|
|
Post by osmiumoc on Jul 11, 2020 19:51:36 GMT -5
So I recently switched to this board and probably will make it one of my main DDR2 boards. I'm still fighting a bit with the transition to cold tho, anyone here who might be able to give me some hint as to what I'm messing up? Third time going subzero, still on DICE atm. I improvised a NB pot which is working decent:   -31°C at the back of the board behind the NB. Board is covered good in liquid rubber, back and front. Insulation worked well, I cut the heatpipe of the NB-heatsink so there was no condensation on the VRM sinks. My issue is that I can not get an FSB beyond 560 stable. It just does not want to do it, I had low multi and made sure there is enough Vcore. I tested the RAM on the same 333 strap that I use for 1:1. I got to 660MHz (https://valid.x86.fr/pn2dgi) on the RAM with the different dividers. I also tried 400 strap but that one does not even post when trying 1:1. So RAM itself is fine, the sticks do high clocks even with the NB and CPU cold. The CPU is at a safe frequency and voltage. I just could not get the FSB above 560. Everytime it fell apart at exactly 562MHz. BSODs like 'Memory Management' and 'IRQL not less ...', hard locks with CE on the debug LED. I have a big issue somewhere in my signal tweaking. What I tried: All voltages up and down. VTT, NB, PLL, IOH. Each one I upped one by one to see if anything changes -> nothing. Same 562MHz wall GTL-Ref tuning. Up and down again, first each one individually and then I also tried them together in different constellations, high and low etc.. -> NO changes, still the exact same 562MHz wall. Clock-skewing: +50ps /+100ps /+150ps And lastly I turned to the only other option, DDR-Ref and termination. I was rather careful here and only tried one step up or down. Nothing changed again and since the RAM by itself does 660 I don't think I have an issue with the signal from IMC to DIMMs. It must be something in the communication between NB and CPU that goes horribly wrong. Since none of the settings in bios had ANY effect on that wall I feel rather clueless as what do or try next. My last straw is that I hit a CPU FSB-wall (Its the same chip that refused to be stable above 560-ish on the P5Q-E too) but I never heard of a Wolfdale having one so low. And VTT did absolutly nothing. I went up to 1.5V VTT and VNB with the same result but I also was able to get to the same 560MHz at 1.3 VTT and only 1.24V NB. I'll try a different CPU next and I really hope this is the culprit since otherwise I seem to really suck at this  |
|
|
|
Post by Macsbeach98 on Jul 11, 2020 22:04:44 GMT -5
A little tip try 1 stick of ram in the outermost slot and see what happens.
|
|
|
|
Post by osmiumoc on Jul 12, 2020 18:30:43 GMT -5
A little tip try 1 stick of ram in the outermost slot and see what happens. Thanks, I gave it a try. In fact I tried many things now and I feel like I'm getting closer to the issue but I just did not find the gremlin yet. So when I try the outermost slot with one stick it manages far worse. Max. stable FSB drops to ~545. I now have three of these boards so I set up another bench with a waterloop to try another EP45-Extreme on ambient side-by-side. I fitted each one with an E8400 CPU (same E0s). Besides that each component is different, another GPU, other sticks of RAM ofc and subzero vs. ambient cooling. One board wrapped and cold prepared, the other one stock. Same bios (Version F8) on both. Fresh OS installs. Even plugged them into different phases of my power circuit. Both boards hit the same 560 FSB wall. Exactly the same behaviour and at precisly the same speed. I focused more on MCH reference tuning and managed to squeeze out another 2 MHz on top, barely making 562MHz SPi stable. I switched RAMs between, nothing. Both boards hate the outermost slots in single channel, both dropped in max. achievable clockspeed. And not just that, the ambient cooled board completly went into a tantrum after I tried that and refused to post from there on out. I reset CMOS and had to try 5 different memory kits, finally had it posting again with a very slow 4GB kit from OCZ... After setting timings and voltages manually again I could switch back to my usual sticks (G-Skill Micron GMH and GKX). Also both boards can do memory frequency fine when not running 1:1 divider. My bios settings: Vcore fixed 1.5V (1.45V actual) Vtt 1.24V up to 1.5V tested -> no