Intel Xeon L5639 Hexa(6) Core LGA1366 Price:US $79.80 Used Ebay Seller

[Rock&Roll]

I've been watching this thread for a long time now. And I've been debating which one of these I should pick up.

I've got my old main rig with a X58 classified motherboard and a 920 that was overheated too many times, causing the IMC to start doing funny stuff.............I think. I just know that CPU was heavily abused and I was getting a lot of data corruption. 48hrs of memtest never turned up anything at all....

My main rig is currently an i5 based rig and I really miss having the virtualization support of an i7/xeon. So I'm thinking of selling my i5 rig, and rebuilding a couple x58 rigs using these cheap xeons (one for a server/dc and one for overclocking and gaming).

• I've been out of the loop for a while. Would my x58 classified still be a premier overclocking board, or should i try to pick up a better x58 board?

• 
  What's good memory for x58 these days? I've got 32GB of CL8 "Ballistic Tactical" ready to go in something.

• 
  Is there a major upside for going for a 5690 vs a 5670 or lower? IME, it's always just as good to buy the cheaper CPU for overclocking.

Thanks for keeping this great thread alive. It's pretty cool watching the 1366 cpu's still kicking ass all these years later.

 

[primetime]

I've been watching this thread for a long time now. And I've been debating which one of these I should pick up.

I've got my old main rig with a X58 classified motherboard and a 920 that was overheated too many times, causing the IMC to start doing funny stuff.............I think. I just know that CPU was heavily abused and I was getting a lot of data corruption. 48hrs of memtest never turned up anything at all....

My main rig is currently an i5 based rig and I really miss having the virtualization support of an i7/xeon. So I'm thinking of selling my i5 rig, and rebuilding a couple x58 rigs using these cheap xeons (one for a server/dc and one for overclocking and gaming).

• I've been out of the loop for a while. Would my x58 classified still be a premier overclocking board, or should i try to pick up a better x58 board?

• 
  What's good memory for x58 these days? I've got 32GB of CL8 "Ballistic Tactical" ready to go in something.

• 
  Is there a major upside for going for a 5690 vs a 5670 or lower? IME, it's always just as good to buy the cheaper CPU for overclocking.

Thanks for keeping this great thread alive. It's pretty cool watching the 1366 cpu's still kicking ass all these years later.

depends on the price really.....personally i think a 5660-5670 will result in same clocks as 5690.....I wouldn't spend no more than 200 at most regardless of the chip....it just seems the asus rampage series have the least trouble and the highest clocks over then rest of us...some of the evga boards don't even work with xeons(without mods) at all....you need to research this first......i don't think any of the higher priced xeons able to achieve more than 4600mhz(4400mhz for me) tops at most.......maybe with the rampage this is beatable otherwise i kinda doubt it

If you can grab whatever for 100 or so or at least a 5670 you be fine....for now iIm on a misssion be reach 4500mhz....but countless hours and failures aren't looking god...but i keep on trying

 

[Rock&Roll]

Thanks for the info. I might keep my eyes out for a used rampage board.

 

[MN Scout]

Still running well here with the X5670 at 4GHz. Normal temps while surfing the internet and listening to music are 25-30C across the 6 cores. The CPU cycles itself between 2200 to 4000 MHz. Haven't noticed any problems.

I paid $240 from esisoinc on ebay (came well packed). That is the price of having to have it now. Has been worth it. Feels like a completely new computer.

 

[primetime]

hated to do it...but i dropped back to 4200mhz.(21x)..while a lot of suggestions helped it was the 1% failure that bugged me. I think those who saved money and bought chips that allow 21x multis were the smart ones

 

[MN Scout]

Well primetime your tried hard. I think you did good in dropping it down. Errors are not welcome.

I did a little research before I bought the X5670. The L5639 the namesake of this thread looked tempting, but wasn't getting as high of clocks (I don't even remember the other reasons). Higher Xeons than the X5670 were going for even more money than what I paid.

The X5670 just seemed like a good middle ground, and it is performing 2X as good as my poor clocking I7920 did.

 

[primetime]

well did an image restore....back at 4400mhz....well see if got it write this time with bios settings...gonna do back to back stability tests, hyper pi, i burn test max, and occt fro 1 hr...if it passes all good...its good enough for me ....so we shall see...any other test i should add for testing ...prime as well i guess...i do a screen shot if al goes as hoped

 

[primetime]

work in progress.but i think i finally have the bios settings down correct:4400mhz is stays

[/URL][/IMG]

edit:it 99% stable at 4400mhz but thats not good enough.......will keep it at 4200mhz at least until winter......possibly i might shoot for the extra mhz then but probably just stay at is
At 4200mhz it seems bullet proof as seen here:

this was under extreme setting as well as high priority.....had to wait till bed time cause you cant use it in that state

[/URL][/IMG]

will run another all nighter using the software mentioned here per Zoson

You can get the latest linpack binaries here:
https://software.intel.com/en-us/art...npack-download

And you can get the updated LinX frontend here:
https://github.com/sanekgusev/LinX-o...ases/tag/0.6.5

 

[SkOrPn]

Has anyone here ever determined if 2x uncore is safe for the X56xx series xeons? Or is it best to keep it at 1.5x? I'm running 200x20 on theX5650 on a R3E and wondering if uncore should be higher or not. Memory is at 1603 Mhz, and current uncore is at 2405.6 mhz.

