X58 ram oc

[somerandomtechyboi]

welp i think its time to start an actually serious thread since ive been going at this for a few months now and ive actually binned some cpus


ive already blown right past the 3200 and 3300 mark all the way to 3400 so now its just trying to see the highest freq i can stabilize and see what kinda scores i can get on spi

yes this is a binned chip and ive bought 9 w3503 so far and definitely gonna buy more cause i can get em for pennies and i do intend on buying a gulftown at some point just that im saving my money for now and reconsidering my options as i cannot trade for shit on a small forex account and i prefer to have some sort of income source that isnt driving myself insane with broken boards and getting next to nothing for it

for important voltages and settings
vdimm at 2.3v
vtt at 1.94v (seems to be the sweetspot for most w3503)
vpll at 1.4v (1.3v caps bclk oc even with slowmode)
vioh at 1.4v
bclk in bios at 207 for 3312 boot
ram multi 16
qpi slowmode on
pcie 115

havent tried pushing any higher cause i just got the 3400 screenie and ended the session with that then never bothered to push that high again cause im starting to suspect that i may be hitting ram limit or hit it pretty soon considering the timings and already needing 2.3v vdimm and id rather not waste this sample nor the 2nd best cause 1.94v vtt definitely seems to degrade it somewhat quick as in a few windows runs atleast thats my observation on the 2nd best sample before i got my best one so far and im preserving the 2nd best since it has a better 2nd channel (can post 3232 in bios where the best one cant) and im currently using it to see max stable freq on 2 sticks as my i7 930 tops out at 2900 with bclk 182

most of the w3503 samples are pretty shit with around half being able to post at >3000 with the others being complete trash and the only decent samples being these 2 best chips that can run >3300 and another one that is about on par with my i7 930

i have still yet to test max uncore though i think i managed to get my i7 to post 4500 in bios at some point vtt 1.86v just have to test these w3503

and it seems like these w3503 can actually handle >1.7v vtt without being completely unstable unlike the i7 cause i am currently testing 2900 with vtt 1.8v on my 2nd best w3503 and it seems to be helping whereas the i7 would crash instantly as soon as i go over 1.7v vtt, just that i need to keep this thing cooler so i dont degrade it too quick cause mid 70s atm


currently im having some real difficulty getting >207bclk to post without qpi slowmode cause as soon as i go from 207 to 208 it visibly posts slower both postcode and display and it fails to complete post but hits the bios logo thing and same goes for all my other bloomfield chips so im pretty sure im doing something wrong here settings wise as the clock drives dont seem to help nor do the clock skews and vpll doesnt seem to help above 1.4v same goes for vioh, pcie set to 115 and ive also tried 120

im also looking at a board swap cause this x58a ud3r caps at 2800 on the 3rd channel and i do wanna try triple channel oc once i get a gulftown or find a workaround to get 1:1 uncore stable on bloomfield with triple channel cause over 2400 and i need to have uncore multi atleast 3 multis above the ram multi to get it to post otherwise stuck on c3 or c5 code, theres a g1 assassin i can pick up for 40$ but i have my doubts since thats an eatx board, theres also an evga ftw3 for 50$ and afaik like most non amibios boards it should be able to do 1:1 uncore but i have some concerns on if theres a bios thatll run 1:1 uncore cause  x58a ud3r can run 1:1 uncore on fa and fb, ff doesnt work and gets stuck on d8 code, fc fd fe are not available on the gigabyte site anymore and i cant find any downloads for em, so i have my doubts on if theres a bios version that doesnt have a broken 1:1 uncore, theres even a foxconn renaissance at 43$ but im not sure if the bios will allow 1:1 uncore and the only time i did have access to one (crossflashed it onto a shuttle x58 board) i did not think to check the uncore ratio as i had no idea 1:1 was even possible and this was before i even got a working x58a ud3r so i also have my doubts on this board, no flaming blade or bloodrage available atm but ground reported not being able to get below 2:3 working on 32nm chips so also have some doubts on those boards having a bios that works with 1:1 uncore, and x58a oc are unobtainium unlike the common asus rampage boards which cant do 1:1 cause forced uncore ratio so thats not an option either, most promising option out of those boards is the g1 assassin and thats an eatx board

and im also looking for a ram ic thats lighter than samsung gdie as these seem to be pretty heavy on the imc but also clocks high (3600+) and the only options im seeing are 4gbit hynix mfr and 1gbit/2gbit hynix bfr/dfr as everything else either clocks too low or needs to be under cold (good binned psc)

[cbjaust]

Nice one. Ask Mr.Scott if he has the bios files you're after.