changes PLL 1.5V to 1.6V -> usually does nothing anyway CPU Ref1 -> I tried up and down on high and low VTTs (as low as 0.720V on 1.4Vtt and as high as 0.860V on low Vtt) CPU Ref2 -> ^same^ (also tried to swap these for each other e.g. setting ref1 as the higher one and ref2 as the lower one. That had some success on my P5Q-E but not here) MCH Core 1.3V up to 1.46V -> no changes (I had one of these boards doing 600 FSB before with an E8600 strangely enough, that only needed 1.4V but I can't reproduce it with the E8400 nor the E8500. I tried the exact same values as I had on the E8600) MCH Ref -> ! had some limited success with pulling it really low like 0.640V but only gained 2MHz in stability. Going even lower did nothing again. ICH 1.5V ICH Core 1.1V to 1.14V DRAM 2.1V to 2.3V I left DRAM termination on auto. Might give that a shot next. I also thought about it being a hole, that I just hit some unfavourable noise around 560MHz but trying to jump resulted in no post. All in all I'm happy to find out its not the cold, nor me messing up the board by prepping it. It is something that I should be able to tweak once I find the right path. Thats what I like about 775, its never the same. I tried to apply my limited knowledge from the Asus boards and instantly fell on my nose. |
|
|
|
Post by Macsbeach98 on Jul 12, 2020 19:25:31 GMT -5
From memory my EP45-UD3P stops around 560 I havent tried too much to get it to go faster.
My Rampage Formula will go to 620 with 2 sticks in and no tweaking except big NB volts thats X48 though its what I will be using for the TC.
You will get there as you say they are all different.
|
|
|
|
Post by osmiumoc on Jul 21, 2020 10:14:19 GMT -5
I'm making some progress. Currently able to boot at 585 and run some 2M SPI at 590 FSB. 
Took me a couple hours to figure out. Basically you need to set everything manually. And I mean everything. First and most important step, figuring out your DRAM clock driving profile. For max. frequency on the 1:2 divider the 1066 profile did well. For high FSB clocks this no longer works, you need the 800MHz profile. For Micron D9s. Any other IC and its a different story. Atleast for me all other profiles result in a no-post scenario with FSB clocks beyond 500. Second you need to tune your DRAM termination to your sticks. My Microns preferred a little less then auto. Don't forget to set all DRAM references accordingly, they have to be within 20mV of the termination voltage. Once you got these roadblocks out of the way you can probe your way forward. There is no need for excessive voltages, in fact setting NB voltage or VTT high never showed any benefit for me. Currently I can run 585MHz @1.26V NB but I set it to 1.34V in order to have the headroom for 600+. Same goes for VTT which is at 1.24V at the moment. Your main instruments to stabilize the FSB clock are MCH reference and the CPU references (GTLs). And they are all affecting each other, so sometimes you need to try multiple changes together instead of testing them one by one. Currently I'm a bit stuck again and have not found the key to go beyond 592MHz but it's probably just a matter of time. |
|
|
|
Post by ShrimpBrime on Jul 21, 2020 12:55:04 GMT -5
Just run pimod 1m or 32m if going for points. No need to run 2m. Only need to be stable for that 1m  |
|
|
|
Post by osmiumoc on Jul 25, 2020 18:21:31 GMT -5
I just run 2M and 4M to save some time. When things are unstable you can still pass a couple 1Ms but trying 2M/4M trips most BSODs or freezes instantly for me. Now getting to this point here took me some time and bits and pieces of my sanity:  So I started hitting freezes with no memory errors or BSODs at 592MHz which were not related to any voltage or reference. I finally made the connection when I got a rare BSOD with 'A clock interrupt was not received on a secondary processor within the allocated time interval' Thank you windows for this specific hint -> There is clock-drift going on. Once I read that it started to make sense, the freezes seemed to be load related. I could idle rather well @592mhz but loading the bus resulted in a hard freeze (reset signal not getting registered). So I added clock skews aaaand. Well. Same 592MHz wall but this time no freezes at FF but occasional BSODs and stopcode CE (critical error I presume). I already knew this behaviour from RAM OC and I knew that CPU->MCH communication is working as it should