Wow, i'm running +150 for 24/7 @4.6 ghz. could it be due to my low bclk setting?Quote:
Originally Posted by gmat
Printable View
Wow, i'm running +150 for 24/7 @4.6 ghz. could it be due to my low bclk setting?Quote:
Originally Posted by gmat
Whats it take to run 230-235bclk on a daily basis? Been playing with my westmere today and had 230 booted into windows for a few minutes before a blue screen ruined the fun. Stuck at 215 right now and want more. Any idea's what to increase? Temps are a non issue
http://i234.photobucket.com/albums/e...ckTTC6/215.jpg
If you have low bclk, and reasonable memory timings, yeah. I run low blck (142) but 1:14 ram multi = 2000MHz ram frequency. If i leave everything to auto, and put the XMP profile (to run my ram at its rated frequency), it will give me a VTT of +350 ! Madness !Quote:
Originally Posted by hlonipha
I would say, altering all voltages but IOH makes absolutely NO sense for me. IOH is important for high bclks as well (as well as for multicard usage and full dimm population).Quote:
Originally Posted by Hey Zeus
PCIe clocks, cpu impedance and qpi signal compensation would be some other points to consider. We cannot see what they're set at.
correct bro
also for a few procs i tested
for high dram clocks
u need to change ure skew depending on the pwm frequency to cpu frequency
hint to others
there's a relation between ure pll to ioh/qpi 1.1 voltage.
But change by how much and what kind of skew ?Quote:
Originally Posted by cstkl1
which relation ?Quote:
Originally Posted by cstkl1
normally i would divulge everything
but with a few ppl attitude that .."anybody can do it with a good dram and proc"
all i can say test it bro.
took me a month to get my 2110 c7 complete stable at 7-8-7-20 1N exactly like SAE.. the hit it took from 209 to 211.. eh not acceptable bro because of that write to write/read to read at 6
now reverting back to getting 7-7-7-20 1n write to write/read to read (same rank )=4 stable.
I'm also running blck of 141 with a multi of 33 and the memory at 1:14. Are you running at +350 ? or did you manually set it to lower ? are you stable at a lower volt setting?Quote:
Originally Posted by gmat
I completely concur. :)Quote:
Originally Posted by cstkl1
I wonder why you'd need this much vtt at such a low blck and uncore. I only need a set +150mV at 4-4.05GHz Uncore with both W3520 (3844A704) and i7-950... And the Xeon was at 200+ bclk.Quote:
Originally Posted by hlonipha
I often had to conclude, completely sledgehammer settings with voltages taken from some oc-how-tos were being used without switching on the thing between the ears (used to have the purpose of thinking).
You can collect ideas and experiences of others, but you should always begin from the beginning for yourselves. Learning by doing/testing. Most of the time newbies (for that special platform) come and ask, why this setting and that setting should be like that. They hardly get why as they did not dive deep enough to understand.
I am sure not everybody has the time or is willing to spent this much time in tweaking, but at least you should not use factory/auto setting like xmp not exactly knowing what they do. With vtt you should be very careful and not to exuberant. For many CPUs with D0 stepping 1.35-1.4V should suffice for 4GHz Uncore. And with air/watercooling I'd avoid to go higher than that for 24/7.
Hi,
I have just start playing with my new beast. Any advice here regarding temps and voltage? VTT max?? 980x run very cold..
GT's max Vtt is 1.4V referring to its datasheet.
+350 was the full auto setting, with XMP profile for the Corsair GT's.Quote:
Originally Posted by hlonipha
I manually set it to lower, managed to lower to +225 @ 4.14GHz, or +200 @ 4.00, anything lower would mean 0x07F random BSOD's under full load. I had to boost the QPI PLL a bit too... Running blck of 142 with 1:14 ram ratio.
Well i have a very rotten luck, all the CPU's i've got were lemons. So i got that EX in order to forego the blck limitations (and i did well, it peaks at like 175) and still run fast ram and fast qpi. I'm glad for you that your CPU's were golden but i had no such luck...Quote:
Originally Posted by SAE
Actually i went from zero, and progressively increased clocks and tweaking settings to maintain stability.Quote:
Originally Posted by SAE
That VTT i got from underclocking CPU, then setting XMP profile on RAM to run it at its rated 2GHz, and reducing VTT down to the point i was getting BSOD's on LinX and such. Wont get stable under +200... Unless you have another solution for me in which case i'm interested.
XMP is recommended by Corsair in order to run their ram at rated speeds.. any other profile will result in no post or quick crashes. I know VTT can be dangerous that's why i was asking here, maybe someone else with the same ram could tell what their settings end up to be with those Corsair modules.Quote:
Originally Posted by SAE
I tried lowering it but even +200 will lead to crashes eventually.
I had 4 D0 CPUs on EVGA boards (X58 SLI and Classified E759) and none needed more than +200 for 4GHz Uncore. You seem to know how to do your testing, but many DO NOT. ;)
BTW, Corsair is being conservative with their XMP profile to be sure those GT can be run at almost any system. I also own Corsair GT ram (2000 CL7) and it's the best ram i ever had! I could run 1.6Vdimm and 1.324Vtt @7-7-7-20 1T RTL 53-55-57 fully stable with my Xeon on water. It's a great cpu as is my 950, but I really cannot understand why most of the XE cpus need more voltage to run same speeds as good 920/W3520 cpus. :shrug:
That's the interesting point. My Vcore is not excessive, but i really have to get so much VTT (mine is the 2000 CL8 variant as per my sig), that's the weird part.Quote:
Originally Posted by SAE
Everything but ram is watercooled... I'm getting rather nice temps, although they peak fast around 4.3-4.4GHz where i have to run over 1.4V Vcore. Still glad about that 975, i reached my goal to get over 4GHz stable daily (with HT on and all bells and whistles), and to run 2GHz ram.