[somerandomtechyboi]

uhh cant upload files but i did upload some photos on an ltt thread linustechtips.com/topic/1570137-w3680-bclk/#comment-16410202

basically ive went ahead and bought a w3680 and yeah pretty dissapointing as 3400 wont even boot in bios, just the same 3312 as my best w3503, though i have managed to get 3000 booted to windows with 3 sticks uncore multi at 20 and ram multi at 16 (5:4), might potentially be able to stabilize it as running prime95 doesnt instacrash all the threads like on bloomfield when i run anything inbetween 1:1 and 2:1 uncore and 1 actually stayed running despite how unstable the oc is

ive also run into bclk issues with the w3680 but now im starting to suspect the board as 32nm will do 250bclk on average for most cpus afaik so unless i have a complete dogshit sample might be the mobo, luckily i have a spare x58a ud3r so ill test on that see if anything changes bclk wise as it struggles to even post 220bclk with slowmode on which is kinda reminiscent of my trash p6x58de that isnt stable above 199bclk, and now that i think of it with the 207 bclk limit without slowmode i think this thing might just be as trash as my p6x58de, damn i hope binning mobo ram slots isnt a thing otherwise im literally gonna have to build a bga rework station and salvage an x58 chipset off another board which i actually have the money to do now like 220$ ish but im pretty close to my target 300-500$ for forex shenanigans so i dont insta blow my account so probably gonna hold off on splurging too much on hardware till i have a consistent source of income though ill buy some on occasion as ill drive myself insane if i dont =p

if this w3680 turns out to be trash which its kinda looking like it is considering i cant hit 3400 in bios then ill probably just resell it and buy a w3670

im also getting close to ram limit as 3400 already needs 2.3v and thats on my best stick and i need a 3rd stick thats good so i can properly test 3000+ triple channel as my crappiest stick (middle stick in the photo) needs 2.3v just to boot 3000 so im gonna have to find more gdies, and im also interested in buying some hmt112u7bfr8c as 1gbit bfr is apparently pretty good for 3000+ aswell and an ic thats lighter on the imc would help, apparently elipda bbse and bdbg might be able to run 3000+ but i have my doubts that theyll beat gdies on air and id rather run on air till im really maxed out then go subzero cause its fun destroying subzero frequency with aircooling


currently im gonna test 5:4 ratio stability as triple channel works and boots to windows, but since one of my sticks is trash ill test dual channel first, doesnt help that the triple channel profile doesnt play nice with dual channel so im gonna have to make a specific tune for 2 sticks, dont have a particular aim other than as high as possible but it would be nice to hit 3200 maybe even 3600+ on subzero for triple channel, and 2 sticks does run at 3312 unlike my 2nd best thatll post 3248 on 2 sticks (best w3503 has a weak 2nd channel) so i also have suspicions that im being ram limited alongside bclk limit

ill also test different bioses for this board as i wanna get the 18x multi working to deal with the stupid bclk limit if the cpu turns out to be trash for bclk and not the mobo, and i have managed to find some beta bioses over here www.tweaktownforum.com/forum/tech-support-from-vendors/gigabyte/28656-gigabyte-latest-beta-bios/page2#post432188
just need to find the other deleted stable bioses (fc fd fe)

as a reference as to what bios versions might work i found this video  where someones managed to get 18x multi working for 2400 ddr3 at 133 bclk, bios is f2q which seems to be a modified bios for the ex58 ud3r rev 1.0, i suspect its a modified f12q but id like to verify what its actually based on, apparently also available on tweaktown but the link i found in this thread is dead seemingly due to forum changing its address so that doesnt help www.overclock.net/threads/xeon-w3690-on-ga-ex58-ud3r-1-0-bios-f11.1734662/

hesitant of newer bioses due to broken 1:1 uncore but if 5:4 works and is stable alongside 1:1+1 then i could care less if 1:1 doesnt work as i dont think ill be uncore limited till 4000+ and i dont expect to hit that on air but damn ill still try anyways

after i get either 18x multi or bclk working then ill screw around with max 1 stick freq, anything above 3300 for 2 stick gets stuck on c1 and same goes for 3450-3500 on 1 stick which for whatever reason wont work at all at <3450 and rolls d8 02/04 code

ill also wanna test max uncore and max freq at 3:2 with my aim at 2600-2800 stable

as theres lots of stuff to test and do ill probably make another post pretty soon