and it was time to tidy up the RAM->MCH connection. After tinkering with the termination and DRAMtoMCH ref voltage some more I managed to get to 600FSB. But barely. On the run above I had the explorer stop working in the background. So its far from stable. Thing is I can't figure out any way forward from here now. I'm thinking I might max out my 800MHz driving profile since I'm now going past 1200MHz DDR2 already but none of the other profiles work. I tried them all with different termination voltages and references. Some kinda make it to a boot attempt but fail there. Others just end in the board beeping a 'reliable power issue' beep-code immediatly after turning it on. My current signal tuning is so much on the edge that even going down OR up just one tick in DRAM voltage (currently at 2.20v, so trying 2.18V or 2.22V) result in a non-post scenario. I tried to let the board set a few things on auto again but it just won't post with anything on auto. Last resort now is touching up on signal pull-ups and pull-downs. I'm kinda afraid to go there as this means MANY new different settings and possible combinations. I already spend 20 hours getting here and trying all of those with all the possible references is going to be a 200+ hour job since I have no clue where to start. Today I found the first white hair in my beard and I started to read the 600-page Intel technical documentation for the P45 chipset family before sleep. I also started to memorize the Samsung DDR2 SDRAM 'Device Operating and Timing Diagram'. If I ever start to write on my walls someone has to stop me EDIT: Gonna use this post to hold a few thoughts before they slip. Went back to 500 FSB on the 5:6 divider to test my max. memory clock. Went down on DRAM voltage to 2.16V. I kept all my voltage settings manual at the point they were @600fsb. My RAM clocks to 624MHz stable for 2M passes on the 5:6 divider. I have set everything manually besides the DRAM driving profile and pull up/down levels for Data, Cmd, Ctrl and Clk. The auto driving profile is NOT 800MHz, as it turned off to switch over. Found out its the 1200MHz driving profile that it sets on auto as this one is now working when I choose it. I can boot at 507MHz FSB and pull the FSB up in windows to 520MHz to reach my 624MHz DRAM clock. I can run 2Ms on this profile. But as soon as I try to boot at 508MHz I get no post with various stopcodes. So. The board must be changing something when I try to boot at 508 instead of 507. Confirmed this multiple times. Only things the board can change there are the Pull-up levels and DDR channel A/B clock skews as these are the only settings I have not yet controlled manually. If its not that, it must be some internal MCH timings that I have no control over. I set tRead manually, I set primary and secondary timings manually I set Phase Adjustments for both channels manually I set the same bank / diff bank timings manually and disabled write training. Still I can boot at 507 and use setFSB to get to 520MHz stable but I can't even post at 508MHz. Mamma mia. My P45 Chipset has Rev. A3. |
|
|
|
Post by alpi on Jul 27, 2020 10:48:21 GMT -5
Well, I guess You should ask, what Nb voltages used when even the nb is subzero and want to real high fsb ! If I remeber well, it's much more close to 2V than any 1V ! To tell the truth, I had a P45T-Extreme in the past but I sold it. This was nearly the only unique, beautiful, older hw what I sold. Not matter on performance. I can't like that board. Actually I felt more and more antipathy with every time I tuned on... I've seen a few ppl was able to do insane clocks with them but at the end I didn't even was curious how they did.  Noxinite used to work on his one and (as usually) can do nice things, if You familiar with Him, maybe worth an ask. |
|
|
|
Post by osmiumoc on Jul 28, 2020 14:45:33 GMT -5