One interesting thing with the 975XE is the very different OCing, additional RAM multis, and the faster QPI (it has higher multis). I'm running the max multi on QPI, that may have some influence too, but with 142 blck it's not ending up at extreme speeds anyway... I'll post my complete settings this weekend to have some feedback (from other XE owners if possible) as this 975 seems to be a very different beast from the regular i7 breed...
(edit) oh RTL 53-55-57 is the correct setting for 2000MHz right ? i'll double check that
Nope, the RTLs are set automatically every bootup. They can even differ from boot to boot if you do not fix them. Iirc mine where at 55-57-59 or 55-58-60 or even at 56-58-60 when at 2000 7-7-7-20-1N. I manually set them to the lowest possible value for the first channel. You then got to set up the 2nd and 3rd channel either like: +2 +2 or like: +3 +2.
Hope it helps.
BTW, I think it's intended that XE clock differently from the lower "crap" ;)
But voltage-wise it's a miracle. Wonder how they pick them.
It helps yes ^^
I'll have to take a look at other detailed settings, getting over 4.3 (HT on still) is not that easy. I also run 2x GTX 285's in SLI in the same loop (with 2x140 and 3x120 rads) , and the motherboard of course, so it's getting hot fast above 1.4V...
SAE: read above this post I only use +150 for my 4.6 ghz. I was asking him if he is running at +350 or if he manually set it to lower and with that if he is stable. been clocking for awhile and do use that thing between the ears.Quote:
Originally Posted by SAE
I know how fast it can get to 85°C, then showing first effects of thermal throttling. A good combination of voltages can help you to keep temps lower. You might have hit a thermal limit for your chip/cooling though.Quote:
Originally Posted by gmat
Hehe. Don't mind, it was not directed to anyone, but to the general public. ;)Quote:
Originally Posted by hlonipha
Yeah the thermal limit for me seems to be around 4.4GHz.. I wouldnt mind running 4.4GHz 24/24 but i would need to find the sweet spot for voltages in order to prevent throttling. The challenge is, dual 285's also add a lot of heat when pushed, so i have to keep a good security margin on CPU temps. It's the first CPU i've ever had that reaches a thermal limit, for the 10 years i've been watercooling them !Quote:
Originally Posted by SAE
:D
You sometimes can decrease vcore finding the right vtt-vdimm balance. Then it might help upping the pwm frequencies to a value that works best, as vrms dump some heat into the loop as well. Set the PLL voltages (IOH and CPU) to 1.35V-1.5V. IOH voltage also has to be upped some for SLI and high (base)clocks. 1.2V is a good starting point, but do not increase it, when you don't need it. Minimizing voltages ain't an easy task, but it may pay off in the long run.
So far i've left the default 1.65V Vdimm.. it runs stable at current step (4.14) so it's fully ready for the WCG marathon week that begins tonight :D (Go xtremesystems !) I dont want to risk any unstability before it's over.
Any clue why the default CPU and IOH PLL are at 1.8V ? I'm still feeling uncomfortable decreasing them so much... Even 1.5V is a huge decrease from 1.8V...
I have IOH at 1.2V and ICH at 1.1V (as recommended for USB issues), i dont feel the need to touch them as even at full blast on graphics (furmark, crysis) it's completely stable. Only other voltage i pumped was QPI PLL it helped go over the 4GHz barrier.
Yeah, I also thought like that about decreasing them this much. But it just works great. I was priming full throttle for several hours with cpu pll @1.5V. IOH pll has also positiv or no negative effects on stability in my eyes.
http://www.abload.de/img/xeon4315-4110-10277-7-4jdx.jpg
24h prime x64 w/5.5GB RAM usage, QPI PLL didn't have to be increased for that.
http://www.abload.de/img/xeon4420-4210-1052_7-8a9bd.jpg
26h prime x64 w/5.6GB RAM usage, QPI PLL in synch with IOH
//edit.
another one with max primestable ram clocks cl7:
http://www.abload.de/img/xeon4441-4229-1057_7-8hlxd.jpg
Very interesting.
Is it a rule for getting full stability ? Is it based on empiric evidence (various overclockers finding out it works best that way) or more technical reasons ?Quote:
QPI PLL in synch with IOH
I think my QPI PLL is at 1.3V now (if i remember) but my IOH is 1.2V, would have to raise IOH to 1.3V then ?
It works best, just a rule of thumb and a consensus between ocers, I would say. It may work the other way as well, but for full stability in 2d and 3d I had to set equal values for both. The first screen settings worked well for prime an linx, but 3d seemed a bit unstable. I then noticed I had not increased qpi pll and catched up on setting it to IOH level. Worked! :D
/edit. What do you need 1.3V QPI PLL for?? I needed 1.3 - 1.35 for 235MHz bclk.
The 975 has a higher QPI than regular i7's (6.4GT/s vs 4.8GT/s) and i'm still running it at full multi with my slightly higher than stock bclk. And also running 1:14 ram multi (and XMP profile) seems to put a lot of stress on the ram controller. I had some weird memory corruption crashes when running it lower than that - i'll try to lower it again eventually as i added some Vcore too after a 0x0101 BSOD.Quote:
Originally Posted by SAE
Again it's a problem kinda specific to 975 XE, which makes it interesting in its own way :D