[ShrimpBrime]

This you??

www.reddit.com/r/overclocking/s/Be5dEiZyb0

Triple channel C10 2830mhz?
60 hours tested?

I require a screen shot to believe it son. So far you've done single channel here at these speeds.

[Macsbeach98]

2830 mem as loose as diarrea isnt fast.
I will race triple channel Hypers 6-7-5-20-1T 2050mhz.
I will be playing with secondaries today and its fast it will easily be No1 SPi32 on HWBot W3690

[ShrimpBrime]

I'm ready for that 60 hour P95 Time stamp here....

[somerandomtechyboi]

Jul 10, 2024 16:31:20 GMT -5 ShrimpBrime said:

This you??

www.reddit.com/r/overclocking/s/Be5dEiZyb0

Triple channel C10 2830mhz?
60 hours tested?

I require a screen shot to believe it son. So far you've done single channel here at these speeds.

ive already done 2800c11 stable on bloomfield albiet dual channel as triple channel needs +3 uncore and anything between 1:1 and 1:2 boots but is completely unstable, dunno if i have a screenie of that dont think so anyways maybe ill bother rerunning with some better ics for close  to 2900, as for triple channel 32nm can run 5:4 and 11:8 i use the former as the latter doesnt seem to work past 2800 even if i set 1.6v vtt in bios

i was intending to run it for longer till i decide i want to toy around with it more so 2 days went by just fine but 3rd day fell asleep at noon woke up at dusk and for some reason it was off so yea fucking power outage as it was off but powered

dual channel 3008c11 runs for 2 hours till it starts erroring out and i dont think its worth doing an spi run for funsies till i get it to 3100 and for that im gonna have to remount this thing again cause inner channel even though its better than the previous mount (stuck at 2720 ish for full stability) its still garbage and doesnt clock past 2850 stable which i can test with my 3008c11 profile where thatll run for 2 hours on the 2 outermost channels but if i move the middle stick to the innermost channel it crashes within a minute and i have verified that middle stick doesnt hate the inner channel infact its the stick that can clock the best in the innermost channel in its crippled state

ill post a screenshot when i get it higher as i cant be arsed to go run the same 2832c10, no fun when its already proven stable, if i cant get it to clock higher oh well theres my w3670 thatll run 2800 on 3 sticks aswell so i can just take a screenie of that and be done with it


for now ill just have these screenshots, ones just an spi run at 2960 triple channel cpu at 4.6ghz just seeing how far i can clock it and still pass spi and the other is that aformentioned w3670 but 2800c11 for only 12 hours so gonna have to run it longer, i dunno why i set it to c11 when its running at 2.3v same volt i need for 2832c10 but oh well gdie just so happens to be very overoltage tolerant given the frequency, both on air cause im pretty sure theyll clock higher under cold and with a better board as this ud3r hardwalls at 3008 on the outermost channel for bios boot

im just doing this shit for fun rather than max perf since high ram oc on x58 is uncharted territory due to crappy amibios boards with their forced uncore : memclk ratio alongside whatever that 1.35v max vtt nonsense is which im starting to suspect is a result of one of the two peculiarities ive found on this platform

mounting definitely has an effect at these clocks dunno about 2600 and under but 2600 is close enough and even at 3:2 a decent bin cpu should be able to hit 2700, its most obvious on the w3670 while getting that 2800c11 to run where one mount crashed within minutes, another insta crashed, and the mount im running now resulted in the screenie above, cooler mounting might also have an effect if you push down on it too hard and screw the mount up now all of a sudden what used to be stable isnt stable anymore, first suspected on the w3680 as its downright strange for the outer 2 channels to be so much better than the innermost channel which was complete dogshit in comparison (2700 ish max vs somewhere around 2900 max stable)

