Core i7/X58 Overclocking Thread

[JWMc]

suicide everest memory bench below - I have no clue if this is right on no -

That is nuts Steve, you must try to catch Eva!

I wish we could see the trial-blocked values.

[GenTarkin]

So, heres a kicker...
The last couple days 24 hours these have been my WCG stable settings on all 8 "cores" : 21x175 (3.67ghz) @ 1.23v idle vcore, 1.335 QPI/VTT, Uncore @ 17x175 for clock of 2975mhz.

If I upped my BCLK one more mhz, my syste on both WCG and Linx/Prime blend would BSOD to the "an uncorrectable hardware error has occurred"

Since those stable settings, I decided to play with my uncore clock, I put it at a multi of 18x to give it a clock of 3150mhz. While at the same time I actually dropped QPI/VTT to 1.25v
All other settings listed above remain the same.

Its been 24hrs of WCG and no crash yet!!!

So, what do you all think? Uncore / memcontroller isnt my processors weakness or is it somehow? Whats preventing me to do 176mhz BCLK and higher with HT enabled. With HT off I can reach 178 BCLK stable...

Think its simply my 1.23v on vcore? I will be testing this sometime over the weekend, because I will be gettin my mugen 2 heatsink and that will allow me to comfortably raise my vcore higher.
So thats a theory I have ... what do you all think is my limit here?

[JWMc]

GenTarkin your main limiter is probably heat on the stock cooler if that's what you are still using until your Mugen arrives.

[GenTarkin]

GenTarkin your main limiter is probably heat on the stock cooler if that's what you are still using until your Mugen arrives.

Yes I am =) lol, thought it was concluded these things really didnt care about heat under 90C or something. My max temps were around 83-85C...still too hot?

[SteveRo]

That is nuts Steve, you must try to catch Eva!

I wish we could see the trial-blocked values.

eva has huge experience and knowledge gap over me.
I am a huge nubi when it comes to memory oc.
I seem to be having some success with max oc ht off and 2k memory (1 hr prime @ 4.43 ht off with mem at 2k, timings: 9-9-9).
Not so good with ht on - why is ht on always so much harder?
Best stable ht on (if you can call 1 hour prime stable) is 4.3 ht on with mem at 1720, timings 7-8-8 - not so good for 2k memory.

[JWMc]

Yes I am =) lol, thought it was concluded these things really didnt care about heat under 90C or something. My max temps were around 83-85C...still too hot?

Hmm, no that's not too hot. You are right: that is concluded. It seemed like 93-95C was just about too hot when I was testing at 4.2ghz with HT on. I finally figured out that sweet spot on the vCore that would keep it from overheating but just barely be enough to keep it from crashing.

If your temps are under that and you're brave then up vCore a bit. I think I saw someone run their chip to 100C on the stock cooler earlier in this thread without it dying or anything.

eva has huge experience and knowledge gap over me.
I am a huge nubi when it comes to memory oc.

Hehe, I guess I was just excited that someone else might have a system they can push great mem numbers on besides Eva. His results are simply unparalleled so far are they not? Maybe there's something to that DFI board. He's obviously a true expert but it also seems he knows how to get his hands on modules better than most of us will ever see but I'm just guessing.

I seem to be having some success with max oc ht off and 2k memory (1 hr prime @ 4.43 ht off with mem at 2k, timings: 9-9-9).
Not so good with ht on - why is ht on always so much harder?
Best stable ht on (if you can call 1 hour prime stable) is 4.3 ht on with mem at 1720, timings 7-8-8 - not so good for 2k memory.

HT on is rough because you are actually saturating the memory bus, esp as you overclock which increases the amount of memory bus utilization possible. The Uncore/QPI is well built to keep those 8 threads fed, these chips are just... .

[GenTarkin]

eva has huge experience and knowledge gap over me.
I am a huge nubi when it comes to memory oc.
I seem to be having some success with max oc ht off and 2k memory (1 hr prime @ 4.43 ht off with mem at 2k, timings: 9-9-9).
Not so good with ht on - why is ht on always so much harder?
Best stable ht on (if you can call 1 hour prime stable) is 4.3 ht on with mem at 1720, timings 7-8-8 - not so good for 2k memory.

I would take the 1720 =) with higher clocks on the cores (HT Enabled). These i7's are such effecient beasts. They really dont need huge memory clocks. Plus if you are willing to take the lower clocks, tighter memory timings may be better...which is just how you are running them =) Cuz...latency is pretty important as well. Even though these i7's dont really have an issue with latency either ROFL!!

[SteveRo]

I would take the 1720 =) with higher clocks on the cores (HT Enabled). These i7's are such effecient beasts. They really dont need huge memory clocks. Plus if you are willing to take the lower clocks, tighter memory timings may be better...which is just how you are running them =) Cuz...latency is pretty important as well. Even though these i7's dont really have an issue with latency either ROFL!!