Should I go to 3200 or Auto, or keep it as-is?

 

[primetime]

Has anyone here ever determined if 2x uncore is safe for the X56xx series xeons? Or is it best to keep it at 1.5x? I'm running 200x20 on theX5650 on a R3E and wondering if uncore should be higher or not. Memory is at 1603 Mhz, and current uncore is at 2405.6 mhz.

Should I go to 3200 or Auto, or keep it as-is?

i have almost always kept it at auto with is 2......but i did find i was able to pass quite a few test like 8/10 at 4400mhz stable with it at 1.5.....but at the same time the benchmarks seemed to reflect a much lower clock....so yea its perfecty safe....and at 1.5 you very well could get a couple hundred mhz......but in the end i put it back to 2 and just dropped the overclock to 4200mhz....where its passes 100% of any test i have thrown at it so far....

so yea i think lowering it to 1.5 helps you get higher overclock that scores the same or lower than lower overclock at 2x .......i got higher scores at 4200mhz with it at 2x than i did with it at 4400mhz at 1.5.....so to me it made sense to keep it at 2x (auto)

but......could be different on your motherboard

 

[Zoson]

Holy shit, please actually read what I said about uncore.
It is NOT SAFE to run at 2x uncore if you have to exceed 1.35v QPI.
PERIOD END OF STORY.
YOU WILL FRY YOUR CHIP.

Secondly, Auto *SHOULD* set 1.5 NOT 2. Your board is fucking up if it's setting 2 at auto. I tried to say this nicely before but it apparently wasn't clear.

 

[SkOrPn]

Hello Zoson,

I want to say thanks for pointing me in the right direction for getting a xeon for this mobo, it has been the best purchase I have ever made on this R3E platform. I never realized there would be such a big difference in how the system feels. Even at default this X5650 feels every bit as fast as my 930 did overclocked.

THANK YOU again man...

 

[primetime]

Holy shit, please actually read what I said about uncore.
It is NOT SAFE to run at 2x uncore if you have to exceed 1.35v QPI.
PERIOD END OF STORY.
YOU WILL FRY YOUR CHIP.

Secondly, Auto *SHOULD* set 1.5 NOT 2. Your board is fucking up if it's setting 2 at auto. I tried to say this nicely before but it apparently wasn't clear.

There's no denying you have a lot of knowledge about this.....no arguing there...he asked if it was safe to run at 2x.......my motherboard defaults to 2x.......you say thats incorrect.....apparently you know more than the asus engineers who wrote the bios for my board

I NEVER told anyone to run the voltages im using (that risk is mine alone)....I simply said that 2x is/has been running fine on mine...IF and when my chip FRYS because of the voltage im running...I'll be sure to tell the world Zoson was right!!!!!!!!!!lol...you have to remember the 920s we all bought years ago and overclocked where 325 on sale

these days a 5640 can be got for 90 bucks.....I know you said you personally burned up a couple of 1500 dollar chips.....learning the hard way....i get it....if i couldn't afford 90 bucks then i wouldn't be using the settings im using (nor should anyone else)...end of story

 

[SkOrPn]

I don't plan on going over 1.3v (actual in bios is 1.29v), but like our cpu, bus etc, we are trying to get more out of it, so if I can "safely" move up from 2405 mhz uncore to a max of 2x (3200 mhz), I would obviously want to do that. If it BSOD's, or fails IBT half way there, then I just discovered my max uncore speed. RIGHT? Can I run 1.6x, 1.7x, 1.8x etc etc? lol, there are quite a few steps between 2400 mhz and 3200 mhz.

All I am wanting to know is if 2405 mhz (1.5x) is absolute must for safety if I am not planning on going higher than 1.3v? What Zoson has said seems to suggest exactly this but he still has not directly answered this question with a yes it is safe so long you stay below 1.35v. I see people suggesting 1.5x is an absolute must no matter what the voltage is, and that seems wrong to me. I have read half a dozen threads and articles on the subject and I still have not seen anyone specifically say what the highest safe uncore is on Gulftown. For example, would 3200 mhz (2x) be perfectly safe if it runs at 1.3v? YES, if you are at, or over 1.35v, then you are in very dangerous territory as is evident by the many dead CPU's. But I'm not ever going to venture into that territory, lol... I just want the highest perfectly safe uncore possible.

BTW, I am running 200x20 on my X5650 (R3E platform) and my 3x4gb memory is 1603 mhz at 1.6v. I'm not interested in a higher OC on the CPU. I just want more uncore if safely possible. Thank You guys for a wonderful information filled thread. Great read...

 

[primetime]

I don't plan on going over 1.3v (actual in bios is 1.29v), but like our cpu, bus etc, we are trying to get more out of it, so if I can "safely" move up from 2405 mhz uncore to a max of 2x (3200 mhz), I would obviously want to do that. If it BSOD's, or fails IBT half way there, then I just discovered my max uncore speed. RIGHT? Can I run 1.6x, 1.7x, 1.8x etc etc? lol, there are quite a few steps between 2400 mhz and 3200 mhz.

