DFI UT X58-T3EH8 continued - Part 2

[Peen]

My 2nd board (2nd RMA) should be here tomorrow. Actually kinda scared to try it again though, because I gotta leave on the road with it in a week. Might just bring the new Athlon rig instead

[ChaosMinionX]

memtest86+ 3.0

approved..
finally a good memtest to find mem skew aka bootup bclk
vtt voltage
and dram voltage..

what u get here is the same as prime.

I used Memtest86+ 3.0 beta 3 and I passed 15 passes of it without any error with my settings, when I get into windows and run LinX max memory I fail with a BSOD and get code 124, which I have found to be VTT/Bclk related.

Could it be that this code is also vcore related as well?

[Peen]

Damn, got my board back I think it's probably from someone on here from RMA lol!! The back metal pieces are all rusty probably from phase and the bow is bowed like a U. But it works, this one buzzes too though. Guess they just all buzz underload. But man, finally got a SSD and it freakin flys now

[cstkl1]

I used Memtest86+ 3.0 beta 3 and I passed 15 passes of it without any error with my settings, when I get into windows and run LinX max memory I fail with a BSOD and get code 124, which I have found to be VTT/Bclk related.

Could it be that this code is also vcore related as well?

think u guys dont know how to use it

to find the vtt voltage
u need to run lower mem skews
u will see the difference

its the same way we use to test skews for REX.

remember the board bios has its limit depending on ure ram chipset
my rams will not do more than 2050 CL7 2-4 hours prime stable.
2100mhz is a no go

but it was a different story with my kingston T1's. which did 2150mhz CL8 stable

[ChaosMinionX]

think u guys dont know how to use it

to find the vtt voltage
u need to run lower mem skews
u will see the difference

its the same way we use to test skews for REX.

remember the board bios has its limit depending on ure ram chipset
my rams will not do more than 2050 CL7 2-4 hours prime stable.
2100mhz is a no go

but it was a different story with my kingston T1's. which did 2150mhz CL8 stable

Could you elaborate on this? What do you mean you need to run lower mem skews?

[jcool]

Damn, got my board back I think it's probably from someone on here from RMA lol!! The back metal pieces are all rusty probably from phase and the bow is bowed like a U. But it works, this one buzzes too though. Guess they just all buzz underload. But man, finally got a SSD and it freakin flys now

Mine's got a rusty socket, but I won't RMA it...

[Peen]

Mine's got a rusty socket, but I won't RMA it...

That's good lol! I shoulda just asked them to fix my old board cause this one gets about 25c hotter on the PWM temps

[jcool]

Try leaving Vdroop control on enable... it droops but it doesn't buzz (as loud)

[Sam_oslo]

After trying 4.3GHZ(205x21) 1.28v and 4.2GHZ(210x20) 1.24v for a while, missed my low-volt summer OC, and had to check it out again.
Summer is over and it is running so cool at lower ambient that it is really tempting to run @4GHZ (with passive cooling ? ) for 24/7.


UnCore:QPI 1:1. both at x18, still running strong with VVT=1.21v(BIOS set)
RAM at x4, @1688 7-8-7-20 1T 1.65v

@4GHZ(211x19) 1.15v

[cstkl1]

Could you elaborate on this? What do you mean you need to run lower mem skews?

well since i am using elpida hypers
the best bios is 04/28

the mem skews are based on bootup clocks
and the vtt is corresponds to it

example for 2050 CL7 to get RTL 53 on Channel A u need to use bootup bclk 185
for set bclk 205
this is tricky as its borderline.
bootup bclk is between -20 to -5 of set bclk. vtt is corresponding to this
so what vtt or dram dq vref i should set??
this takes time and need to do a lot of testing
but overall its either channel dqvref Channel A + 3.0% with cmd Vref + 120 or -2.5% with cmd vref 110
this will pass memtest86+ but will have some i/o bsods aka nvidia/raid etc on boot ramdomly. so channel B and C need to be set also. to get optimal rtl of 53/56/58 at vtt 1.44v set.

for X58 u need to always test on different boots. cold boot, hot boot. resets etc.

testing of vtt. easy set bootup to 185 for the above with no adjustments and lower vdimm.
for me i need 1.65v which is proven by prime. so i set 1.635.
u can see errors in test 2. which correct vtt will reduce it.

in REx 775 mobo's we tested the same way.undervolting certain voltages to find the others. thats how u test with memtest86+ 3.0. previous memtest versions was not very sensitive. this is especially to mem skews and rtl's of the uncore.

look at everybody overclocking its so boring all high cpu clocks with slow mem ..
progress on x58 overclocking would be with higher cpu clock with high mem speeds
until to date only one person has done 2050 4.3ghz 24hours prime stable on evga classified
what about 4.4ghz 2100mhz etc..

