / +50mV.

I´m running out of time for today, but it will only go up from here. Might have to do another little mod and replace the thin and brittle mobo fan with a proper 60mm, tape over the gaps on the side between fan-frame and heatsink to channel the airflow. I would like to avoid temps over 80°C, especially when during summer my ambient will be 5-7°C higher. If this still does not help and this mainboard temp is the limiting factor for my overclock, then I will have no choice but to take the system apart again and fix that heatsink by either using liquid metal or installing a chipset watercooler. Which I would love to avoid since it will be a pain to undo all the things I did already...

 

[Edwired]

At this moment im trying to test resident evil 2 remake by steam i can tell you it runs on my system no problem the only problem i am having is the graphic card crashing in pcie x16 slot so i would say i have to drop the card to pcie x8 slot tonight as i have done that before i was able to play for hours on end. As well i think the pcie x16 slot is faulty to a degree as i noticed one of the pins is nearly missing as well it very loose when i install a graphic card. I do have another motherboard laying about that i could desolder the slot off that and replace the one on the asus p5q premium but that will be another day "facepalmed"

 

[Dinnercore]

Quick Update, I bought another 60mm fan from Noctua and taped it on the board heatsink, replacing the stock fan. I had to use this white tape for now, it is the only sticky thing I have that is rated for temperatures above 40°C since I´m out of double sided 3M stuff. Might fix this later, but it does not bother me too much.

It did help with the heat a little bit, looking at hwmonitor I can see a consistent 3°C drop from idle to load. Now it sometimes even dips below 40°C in idle! And I got 74°C max. @ 3.71GHz. That is better than the 77°C I had at 3.57GHz.

In order to get this stable I had some more adjustments to make, again a slight bump in GTLs, but this time it was not enough (65/55/65/55 now). It still failed during FFT lengths above 700k and I had to lower the memory voltage even further.
Now I´m at 1.825V for memory instead of the 1.85 before. I don´t know if that hints at the ram modules becoming unstable or adding signal noise when running at a voltage above their rating or if this is something the board is struggling with. In which case I would be in trouble and limited to only 1.8V RAM, everything above throws instant errors even at stock speed. As seen with 1.9V, even the 3.3 GHz would fail. To test this I have some 2.1V rated sticks incoming, lets hope for the best.
If it fails to be stable with 2.1V memory I can still try to either increase FSB voltage one step or try to back off the GTL values a little.

However since I´m now stable at 3.71GHz I need go higher.

 

[phill]

You'll never stop chasing the bigger numbers

But congrats and nice work Brings back memories

 

[Edwired]

At the moment im getting more game stable. The crashing is related to ram on the motherboard. All the timings are set manually as auto will crash all games with or without an overclock on the cpu, ram, gpu. The cpu is 4.05ghz, ram 1123mhz, fsb 448, first timings 6-6-6-15-5-50-7-4 second timing 8-6-7-6-7-7-8 third timing 17-5-1-5-5

 

[agent_x007]

Try setting 896MHz on Memory with CL5.5.5.15 timings + Command Rate 2T.
1,9V on DRAM Voltage isn't enough for 1066MHz.

 

[Edwired]

The vtt is already at 1.3v nb is at 1.40v the nb gtl is at 0.615. Troubles starts if any of the ram timings is auto. As it already loose from the start but unstable. So i had to tweak each settings while testing games that are ram related that way i tweaking it in the correct direction at the moment im testing it in project cars and resident evil 2 remake one shot demo.

As the common errors im getting appication error 1000 and display 4101 both are caused if the timings in auto plain and simple. Not driver related as commonly seen on the internet

 

[Dinnercore]

I think my voltages are exactly where they should be now, with the same settings as before I went straight to 4+GHz with no issues. Still got some headroom for FSB, maybe even Vcore. The 77°C on MCP is already high but only during the prime torture tests. In typical usage scenarios like gaming I´ve not hit above 60°C yet.
BUT before I try to go EVEN FURTHER BEYOOOND....AHHHHHHHH (sry) I will wait for my RAM to arrive. I´m at a point where I run Bus Speed 1 : 1 DRAM frequency and any increase in FSB will push the memory too. And I´ve already seen that the sticks I have now behave a bit picky.

Oh and I added something else to the build, I know many have mixed feelings about case badges. I personally like them but I did not want to disturb the clean look of the case so I put them behind the front door that hides the drive bays.

Yes I know I have a Xeon and no Core 2 X, but they are kinda the same. Here is my current CPU-Z validation: https://valid.x86.fr/plcjtl
If I can convince the memory to behave I might go hunting for the single thread performance of these entry level Ryzens...

Don´t worry tho, my CPU-OC adventures will come to an end eventually, then I will get working on the GPUs.

 

[Ruru]

That sticker is hella lot cooler than a Xeon sticker.

 

[Dinnercore]

My hunt for DDR2 ram brought up some stuff, I now have 4 x 2GB Geil Evo rated 800 CL-4-4-4 at 2.0V. But I also found something that really caught my attention.

There are not many 1200 - PC 9600 kits to find today. I could only spot 3 offers atm on multiple platforms, two had been these guys:

2x1GB OCZ stuff for 1200 5-5-5 @ 2.1V. A little low capacity for my Windows 7 install but 4GB are in theory still enough for the games from 2008+. But there were these bad boys too:

A 2x2GB kit from G.Skill. With the date on them being late 2010 they don´t perfectly align with my 2008/9 components but it´s close enough for me. It is still DDR2 and if they work at 1200 with high voltage I might profit from them even if my board fails to be stable at higher memory voltages.
If the stars align I might find another 2x2GB kit like these some day, I will be patient. After all I managed to get the waterblocks for the 295s too.