All I am wanting to know is if 2405 mhz (1.5x) is absolute must for safety if I am not planning on going higher than 1.3v? What Zoson has said seems to suggest exactly this but he still has not directly answered this question with a yes it is safe so long you stay below 1.35v. I see people suggesting 1.5x is an absolute must no matter what the voltage is, and that seems wrong to me. I have read half a dozen threads and articles on the subject and I still have not seen anyone specifically say what the highest safe uncore is on Gulftown. For example, would 3200 mhz (2x) be perfectly safe if it runs at 1.3v? YES, if you are at, or over 1.35v, then you are in very dangerous territory as is evident by the many dead CPU's. But I'm not ever going to venture into that territory, lol... I just want the highest perfectly safe uncore possible.

BTW, I am running 200x20 on my X5650 (R3E platform) and my 3x4gb memory is 1603 mhz at 1.6v. I'm not interested in a higher OC on the CPU. I just want more uncore if safely possible. Thank You guys for a wonderful information filled thread. Great read...

first of all....please be more clear about which voltage your referring to...my qpi.dram voltage is 1.4.....is this the voltage your referring to? Zoson claims a couple of bios settings on my asus p6tdelux are incorrectly labeled by asus....how he knows this god only knows.....its in my opinion every motherboard is different when it comes to these settings...but thats just my opionin

as Zonon's claims a couple of the settings in my board are incorrectly labeled (which makes discussing different voltage settings practically useless) I can only go by what it is actually labeled in my bios...i have posted pictures of my bios settings and we still get cross wired on what voltage is what lol....with this being said then theses no way anyone can say what is actually what? you get where im going with this?

if you actually own an asus p6t delux v1 with same bios im using....then im a perfect genie pig to see if something burns up or not....I at this time can afford the risk....

Im not knocking Zoson one bit....but how he actually knows certain bios settings on my particular board are incorrectly labeled....but it makes discussing voltages.....how do i put it...hes talking about one setting and im talking about what is actually labeled in my bios
and we end up on different pages......so its useless

all i can say is good luck....is you cant afford to replace it don't overclock it...and what works great on mine might not work same on yours....there different motherboards...different bios revisions....Apparently some of the settings are incorrectly labeled by Asus engineers....at least on mine according to Zoson

I am practically running the exact same settings i have been using since i bought this board......now if you see me posting from my laptop tomorrow you can say i told you so lol
keep in mind...i don't fold at home 24/7....my temps appear very good....and it passes all test i have run on it....no weird issues....no software corruption that i know of (but i keep image backups just in case)

tldr: know one can tell you whats absolutely safe with overclocking.....there is always a risk..... This is not Soft OCP lol thats down the street

 

[SkOrPn]

tldr: know one can tell you whats absolutely safe with overclocking.....there is always a risk..... This is not Soft OCP lol thats down the street

LOL, I agree. However, there is risk just turning on the computer, or your TV, or a lamp. I had a Abit BP6 overclocked to 600mhz on both Pentiums and when I came home after work (in the year 2000 or so) there was a funny smell in the air and I noticed Seti was no longer churning away on the screen and the computer was off. I opened the case up only to find the mobo had caught fire and destroyed "nearly" the entire inside of the case. Clearly it was not safe for that system to be overclocked or even on for that matter, lol. The only thing I could figure out was that maybe some small animal or large group of bugs had made a home on the system board, died and then provided fuel for a small fire when a cap exploded. Otherwise a giant black scorched section in the center of the mobo made no sense to me.

Anyway, I have ran my 930 at 4ghz for 4 years, and 2x uncore (3200 in this case). I am hoping that every bios setting can be a "improvement", either with lower V, or with lower latency, or higher clocks. Moving down a frequency within a setting that goes from 2400 mhz to nearly 9000 is a very strange thing to do on a overclocking platform. To me that is underclocking something and seems to suggest to me anyway that it was only provided as a means to try and push more ram or cpu clocks out of it. And so then, if you're not planning on extreme overclocks then up to 2x would be OK if voltage is reasonable.

Further more, Intel tells us that safe voltage on their chips are 1.35v. Memory manufacturers tell us their max V as well, etc etc, so there must be an engineer who knows what the actual max V setting should be for uncore (QPI/DRAM in Asus boards), and not this guesstimate everyone is throwing around. Zoson says its 1.35v, and many of the guides say its 1.4v. Surely this could not have been overlooked by both Intel and then every motherboard manufacturer on the planet. Uncore in my bios is represented by a 4 digit Frequency number (from 2400 to something like 8000+), thus I want to OC it until I reach a point it is unsafe to run (just like we do with everything else), or the computer fails and then back down. I feel that point for my CPU is 4ghz, but that's just me (I know its actually higher, but I see no benefit in doing so). I also feel that my max bclk that I am comfortable with is 200. So, again there must also be the same concept in play when talking about the uncore multiplier. Its obvious that the difference in performance from 1.5x to 2x is huge, so this begs the question, why accept the lowest setting if it increases performance when raising it higher? That is not what overclockers generally tend to do, and I do not target specific parts as the only items I want to overclock. EVERYTHING to me is overclockable to some degree, even my PCIe, which I always set to 101... Why? Because I can.