Which should be faster - looks like I might get prime stable (both with ht on) -

#1. 31x138 for 4.28ghz with memory at 1932 @ 9-9-9-24
(just past 1 hour prime at the moment)

or

#2. 25x172 for 4.30 with memory at 1720 @ 7-8-8-21 (previously had stable for 1.5 hr prime)

1 hour prime is no not enough but it tells me I might be getting close to stable settings.
I suspect #2 is faster because cpu cycles are king but I will bench and verify.

[YukonTrooper]

I'm guessing #2 will yield the best real-world results. That memory bandwidth will only show in synthetics.

[Chicken Patty]

guys, I highly recommend for anybody using the EVGA X58 motherboard to upgrade to the latest BIOS. I myself still can't believe how great it is, improved my memory overclock amazingly and overall clock and stability of my system.

[eva2000]

eva has huge experience and knowledge gap over me.
I am a huge nubi when it comes to memory oc.
I seem to be having some success with max oc ht off and 2k memory (1 hr prime @ 4.43 ht off with mem at 2k, timings: 9-9-9).
Not so good with ht on - why is ht on always so much harder?
Best stable ht on (if you can call 1 hour prime stable) is 4.3 ht on with mem at 1720, timings 7-8-8 - not so good for 2k memory.

don't count ya self out... with time and more trial and error memory testing everyone should be able to get more out of their system - just be prepared for alot of bios/reboots between config testing

HT on = more heat and power output it seems. My i7 920 on TRUE120 air refuses to run 3d benchmarks stable past 4500mhz with HT ON. With HT Off i can do lighter cpu tests i.e. super pi 32m up to 4652mhz !

Which should be faster - looks like I might get prime stable (both with ht on) -

#1. 31x138 for 4.28ghz with memory at 1932 @ 9-9-9-24
(just past 1 hour prime at the moment)

or

#2. 25x172 for 4.30 with memory at 1720 @ 7-8-8-21 (previously had stable for 1.5 hr prime)

1 hour prime is no not enough but it tells me I might be getting close to stable settings.
I suspect #2 is faster because cpu cycles are king but I will bench and verify.

yeah #2 but for general 24/7 apps probably won't find as much difference

guys, I highly recommend for anybody using the EVGA X58 motherboard to upgrade to the latest BIOS. I myself still can't believe how great it is, improved my memory overclock amazingly and overall clock and stability of my system.

before and after examples for new bios ?

[rge]

Which should be faster - looks like I might get prime stable (both with ht on) -

#1. 31x138 for 4.28ghz with memory at 1932 @ 9-9-9-24
(just past 1 hour prime at the moment)

or

#2. 25x172 for 4.30 with memory at 1720 @ 7-8-8-21 (previously had stable for 1.5 hr prime)

1 hour prime is no not enough but it tells me I might be getting close to stable settings.
I suspect #2 is faster because cpu cycles are king but I will bench and verify.

I bet #2 not only faster, but you will be able to run reasonable 24/7 qpi/vtt volts. I think best thing about 2000 rated ram, is not only fun for benching, but also great for 1700-1800's speed range which will probably be the sweet spot for reasonable 24/7 qpi/vtt...unless you plan to be a guinea pig for us with vtt.

my 1600 rated ocz 8,8,8 3x1gb maxes out at 2020mhz 8,8,8,21 with 1.96v dram and 1.62 qpi/vtt (dont want to try higher volts)....but the high volts allows uncore 22x for 4400 mhz. Wish I had bought 2000 mhz to play with instead of 3x1 1600.

[286SX]

Updated my max OC HERE.

4085 [215 x 19] @ CPUv 1.3125; PLLv 1.8; QTIv 1.30. This is only LinX 20 tested, not Prime95. [Temps 98/97/93/91 - OUCH!]
4066 [214 X 19] @ CPUv 1.3125; PLLv 1.8; QTIv 1.30. This is only LinX 20 tested, not Prime95.
4047 [213 X 19] @ CPUv 1.31875; PLLv 1.8; QTIv 1.3. Also LinX 20 tested, not Prime95.
4020 [201 X 20] @ CPUv 1.31875; PLLv A; QTIv 1.225. Prime Small 8 hours.

Looks like I am getting to the top of my cooler's ability. I was able to run 4.104Ghz (216 X 19), but temps got up to 99 in LinX, so I shut it off.

**NOTE** interestingly, in prime blend, temps are 10 degrees cooler than LinX. I may be able to get a higher prime blend stable OC than LinX simply due to temps...

Does that still count??

-td

[GenTarkin]

Well, I have my Mugen 2 installed and I LOVE this thing!
Same OC so far as previous which was 3.67ghz @ 1.23vcore
Temps dropped from a max (linx and WCG) of 83-85C all the way down to max of 63C!!!

Idle temps in the 30's now.