theres also just the thing randomly destabilizing after awhile seems to happen after a p95 run, i continue to tweak the thing and all of a sudden the settings i was just running stable arent stable anymore and this is with a bios profile saved so i know its not bad bios settings, ive observed this the most running that dual channel 3008c11 where after i run it on prime95 for awhile and start tweaking again it just destabilizes and i have a bios profile for it as a reference thatll do that 2 hour p95 run so no fucking up the bios settings when its the exact same bios settings, the fix is literally just unplug power leave it for a few hours and boom now its stable again as if nothing happened, 3008c11 specifically needs 1.6v instead of the 1.52v that i run for 2832c10 so maybe vtt has something to do with it but im not sure and idk if this happens at lower voltages cause if it does someone could definitely mistake it for degradation


as for the reddit thread yea need some 2gbit d die timings, i think theyre better than cfr but probably not by much, i hope they atleast dont degrade past 2v cause thats an automatic win if they dont, gdie is great and all but id also like to do some frequency shenanigans on bloomfield aswell with an ic thatd be lighter on the imc cause imc is quite the limiter especially for 2 stick, maybe go past 3400 on my best w3503 before it degrades into an average sample at 1.94v vtt =p  not like i cant bin more of these anyways theyre dirt cheap

[ShrimpBrime]

Well OK then! It's something at least.

[somerandomtechyboi]

Jul 12, 2024 9:21:50 GMT -5 ShrimpBrime said:

Well OK then! It's something at least.

currently working on that w3680 see if i can get it to 2900 triple channel stabily and therefore 3008+ spi to actually run and maybe even go past 3232 for max frequency, not sure if im starting to hit stick limit or not at this point think i should grab haswell voltmod board to run 2.3v vdimm see what these are actually capable of as im basically doing a blind oc past 2900

my sticks are probably too garbage to actually bench and get decent scores, already needs 150trfc for some semblance of stability at 3008 and those are among my best sticks and i have my doubts that elpidas on oem bare pcbs are gonna be any good considering my best psc only does 2656c11 stable at 2.2v

speaking of which heres the only other interesting thing ive done so far maybe i can get it running at 2800c11 now that i have better sticks available, its a cursed mix match overclock with the innermost stick being 1gb psc middle being 1gb gdie and outermost being a 4gb 2rx8 cfr stick, only reason this is possible is because this board allows for seperate timings per channel and the psc is the worst out of them

psc iirc was all auto timings but cl11 and cmd2t cause for whatever reason this board is really good with running pscs on auto timings whrre i could get 3100 posting with just full auto timings, this stick is the limiter on this oc

gdie was just my regular gdie timings iirc (basically hwbot gdie timings i just loosened) at cl10 instead of 11 given the voltage

cfr was 11-14-15 everything else auto should be good till around 2800 ish

trfc and command rate are all matched cause these arent 4gbit ics that cannot run any decent trfc, vtt was either at 1.5 or 1.45v kinda forgot but definitely not above 1.5v, vdimm at 2.2v to get that stupid psc stick stable at this speed, dunno why i had ht on but oh well, bclk is really low cause 16x seems to be the best multi for ram oc on x58 and if i set lower ram multi itd probably crash but usually the outermost stick just stops detecting so thats kinda annoying, id like to try the 18x multi but it seems to be broken

also forgot to mention that the cpu mount can just randomly fuck itself over as i observed yesterday with my w3680 when i was testing my reference 2832c10 oc for triple channel where it crashed within minutes (same profile i ran for ~60 hours), i just did a half remount by unlatching the lever but not fully unmounting it because its convenient it shouldnt cause this thing to 00 or ram channels to die then i have to do more remounting and it shouldnt change its characteristics that much, as of now its around 10 hours (started ~6pm yesterday now its 4:13 am) stable and still going but ill probably mess around with it later today see if i can get it clocking higher just dont feel like messing around with it  right now

[ShrimpBrime]

Need to blow some cold air on them or something. Helps when the memory controller gets cool too. Maybe a water block on the cpu with some ICE water could do the trick?