Here's another question that may be easier to answer from someone who is an expert overclocker, because I am surely not. Is UNCORE really that mysterious after all these years having dealing with it?

One last thing, I just read several Gulftown overclocking guides and one of them begins with, "You can go as low as 1.5x uncore with Gulftowns" and then later in the guide he tells you "You must set the Gulftown uncore to 1.5x". There is no such thing as you "can" and or you "must" in the same context. It has to be one or the other, because "can go as low" suggests it is not an absolute requirement. lol

I will just up uncore and keep voltage at or below 1.3v and then do some tests and if it dies I will buy another cpu and try again. By the way, I just spoke to someone who has been running their Gulftown Xeon at 1.35v QPI and 2x Uncore non stop for four years, and he still has not seen anything go wrong with it, and his board isn't quite the overclocking platform mine is. Kinda gives me some hope... lol

By the way, I am not trying to start an argument, only trying to weed out all pertinent truths on the subject of uncore. Zoson is the guy responsible for me upgrading to a Xeon and I thank him for it. Thank you to primetime for providing your experience in this thread. You guys rock!

 

[Zoson]

OK, lets try to clear all this up in one post. Remember, I have an asus board and it has all of your bios adjustments and more, with better control. I can go below 1.8v PLL down to 1.20v PLL, for example, and I've had it for several years with a lot of different CPUs! You can see a lot of that here in my post history on hardforum if you really want to verify. That said... PREPARE YOURSELF FOR A SHORT HISTORY LESSON!

The first generation intel i7 architecture is divided into two node categories: 45nm and 32nm.
The 45nm cpus were the first on the market known as "Bloomfield" consisting of the i7 920 up to the i7 975x and 4 cores with 8 threads, with the i7 980x and i7 970 coming later on the 32nm node "Gulftown" process with 6 cores and 12 threads. Essentially the same architecture, but with a transistor node shrink, the corresponding changes, and additional features. What you have, all of these 6 core xeons, are "Gulftown" cpus with a second QPI link bolted on as yet another feature for SMP operation.

Another one of those features was lowering the previous REQUIREMENT of 2x uncore to dram speed down to 1.5x(so for ddr3 1600mhz you could run 2400mhz uncore instead of 3200mhz uncore). The reason was because the architecture of the integrated memory controller(part of the uncore and forward:IMC) could not sustain 2x uncore at full memory loads at below the voltage spec! This was a HUGE problem for people wanting to run >12GB loads on the 45nm i7 CPUs! With my i7 930 I remember having to give the IMC 1.58175v just to get my 3x2GB and 3x 4GB modules running together at 1600MHz CAS 8, and the spec was 1.45 on that cpu!!! As soon as I got my first 32nm i7 970, i could run that same exact kits at 1.275v as long as I left my uncore at 1.5x. It freed the memory speed from the uncore.

So there's the problem, two different generations running on the same x58 chipset that need different defaults and have different voltage requirements. Boards that were designed for 45nm chips are in the fray and will run the chips with a timing update to the bios. That doesn't necessarily change what the board will set its defaults to. Nor does it make certain hardware able to go below what it was designed to do... That's what it was designed to do after all... Power 45nm CPUs. That is one of the main reasons why there is the Rampage II Extreme and the Rampage III Extreme - BOTH x58 CHIPSET!

All of this is a result of the transistor node size shrink, which has the benefit of all the transistors using less power to run at a certain clockspeed. However, it also reduces their ability to tolerate higher voltages because it literally damages them through heat decay and electromigration.

Anyway, getting to the first gen i7 voltages:
First we have the 'core' of the cpu which includes the actual cpu cores and L1 and L1 caches, this is the 'core voltage' in bios and nobody disputes that. You can go up to 1.45v and be within intels 'spec' and with good cooling you can actually go quite a bit over this for short periods with no penalty.

Then we have the 'uncore' of the cpu, which is the L3 cache, IMC, QPI controller/interconnect, and a few other things. This voltage tolerance was reduced to 1.35v. If you take the time to read the reviews of the 980x you'll see several reviewers commenting that it was strange that intel had only reduced the uncore voltage limit from 1.45 to 1.35, and left the core voltage limit at 1.45v. It's also where we saw a lot of cpus die in the extreme overclocking threads. New gulftown owners that were experienced bloomfield overclockers frying their cpus because they weren't aware of the change and how it became a hard stop instead of a 'general guideline unless extreme cooling.' Even people with LN2 were reporting dead cpus because of high uncore voltages.

This is also where there is a bios confusion. Everyone except ASUS decided to call it 'uncore' while asus decided to call it 'qpi/dram'. ASUS said that it was more descriptive because for overclocking the user didn't care that it was inaccurate - it was a setting that would be used to increase qpi and dram performance. Pure marketing spin.