I finally grew tired of the RAM sticks I had in there and decided I'll give the next kit a chance. I initially used my generic Micron GKX sticks and switched to Transcend Axeram (also GKX). For this I had to adjust a few things but I noticed a big improvement -> The auto driving profiles work with these Transcend sticks. And I got a little bit further already:  Whats interesting to me is, that both the generic GKX and the Transcend GKX perform rather similar on the 1:2 divider for max. DDR clock. BUT the Transcend can run more sub-settings on auto and reaches a little bit higher on 1:1. So my theory: The different SPD profiles for these sticks result in the mainboard reading them different. The Micron sticks only have a profile up to 400MHz and stop there while the Transcend sticks have a profile up to 600MHz. So with the Transcend RAM it has some reference what internal timings to use for higher clocks while on the other sticks it doesn't have any reference and probably tries to use the default 400MHz settings. This would also explain why it ran fine on the 800 driving profile (400 actual clock) but none of the others. Since both kits can reach very high RAM speeds with a divider I want to try spd-flashing in the future. Maybe I can make these Micron modules work when I give them the profile of the Transcend RAM. For now I'll use the Axeram until I hit a wall again. alpi I might ask around on the HWbot forum but messing around like this with the EP45-Extreme has been fun. Some days I wish I could just copy someones settings and be done with it but usually I enjoy the journey, fighting the settings until it finally works. It gives a nice sense of achievement. |
|
|
|
Post by alpi on Jul 29, 2020 3:19:53 GMT -5
"usually I enjoy the journey, fighting the settings until it finally works. It gives a nice sense of achievement. " Absolutely this is what gives the real enjoyment in this "game" for me too ! No wasted time exists, useless experience isn't really exists I think. Sooner or later it will useful or simply just helps to understand some other mechanisms, connections a bit later ! Easy win isn't a real win ! |
|
|
|
Post by osmiumoc on Aug 1, 2020 10:58:45 GMT -5
My next wall I hit was at 617MHz. I could run a 1M at 617MHz and barely made a validation at 623MHz. I noticed a weird behaviour of the RAM timings. As soon as I put some load on the RAM the the tRP value jumped from 5 to 13 automatically. It happened everytime I validated or ran SPI at 617 or beyond. I also had to increase my NB/MCH voltage for the first time, 1.36V was just enough for 615MHz. I'm on 1.40V from now on. VTT is 1.28V atm. I suspected the RAM again as the limiting factor but higher voltage then 2.20V made it even worse, crashing at 610 already. Instead of spending hours to tweak this one I decided to just try a few sticks out since I now have a small collection to choose from. Next in were Adata Vitesta Extreme 1000+ sticks. I had 2x kits in 2x1GB config and both kits failed to boot at 612MHz. They posted fine but could not boot. Again higher voltage was no use. So I tried my Crucial 10th Anniversary kit and this one worked.  It does 620MHz fine and I could even lower the voltage from 2.20V to 2.16V for this. Since these sticks seem to be rather special I put them to the side as spare. I'd rather not burn them out yet. Next in went some SUPER*TALENT sticks and they failed to post at 2.18V but also managed to run 620MHz fine at 2.24V. Since these seem to scale well with voltage I'm going to keep them in and try further Maybe I'll also take a break from hunting FSB clocks and continue with my GKX vs. GMH comparison. EDIT: Noticed a weird behaviour with GKX. It works very well in the first slot (up to 625MHz fine) but cant even post @600mhz in channel B. The further I move the stick away from the CPU the worse it gets and gets bad fast. In the outermost slot it does not even try to post it just beeps no RAM detected even at 580MHz... I tried channel B and individual DIMM clock skews, no success. ANY setting besides auto = no RAM detected. I tried tuning the individual channel reference voltage up and down = no changes. I find it rather intriguing how so far only my generic Micron GKX show this behaviour. I tried other sticks from kits like my Crucial Anniversary RAM and the SUPER*TALENT and they don't care that much in which slot they are. EDIT 2: This lead me to an idea. If they only work in the first slot I can mix them with my SUPER*TALENT kit and create two working kits so I have another spare! Well usually its a bad idea to mix and match different RAM with different timing profiles... but not in this case!  No I did not flash the SPD to make them look different, those are two completly different sticks (pcb is a bit different too):  This even got me up to 625MHz.  But beyond this point I seem to need more and more VTT and MCH/NB voltage. So I'll probably move this back to dry ice now. I feel like I'm a bit more prepared this time |