 

[Edwired]

I was able to boot at 1200mhz on 4 dimms with g-skill f2-8500cl5-2gbpk

 

[Dinnercore]

Sticks are in now, replacing my current OCZ ones. Before that I took each module by the hand, we made a big circle and had a brief discussion about which ones would like to work together.
The 2 x 2GB G.Skill run side by side with one of the 2 x 2GB Geil kits. It was plug and play, I immediately went to bios and put them on 2.0V.

(OH no they are burning! JK, red LEDs. Measured 29°C on the heatsink after a test run.)

Currently running some tests, no errors or crashes yet. Primary timings I choose a bit loose: 5-5-5-15 2T
Later I can decide if I want to push for CL4-4-4 or higher frequency.

I was able to boot at 1200mhz on 4 dimms with g-skill f2-8500cl5-2gbpk

Nice, did you get them stable at 1200? What timings and voltages do you use?

 

[Edwired]

Just for a brief moment i got into windows then it crashed in prime95 i was somewhere between 5-5-5-15 to 6-6-6-18 with voltage between 2.10v to 2.20v that by the custom pcb reading with dmm. But that when i lost track what i was doing. I will try again another day just too much going on lately. If i had prime95 going with valley heaven benchmark the graphic card keeps crashing but the cpu and ram kept going like nothing happened. The way i see it it somewhere between the nb gtl, ram clock skew, cpu clock skew and northbridge clock skew. It like playing whack a mole all over again

 

[phill]

Always overclock or tweak one thing at once, you'll just cause yourself more headaches with trying to do CPU, RAM and GPU together, stick with a solid CPU speed, then try the RAM if that's what your aiming for.
Get to your desired CPU speed, then tweak the ram, that's what I do Years of benching on HWbot you learn how to make things easy for yourself

 

[Dinnercore]

Just for a brief moment i got into windows then it crashed in prime95 i was somewhere between 5-5-5-15 to 6-6-6-18 with voltage between 2.10v to 2.20v that by the custom pcb reading with dmm. But that when i lost track what i was doing. I will try again another day just too much going on lately. If i had prime95 going with valley heaven benchmark the graphic card keeps crashing but the cpu and ram kept going like nothing happened. The way i see it it somewhere between the nb gtl, ram clock skew, cpu clock skew and northbridge clock skew. It like playing whack a mole all over again

Ahh yeah that happened to me too, I tend to do multiple things at the same time and then I don´t know which of these things cause issues. I benefit from my board as it does not have all these settings you have. I can´t adjust clock skew for example.
I hope you get those 1200 stable at some point, I would say stick to what @phill is saying.

Back to my project:

I start to doubt the use of prime95 for testing my system. Don´t get me wrong, it is really nice for finding issues before they show in a BSOD or failing to boot. BUT the temps it produces in torture tests are so far off from real world usage, that I´m beginning to think I´ll just blow up my VRM with it when it actually would be fine for gaming or even rendering.

Here are temps after an hour playing Sacred 2 and running 3x Far Cry 2 benchmark afterwards:

And during Cinebench:

In gaming the MCP (still not sure if that temp is actually MCP or VRM) is over 20°C lower, cores are ~7°C lower but they are not critical in p95 so I will not focus on them.
To get to the numbers I see during torture test I would need an ambient temp of 42°C+, I don´t exceed 30°C during hot summer days. I might see 75°C on that sensor during summer when I run 24/7 render workloads.

I say that because prime95 feels uncomfortably and unnecessary hot on that part of the board and going further from here I might see 68-70°C during render workloads, but in prime95 hit 90°C+. If this is the MCP-temp then 90°C would be the limit it can take.

 

[phill]

@Dinnercore - What sort of fans do you have in the system that hits 28000RPM??! Does the PC float at all??......

Also just a question really, with the temps you are getting with VRMs etc, are you actively cooling them when testing with fans etc? Stability tests are fine but they do put a lot of stress on the hardware (hence being a torture test) but if you not cooling those areas they might be causing you issues

Just yesterday a mate from work was having issues with a battery charger he had made, wasn't getting the full amount of the charge into the batteries because one of the chips were getting too hot. With a tiny heatsink over it, it was getting more charge which in turned charged the batteries a little higher. Cooling when testing is so important, you can't really ever have enough I think when I used to use the 780i board, I made a very dodgy wooden frame to hold 3 120mm fans, it cooled over the two GPUs and the motherboard a little as any flow of air (big or small) can make a massive difference (Even with hard drives but I digress...) lol

As seen here by my very poor picture lol Temps on that motherboard were horrendous... When I moved over to the P35 board after it, it was a completely different beast...

Ah, back in the days

 

[Edwired]

I know pain in my face as im getting stable for the cpu no problem there it when the graphic card doing the heaven bench at the same time with prime95 it will disappear after a few mins with errors in event viewer appication 1000 and nvidia driver error 4101. Im beginning to think theres a problem with pcie x16 same with pcie 8 both will crash the same way as before it seems i have to use pcie x4 1.1 that will make games slow and stuttering and ya know i cant stand looking at 1910 film reels. Look like i have to try that tomorrow night if it crash one more time im gonna just overvolt everything and overclock everything till it bleed it guts on the floor