They ARE RIGHT though, that is EXACTLY what you use it for. You use it to obtain higher IMC clockspeeds, which as you saw really benefits your performance! Yes, now you have your speedy memory, but your IMC is holding it back. It's not fast enough. So you increase your uncore(QPI/DRAM) voltage to 1.35v and see how fast your IMC will go at that voltage. That may be 1.5x, that may be 2.2x, or it may be somewhere between at 1.7 or 1.8x. This all depends on the speed of your ram. If I remember correctly most IMC's could do 3200MHz at 1.35v

So again, you're now playing in a different court. With your old CPU it was perfectly fine to run up to 1.45v uncore(QPI/DRAM), but now it's not. Your uncore multiplier bottom limit was lowered to help assist with the problem of being unstable at 1.35v uncore(QPI/DRAM). You also have to, again, consider LLC and/or voltage droop. If you have LLC not fully enabled, you could be drooping to 1.35v under load from 1.4v idle, which is less bad for your cpu. Gulftown cores DO like LLC though. I really suggest turning it up to the max and setting your voltages to the max specs and seeings what your chip can do at 1.35v uncore(QPI/DRAM).

 

[SkOrPn]

Damn Zoson, where is the props button when you need it?

Like I said I am not going to use voltages anywhere near what your talking about. I just reduced QPI to 1.3v and upped uncore to 2606 (1.62x) and it ran a full 10 runs of IBT at very high stress. What was really exciting was the fact that at 2400 mhz uncore it was in the mid 50'ish Gflops, but at just one tick up it jumped to a staggering 71 Gflops. I call that staggering because it is like a 15 to 20% increase going from 1.5x to 1.62x. VERY difficult to not play around with it when you get results like that. lol

So, even though I am only playing around with 1600 mhz memory and not planning on running my chip over 4ghz, you still think it would be beneficial for me to run max LLC?

Also, from what you just said, it sounds like it is the voltage ONLY that is the IMC killer, and had nothing to do with the high uncore speeds. If I can run 3200 mhz uncore (or even just below it) and somehow drop my QPI into the 1.2v range, I would be a happy camper. My friend is running his at 3195 at only 1.2v QPI, on a x5650 running at 4.5ghz and he has 24GB of 1600 memory (6x4gb). Seems like I should easily do what he is doing since I have half his memory and much less overclock.

Thanks again Zoson, and thanks again for telling me about these xeons over at ES...

 

[primetime]

Zoson......let me ask you this.....with all the different boards, different bios revisions, especially since apparently Asus miss labeled bios settings, how do we even know that the voltage set is really what is set....you know people always saying don't trust what certain software is saying your power supply is producing...it can say one thing.....and be something a lot lower or higher...causing lots of problems

Also would there be any tell tell signs things are about to die? or basically it just doesn't post one day and thats it? is there any software that can double check the voltage in question? is it remotely possible that its not really as high as i have it set in bios? or is that just wishful thinking? guess i can drop my overclock to 4ghz and try with the lower voltage.....not like 200 mhz is really a bottle neck..

As always....thank you for your input.....

 

[SkOrPn]

Ok, I take back what I said about the large increase in Gflops on IBT. It was user error. I did test with IBT (high stress) using uncore from 2400, 2600, 2800, 3000 and 3200 and it scaled appropriately. It wasn't until I increased the stress level that the Gflops had a huge increase. I have decreased QPI voltage down to 1.225v (actual in bios shows 1.204, and HWMonitor shows something like 1.196), and I am running uncore at 3208 mhz (2x). This has got to be good, no?

Now what about QPI Link? lol

 

[primetime]

Ok, I take back what I said about the large increase in Gflops on IBT. It was user error. I did test with IBT (high stress) using uncore from 2400, 2600, 2800, 3000 and 3200 and it scaled appropriately. It wasn't until I increased the stress level that the Gflops had a huge increase. I have decreased QPI voltage down to 1.225v (actual in bios shows 1.204, and HWMonitor shows something like 1.196), and I am running uncore at 3208 mhz (2x). This has got to be good, no?

Now what about QPI Link? lol

for what its worth mine is running about 78gflops.....obviously if i were running under 2x or 3200mhz uclk tthen it be a lot lower.....now qpi link data rate i run as low as possible 7218mt/s.....i just always have

[/URL][/IMG]

edit: where exactly are you seeing qpi in HWMonitor?

[/URL][/IMG]

this is the same program your referring to right?

 

[SkOrPn]

now qpi link data rate i run as low as possible 7218mt/s.....i just always have

That is because the QPI bandwidth is already insanely fast, not even x79's DMI 2.0 can touch it. I just wondered if it would hurt anything going higher? Would my system blow up, or kill the cpu, or kill the board if QPI Link was higher, etc?

edit: where exactly are you seeing qpi in HWMonitor?

[/URL][/IMG]

this is the same program your referring to right?

Yeah its the same program, but you're using the pro version.

The temps are low and the voltages are low on everything that matters and it still passed 12 hours of P95 and 10 passes of high stress on IBT. Look at the uncore speed in CPU-Z (NB freq of 3207.3). Is this something I can run as a daily for the next several years with the volts at 1.2v or less, and not worry about it?

 

[Zoson]

So it looks like the P6T does not have a QPI sensor that HWMonitor can read. Both Skorpn and I have the R3E, so we get a lot of 'enthusiast' sensors not available in the workstation boards.