|
|
|
Post by osmiumoc on Aug 4, 2020 17:32:22 GMT -5
Sorry for double-posting but I also don't want to stretch my posts towards infinity so that a single thread page takes minutes to load. Went back to dry-ice and it ran perfect today. I'm starting to really like this board. You need to take some time getting comfortable with it but it will reward the efforts. BTW, this is how I used the original NB heatsink and slapped a crudly home-made pot on top:  
By cutting the heatpipes to the VRM section you ensure that there is no nasty condensation on top of your VRMs. You can leave the southbridge part on, it adds stability when mounted and due to the heatpipe freezing + heat from the southbridge it does not build condensation. It is a good idea to wrap the pipe thats running towards the southbridge tho.  I ran like this for 2,5 hours today. Back of northbridge reads -34°C @1.46V. I got all the way up to 680MHz FSB: valid.x86.fr/kxabjwThe RAM is my limiting factor still, NB was fine @1.46V for the 680. I did not check if a lower voltage was sufficient too, I just let it sit at that level and did not care about it.   It did 32M fine above 610MHz, for this my CPU was the big roadblock. I did not use a binned chip I just wanted to see if I can control the board now. So the CPU was a bit weak. And I pushed it up to 1.97V at some point which it hated. Lost a few MHz after that. Probably to be expected from a 45nm. Single core workloads did okay-ish but multi-thread tasks it tapped out at 5500MHz max. at 1.88V. Time to dig up some decent CPUs. And I need to get XP or Server 2003 on a disk. Still need to find some install media for those. My PI scores suffer a lot from running Win 7. |
|
|
|
Post by Bones on Aug 4, 2020 17:45:59 GMT -5
Looking good from here!
|
|
|
|
Post by Mr.Scott on Aug 4, 2020 18:50:50 GMT -5
Getta quad core on that and sub us a 32M for the team cup You are very good on 775 and DDR2. |
|
|
|
Post by Macsbeach98 on Aug 5, 2020 0:01:36 GMT -5
|
|
|
|
Post by osmiumoc on Mar 4, 2022 18:10:58 GMT -5
Man I really wish I could find some good material on DRAM driving strength settings and what exactly they are and do. My EP45-Extreme has them as profiles from 667 to 1333 (OC). And on air cooling I can choose the 1066 profile and fly past 1200 on the RAM easy. But as soon as I freeze the RAM it needs the highest setting (1333) for 1200 and stops to work beyond 630/1260. It hangs in the same manner on code 52 which is final memory test indicating that the issue is again with the driving strength settings. Any other setting has no change. If I keep the RAM ambient I don't run out of driving strength, so is there a way to increase or otherwise influence this? If I run out of driving strength, to my limited knowledge that indicates there is a lot of signal noise or interference on either line of the RAM. Like data, clock, command etc.. Question is where does that noise come from and how can I tackle it. Does anyone know a place to deep dive on information about such things? I can see how signals from RAM are supposed to look from microns documentation about DDR2 but I did not find any good guide on what tools there are to manipulate signal lines. I wonder if I could somehow rig a logic analyzer or oscilloscope to a DDR2 module under cold and see how it behaves. Wishful thinking that is, I'd never be able to pull that off or afford the testing equipment |
|
|
|
Post by Macsbeach98 on Mar 4, 2022 20:56:19 GMT -5
P45 and DDR2 has always been a pain in the ass I have virtually given up on them especially after running that P5Q
I much prefer X48 DDR2 Rampage Formula drop the ram in and run the CPU cold and go straight to 620 bus 9.5 multi with a E8500 without any drive strengths or other bullshit and be completely stable.
|
|
|
|
Post by osmiumoc on Mar 4, 2022 21:09:37 GMT -5
On some rainy days, even I wish to try a DDR3 775 board. But my heart is with DDR2. I have no idea why. Some boards seem to deal with these 'hidden' settings on their own pretty well. Might have to give Asus another go. Still, somewhere in all these settings must be a way forward. Like I speculated in my other post on general comments, cooling the RAM was probably not an improvement. I just wish I could 'see' into these parts as they run and keep an eye on how the signals interact and change. All we can do as mere mortals is to feel our way forward in the dark  |
|