[anzial]

try 4ghz @ 1.33vcore If possible, at 191bclk and 21 multi

[Chicken Patty]

before and after examples for new bios ?

well before on the old BIOS i needed 1.341v to be stable at 4.0 GHz. Now it only takes me 1.305v.


The biggest difference was that for some reason I had a really big problem running the RAM divider at 2:8. I had to run it at 2:6. However, with the new BIOS I can easily run the RAM at 2:8 and it is rock stable. THats the biggest difference i've known, and this BIOS definitely seems more stable.

[Chicken Patty]

a shot of my stable overclock for now, still havent tried to push higher.

[YukonTrooper]

a shot of my stable overclock for now, still havent tried to push higher.

Have you run any stability programs?

[GenTarkin]

try 4ghz @ 1.33vcore If possible, at 191bclk and 21 multi

I am testing 3.8ghz @ 1.27v, 1.255v QPI/VTT HT ON
I had to lower my Uncore to 17x multi ... 18x it BSOD'd with the well known "uncorrectable hardware error detected" in linx
Ram is @ 8x 1.6v 7 - 7 - 7 - 21

Little nervous to try higher, may set it to WCG for tonight see if there were any issues come morning..

Max temps went to 70C at this voltage and clockspeed

[eva2000]

I am testing 3.8ghz @ 1.27v, 1.255v QPI/VTT HT ON
I had to lower my Uncore to 17x multi ... 18x it BSOD'd with the well known "uncorrectable hardware error detected" in linx
Ram is @ 8x 1.6v 7 - 7 - 7 - 21

Little nervous to try higher, may set it to WCG for tonight see if there were any issues come morning..

Max temps went to 70C at this voltage and clockspeed

Not familar with other x58 boards other than DFI, but are you folks always running uncore frequency higher than recommended 2x dram frequency ? That could be source of instability tried at 8x mem and 16x uncore ?

[anzial]

I don't even have an option to change uncore frequency (Biostar original bios), I assume the board automatically sets it to twice the memory.. any way to check that in windows? via CPU-Z or something

[eva2000]

Download cpu tweaker (read only) http://www.xtremesystems.org/forums/...ad.php?t=92190 and http://i4memory.com/f79/memset-v4-0-...-timings-2868/

example

[anzial]

Thanks for the link, eva, I know now that uncore is 16x, QPI is 18x, and memory is 4x which I assume is 2:8..

[SteveRo]

don't count ya self out... with time and more trial and error memory testing everyone should be able to get more out of their system - just be prepared for alot of bios/reboots between config testing

HT on = more heat and power output it seems. My i7 920 on TRUE120 air refuses to run 3d benchmarks stable past 4500mhz with HT ON. With HT Off i can do lighter cpu tests i.e. super pi 32m up to 4652mhz !


yeah #2 but for general 24/7 apps probably won't find as much difference
before and after examples for new bios ?

I bet #2 not only faster, but you will be able to run reasonable 24/7 qpi/vtt volts. I think best thing about 2000 rated ram, is not only fun for benching, but also great for 1700-1800's speed range which will probably be the sweet spot for reasonable 24/7 qpi/vtt...unless you plan to be a guinea pig for us with vtt.

Max CPU OC versus Max Memory OC, also - hyperthreading on versus hyperthreading off -

Good morning,

So, trying to figure out a couple things.

1. Is cpu cycles still king?
More specifically - are you better off to give up a little cpu oc to get a significant increase in memory speed?

2. Does the increase in cpu oc allowed - due to turning off hyperthreading - overcome the factor-of-two reduction in threads (8 to 4)? (probably not but worth checking anyway).

A caveat or two – this is just one example by one guy that just knows enough to be dangerous.
Speaking of dangerous – I use very high voltages for this and I am also water cooling my cpu – so be careful trying this.
I know there are many other benchmarks that I could have used , I picked these as they were easy (and free).

So looking at the table below – some observations.
1. Lowest Super Pi time is with the highest cpu OC - no real surprise there.
2. Everest read speed and latency improves with increased DDR3 memory speed - Super Pi performance improves with increased memory speed (compare config 3 to 1).
3. Comparing config 1 and 2 – increasing memory speed from DDR3 1720 to 1932 is not enough to overcome just a 22 mhz decrease in cpu speed (looks like cpu cycles are still king)
4. Turning hyperthreading off (reducing threads available from 8 to 4) allows for an increase in cpu oc of about 133 mhz.
This will increase performance in single thread (and up to 4 thread applications) but it is not enough to overcome the reduction of 8 to 4 threads for an application that can take advantage of more than 4 threads - as it appears is the case with the Cinebench multi cpu render test.

Please correct anything I got wrong and make further observations as they occur to you, thanks!

[eva2000]

SteveRo, for testing use super pi 32M, 1M is too small a test to test system memory bandwidth and processor