As far as knowing exactly how well a motherboard reads the voltages... The R3E is calibrated very well. It also has those exposed voltage points to verify the voltage levels with a digital multimeter. On the R3E I found that most of the voltages were within .01v of the readout I got on the DMM. So when I was setting 1.35v uncore in bios with full LLC I was getting between 1.34v and 1.36v real.

Also correct, you can run ANY uncore speed without killing your cpu as long as you stay at/below 1.35v QPI.

The uncore figure I gave 'could run 3200MHz' was assuming a 6x4GB memory load at 1600MHz. Memory with tight timings will also affect this. For example, a 24GB 1600MHz CAS 9 kit may allow you to run 3200MHz uncore, while a 24GB 1600MHz CAS 7 kit may not.

As far as your IMC dying, you'll start getting BSOD's with stop code 101, 124, or 104 and be forced to give the IMC even more voltage(thus causing it to die even faster) to get stability back.

And yeah Skorpn, nice low volt OC there for sure.

 

[SkOrPn]

Thank you Zoson for the confirmation on the safety of the IMC below 1.35v.

Now, about the solder testing pads on the upper right of my board. The very top one say "GND", is this the one I put the black lead from my fluke meter on and then obviously the red one to the pad I am interested in testing? Having the HWMonitor, BIOS and TurboV all read differently makes me want to know the truth, lol...

Also, I plan on tightening my timings a bit here soon, and will probably bring the memory back up to 1.6v (R3E default). I just set it to 1.5v because that is what G.Skill advertises it as, but since I would like to tighten up the timings and try and push at least 9-9-9-24-1N@[email protected], I think I need the extra tenth of a volt. So, this "could" possibly make it harder to run uncore at 3200mhz? If so, what would be best to remedy that? Dropping uncore down a notch or two, or running slightly looser timings?

 

[Zoson]

Thank you Zoson for the confirmation on the safety of the IMC below 1.35v.

Now, about the solder testing pads on the upper right of my board. The very top one say "GND", is this the one I put the black lead from my fluke meter on and then obviously the red one to the pad I am interested in testing? Having the HWMonitor, BIOS and TurboV all read differently makes me want to know the truth, lol...

Correct. black to GND and red to one of the marked solder dots.

Also, I plan on tightening my timings a bit here soon, and will probably bring the memory back up to 1.6v (R3E default). I just set it to 1.5v because that is what G.Skill advertises it as, but since I would like to tighten up the timings and try and push at least 9-9-9-24-1N@[email protected], I think I need the extra tenth of a volt. So, this "could" possibly make it harder to run uncore at 3200mhz? If so, what would be best to remedy that? Dropping uncore down a notch or two, or running slightly looser timings?

You're good to run your vDDR up to 1.65v without damaging your IMC. And yeah if you're at 3200MHz uncore and lose stability after tightening timings you either need to up ram voltage(if the problem is the ram) or up uncore voltage(if the problem is your IMC)... **OR** if you can't increase uncore voltage any more because you're at 1.35v already, then reduce uncore speed.

 

[SkOrPn]

You're good to run your vDDR up to 1.65v without damaging your IMC. And yeah if you're at 3200MHz uncore and lose stability after tightening timings you either need to up ram voltage(if the problem is the ram) or up uncore voltage(if the problem is your IMC)... **OR** if you can't increase uncore voltage any more because you're at 1.35v already, then reduce uncore speed.

Ok, thanks for that. I think I want to maintain the lower voltages over anything. I like to overclock, but like you said earlier in this thread, a 4ghz OC is only the top tier of a moderate overclock and not considered extreme overclocking (which is what my board was originally intended for, extreme, lol). Which is why I limit my bclk to 200, when it will go higher and cpu to 4ghz when it also will go higher. The memory speed of 1600 is also considered an overclock on this platform, so I run the sticks at their default because as far as my CPU is concerned it is overclocked. So, if I lose stability with tighter timings I will probably either drop uncore or loosen the timings, which ever gives me back stability at the same voltage. Also, 1.65v is getting near the threshold of the 0.5v requirement difference, right? If QPI is at 1.2 or just below, than anything over 1.65v is starting to flirt with the half volt spec? How does that play into it? Is that another dangerous setting? What if I drop the QPI and CPU to a solid 1.1v? Would 1.65v on the RAM be dangerous?

 

[primetime]

Ok, thanks for that. I think I want to maintain the lower voltages over anything. I like to overclock, but like you said earlier in this thread, a 4ghz OC is only the top tier of a moderate overclock and not considered extreme overclocking (which is what my board was originally intended for, extreme, lol). Which is why I limit my bclk to 200, when it will go higher and cpu to 4ghz when it also will go higher. The memory speed of 1600 is also considered an overclock on this platform, so I run the sticks at their default because as far as my CPU is concerned it is overclocked. So, if I lose stability with tighter timings I will probably either drop uncore or loosen the timings, which ever gives me back stability at the same voltage. Also, 1.65v is getting near the threshold of the 0.5v requirement difference, right? If QPI is at 1.2 or just below, than anything over 1.65v is starting to flirt with the half volt spec? How does that play into it? Is that another dangerous setting? What if I drop the QPI and CPU to a solid 1.1v? Would 1.65v on the RAM be dangerous?