[Sniper]

Why do you want high mem speed over cpu Mhz ? Most test show that increasing cpu Mhz gives way more performance than increasing mem speed.

Running high mem speed also requires high uncore clocks, wich normally requires high vtt volts, something not everbody feels comfortable with.

[ChaosAD]

look at everybody overclocking its so boring all high cpu clocks with slow mem ..
progress on x58 overclocking would be with higher cpu clock with high mem speeds
until to date only one person has done 2050 4.3ghz 24hours prime stable on evga classified
what about 4.4ghz 2100mhz etc..

Whats the point of running such high ram clocks on i7 when either 1600c8 or 2200c7 wont make any diffenece in everyday pc use other than benchmarks

[cstkl1]

Whats the point of running such high ram clocks on i7 when either 1600c8 or 2200c7 wont make any diffenece in everyday pc use other than benchmarks

right
then whats the point in spending USD500 on some rams.
or even bother getting 1600mhz C8.. why not just get 1333mhz C9 sticks since
dram multi 10 bclk 220 results in cpu speeds of 4.6ghz.
again ure finding the limitations of a certain hardware.
and not understanding at all the rest.
why 2100mhz C7/C8 4.4ghz is a hard thing to do.?? why??
u have good cooling/ good scaling proc/ good mobo
whats stopping anyone??
this is where the term " know thy limitations"

if the rams can do 2050 C7 1.65v
it can do 2150-2200mhz C8 1.65v
so whats stopping it??

dunno about u but when i overclock i want everything to be overclocked equally.

hence ram, qpi, uncore and proc.

overclocking proc on x58 even a retard can do it by just upping voltage with a good proc and good cooling
the rest is different story. and the ppl who do it will actually feel the difference in each bios.
most bioses for any mobo the biggest difference is the ram performance and overclocking

[Sam_oslo]

High MEM speed is cool for testing&benching, but it is not exactly the sweet spot for 24/7. My sticks can run stable at 2150MHZ CL9 with much better benching numbers, but it needs a good deal more VVT to run stable for 24/7.

I had a hard time to track the effect of high RAM speed on my daily applications, but it is cool to run on the edge with x5 MEM, just for the heck of it. The only problem is that x5 RAM is far from the sweet spot for total 24/7 OC.

In my experience and on my setup, the 1.32v VVT needed for stability of x5 MEM puts 3 degrees on CPU and 5 degrees on PWM/temp, both at Idle. This extra heat saved on a 1.2v VVT for x4 MEM can be used for better OC on CPU, for sure. So it is safe to say that running MEM at x5 is not exactly the sweet spot for 24/7, at least on my setup.

[cstkl1]

Why do you want high mem speed over cpu Mhz ? Most test show that increasing cpu Mhz gives way more performance than increasing mem speed.

Running high mem speed also requires high uncore clocks, wich normally requires high vtt volts, something not everbody feels comfortable with.

most test also shows theres no point in going above 4ghz on a i7..
so why do u??
because ure cooling can take it.

same theory applies if ure rams can do it.. whats stopping it??
mobo?? bios?? graphic card???
this is what overclocking is all about..
understanding limitations.
what happens when u havent surpass cooling/voltage.. whats stopping u.

whats the dqvref for?? whats the cmd vref for?? whats the rtl settings has to do with uncore and dram skews and vtt voltages??
.. it seems alot of ppl dont even try

[cstkl1]

High MEM speed is cool for testing&benching, but it is not exactly the sweet spot for 24/7. My sticks can run stable at 2150MHZ CL9 with much better benching numbers, but it needs a good deal more VVT to run stable for 24/7.

I had a hard time to track the effect of high RAM speed on my daily applications, but it is cool to run on the edge with x5 MEM, just for the heck of it. The only problem is that x5 RAM is far from the sweet spot for total 24/7 OC.