its funny you say that...my board has to be set at 1.66 on the ram to be stable...cause the next lowest setting is 1.64....and while it never bsod's it will actually fail the stress test unless its set at 1.66.....which is actually labeled in the bios as dangerous...but my ram needs it lolalways has....its the only setting i actually run in the red

 

[SkOrPn]

its funny you say that...my board has to be set at 1.66 on the ram to be stable...cause the next lowest setting is 1.64....and while it never bsod's it will actually fail the stress test unless its set at 1.66.....which is actually labeled in the bios as dangerous...but my ram needs it lolalways has....its the only setting i actually run in the red

Hmm, yeah when I set 1.65v I instantly got a RED Warning popup in my BIOS as well. However, it only took me a few seconds to be scared into submission and back to 1.6v, lmao... I never did try 1.61-1.64, I guess being so close to a red warning label actually worked on this old guy.

 

[primetime]

if i were risking a 1500 dollar chip...things be a lot different....honestly i wouldn't even be chancing it...but these days its a lot cheaper if disaster happens

 

[schmuckley]

Ohai!

 

[SkOrPn]

if i were risking a 1500 dollar chip...things be a lot different....honestly i wouldn't even be chancing it...but these days its a lot cheaper if disaster happens

Yeah, and not to mention we both probably have spare chips lying around, I know I have my 930 to go back to if I must. But I don't want to, lol.

Plus, its less than a hundred for some xeons these days, including the one I'm using... I still can't seem to get over that.

 

[Tiporaro]

OK, lets try to clear all this up in one post. Remember, I have an asus board and it has all of your bios adjustments and more, with better control. I can go below 1.8v PLL down to 1.20v PLL, for example, and I've had it for several years with a lot of different CPUs! You can see a lot of that here in my post history on hardforum if you really want to verify. That said... PREPARE YOURSELF FOR A SHORT HISTORY LESSON!

The first generation intel i7 architecture is divided into two node categories: 45nm and 32nm.
The 45nm cpus were the first on the market known as "Bloomfield" consisting of the i7 920 up to the i7 975x and 4 cores with 8 threads, with the i7 980x and i7 970 coming later on the 32nm node "Gulftown" process with 6 cores and 12 threads. Essentially the same architecture, but with a transistor node shrink, the corresponding changes, and additional features. What you have, all of these 6 core xeons, are "Gulftown" cpus with a second QPI link bolted on as yet another feature for SMP operation.

Another one of those features was lowering the previous REQUIREMENT of 2x uncore to dram speed down to 1.5x(so for ddr3 1600mhz you could run 2400mhz uncore instead of 3200mhz uncore). The reason was because the architecture of the integrated memory controller(part of the uncore and forward:IMC) could not sustain 2x uncore at full memory loads at below the voltage spec! This was a HUGE problem for people wanting to run >12GB loads on the 45nm i7 CPUs! With my i7 930 I remember having to give the IMC 1.58175v just to get my 3x2GB and 3x 4GB modules running together at 1600MHz CAS 8, and the spec was 1.45 on that cpu!!! As soon as I got my first 32nm i7 970, i could run that same exact kits at 1.275v as long as I left my uncore at 1.5x. It freed the memory speed from the uncore.

So there's the problem, two different generations running on the same x58 chipset that need different defaults and have different voltage requirements. Boards that were designed for 45nm chips are in the fray and will run the chips with a timing update to the bios. That doesn't necessarily change what the board will set its defaults to. Nor does it make certain hardware able to go below what it was designed to do... That's what it was designed to do after all... Power 45nm CPUs. That is one of the main reasons why there is the Rampage II Extreme and the Rampage III Extreme - BOTH x58 CHIPSET!

All of this is a result of the transistor node size shrink, which has the benefit of all the transistors using less power to run at a certain clockspeed. However, it also reduces their ability to tolerate higher voltages because it literally damages them through heat decay and electromigration.

Anyway, getting to the first gen i7 voltages:
First we have the 'core' of the cpu which includes the actual cpu cores and L1 and L1 caches, this is the 'core voltage' in bios and nobody disputes that. You can go up to 1.45v and be within intels 'spec' and with good cooling you can actually go quite a bit over this for short periods with no penalty.

Then we have the 'uncore' of the cpu, which is the L3 cache, IMC, QPI controller/interconnect, and a few other things. This voltage tolerance was reduced to 1.35v. If you take the time to read the reviews of the 980x you'll see several reviewers commenting that it was strange that intel had only reduced the uncore voltage limit from 1.45 to 1.35, and left the core voltage limit at 1.45v. It's also where we saw a lot of cpus die in the extreme overclocking threads. New gulftown owners that were experienced bloomfield overclockers frying their cpus because they weren't aware of the change and how it became a hard stop instead of a 'general guideline unless extreme cooling.' Even people with LN2 were reporting dead cpus because of high uncore voltages.

This is also where there is a bios confusion. Everyone except ASUS decided to call it 'uncore' while asus decided to call it 'qpi/dram'. ASUS said that it was more descriptive because for overclocking the user didn't care that it was inaccurate - it was a setting that would be used to increase qpi and dram performance. Pure marketing spin.