In my experience and on my setup, the 1.32v VVT needed for stability of x5 MEM puts 3 degrees on CPU and 5 degrees on PWM/temp, both at Idle. This extra heat saved on a 1.2v VVT for x4 MEM can be used for better OC on CPU, for sure. So it is safe to say that running MEM at x5 is not exactly the sweet spot for 24/7, at least on my setup.

post a 2150 C9 4.5ghz. .. u havent even tried stability and gave up because u assumed the vtt was hindering u..

lol i am runnign 4.3ghz 2.1k CL7 24/7 non stop for 5 months. with more than 100 hours of stressing on my mobo/proc and rams.

challenge.. post dram 2050 C7/C8/C9 prime 8 hours stable 4.3ghz HT On/OFF..

hint hint it isnt the vtt thats causing the issue.

the dfi mobo and evga is a dram overclocking king.
compare to gaygay and asus.

[cstkl1]

also another note on some of the xeons and 950
they have higher multis on cpu

whats stopping anybody from running 4.4ghz 200bclkx22 with dram at 2k C7..
have u seen anybody try before??
y 4.4ghz 1600mhz is the same as 920 4.2ghz dram 1600mhz while the 920 has no issue to run 4.2ghz 2000mhz CL7
but the 950 does..
all this is based on idiotic overclocking by increasing voltages only.
whats stopping it.
uncore to cpu speeds??
why high dram speeds above 4.2 is just that hard??
this is the questions i have been trying to answer with multiple kits and multiple boards and multiple cpu's.
everything points to qpi related and dqvref so far

[Sniper]

Sure, I see your point about pushing the edges, and I'm glad someone does. But for me its all about performance, and if I have to choose between cpu mhz and mem speed (and I do), I choose cpu mhz every time.

Personally, I dont have a clue dqvref does, and so far I'v managed ok without it. Would I like to know it, sure, but I dont have the time (yet) to learn all about mem tuning.

I saw someone say that they wished subtimings didnt exists, and thats almost where I'm at too.

[cstkl1]

one last note also..
have u guys actually tried stock cpu speeds with dram 1066,1333,1600,1866 and hardest of all 2133mhz?/

hint hint 2133mhz .. u need to adjust other settings..
not so direct as to voltage alone.

betcha not many has tried.

[cstkl1]

Sure, I see your point about pushing the edges, and I'm glad someone does. But for me its all about performance, and if I have to choose between cpu mhz and mem speed (and I do), I choose cpu mhz every time.

Personally, I dont have a clue dqvref does, and so far I'v managed ok without it. Would I like to know it, sure, but I dont have the time (yet) to learn all about mem tuning.

I saw someone say that they wished subtimings didnt exists, and thats almost where I'm at too.

no u wouldnt be saying that if u think it more thoroughly
ure case is true if all dram manufacturers collapsed and there was only one chipset in the market.
what if down the line u upgraded ure rams by adding another 6gb's but the chipsets were different..
thats when u would be gratefull on subtimings.

[jcool]

I saw someone say that they wished subtimings didnt exists, and thats almost where I'm at too.

That would be me

Seriously I hate ram tuning. Neither do I understand it properly (on x58 at least) nor do I have the time to get into it.
Just buy two matching 6GB kits with either HCF0 or Hypers and you can run high Mhz pretty hassle-free, that's my approach anyways

[Sniper]

Ram tuning is somewhat the holy grail of oc, and I admire those who have the skills and time to learn it, but the time part excludes me.
I enjoy tinkering with my cooling, running prime/linx trying too achive my ultimate goal, as high as possible performance with as little nois as possible.

[p2501]

Thanks for the input, but I'm on linux atm so HyperPi32M won't work. But I found a proggy called memtester which seems pretty thorough. We'll see what and if it finds something.

[jcool]

Ram tuning is somewhat the holy grail of oc, and I admire those who have the skills and time to learn it, but the time part excludes me.
I enjoy tinkering with my cooling, running prime/linx trying too achive my ultimate goal, as high as possible performance with as little nois as possible.

Same here..

[ChaosMinionX]

Oddly enough plugging in the floppy power connector just below the mosfets fixed all of my problems.

Now stability testing 200x20 1.35v (in bios) 1.3v vtt.