They ARE RIGHT though, that is EXACTLY what you use it for. You use it to obtain higher IMC clockspeeds, which as you saw really benefits your performance! Yes, now you have your speedy memory, but your IMC is holding it back. It's not fast enough. So you increase your uncore(QPI/DRAM) voltage to 1.35v and see how fast your IMC will go at that voltage. That may be 1.5x, that may be 2.2x, or it may be somewhere between at 1.7 or 1.8x. This all depends on the speed of your ram. If I remember correctly most IMC's could do 3200MHz at 1.35v

So again, you're now playing in a different court. With your old CPU it was perfectly fine to run up to 1.45v uncore(QPI/DRAM), but now it's not. Your uncore multiplier bottom limit was lowered to help assist with the problem of being unstable at 1.35v uncore(QPI/DRAM). You also have to, again, consider LLC and/or voltage droop. If you have LLC not fully enabled, you could be drooping to 1.35v under load from 1.4v idle, which is less bad for your cpu. Gulftown cores DO like LLC though. I really suggest turning it up to the max and setting your voltages to the max specs and seeings what your chip can do at 1.35v uncore(QPI/DRAM).

Have been following this thread off and on as I've used some x5650s in a couple repurposed home servers - wanted to take the time to post simply because that was an excellent response by zoson and he deserves a salute.

 

[primetime]

Have been following this thread off and on as I've used some x5650s in a couple repurposed home servers - wanted to take the time to post simply because that was an excellent response by zoson and he deserves a salute.

I think we can ALL agree on that!

 

[Zoson]

Ok, thanks for that. I think I want to maintain the lower voltages over anything. I like to overclock, but like you said earlier in this thread, a 4ghz OC is only the top tier of a moderate overclock and not considered extreme overclocking (which is what my board was originally intended for, extreme, lol). Which is why I limit my bclk to 200, when it will go higher and cpu to 4ghz when it also will go higher. The memory speed of 1600 is also considered an overclock on this platform, so I run the sticks at their default because as far as my CPU is concerned it is overclocked. So, if I lose stability with tighter timings I will probably either drop uncore or loosen the timings, which ever gives me back stability at the same voltage. Also, 1.65v is getting near the threshold of the 0.5v requirement difference, right? If QPI is at 1.2 or just below, than anything over 1.65v is starting to flirt with the half volt spec? How does that play into it? Is that another dangerous setting? What if I drop the QPI and CPU to a solid 1.1v? Would 1.65v on the RAM be dangerous?

Running a lower BCLK with a higher multiplier is almost always in your favor. That's why I run only 155BCLK.

As far as the memory thing - yes - if you have 1.65v memory you must run 1.2v uncore or higher or you break the half volt difference rule. While you supply your IMC with a specific voltage to run at, the actual DRAM BUS voltage is determined by the vDDR voltage. I've seen people running 1.35v uncore using ram that's 1.8v without damaging anything other than the ram itself.

In any case, I'd be shooting for 4.2GHz core, 3.2GHz uncore ant 1600MHz CAS 8 if I had your equipment/cooling.

 

[SkOrPn]

The only multipliers I have above 20 are the turbo ones. I'm not sure I understand multipliers that don't stick once you set them. I only need a multi of 21 and at this voltage and cooling room it would make it easily. I'm just not convinced turbo is worth it.

I am going to try pushing 8-8-8-24-1N today at only 1.6v and see what happens. Thanks again

 

[Zoson]

You can't use C1E or SpeedStep if you want turbo mults to stick. So your cpu will constantly run at the overclocked speed instead of being able to downclock to save power.

 

[SkOrPn]

You can't use C1E or SpeedStep if you want turbo mults to stick. So your cpu will constantly run at the overclocked speed instead of being able to downclock to save power.

Well, well, that was the missing information I needed to understand the Turbo function. And to think I always have C1E and Speedstep disabled. So, I have more multis to play with that will stick? I just never tried using a multi that was not specified by ARK, and ARK states my top multi is 20. Hmm, lol I wonder what I am doing today?

The Godzilla movie in 3D or higher overclocks, hmmm....

 

[primetime]

ok.....figured i would try the lower voltage thing...had to down clock to 4000mhz.....no big deal.....passed both major stress tests no problem.....then it cant even complete hyperpi lol...WTF?

[/URL][/IMG]

ran stupid thing twice...just fail lol...but i know its works fine at 1.4 qpi voltage....hell even at 4400mhz....kinda a head scratcher ain't it? it passes the 2 hardest stress tests at 10 runs each but fails hyperpi? that makes a lot of sense

 

[SkOrPn]

Why is your gflops so low at that clock? At the same clocks and less voltage I am hitting near 70-71 Gflops. Ok, I think your chip needs to be pawned off for another, lol...

 

[primetime]

Why is your gflops so low at that clock? At the same clocks and less voltage I am hitting near 70-71 Gflops. Ok, I think your chip needs to be pawned off for another, lol...

you noticed that to lol.....i usually run around 78-80 Gflops......but you wish your temps were that low ...Anyways up the qpi back to 1.4....back to 4200mhz and it passes hyperfine....its as if my board needs the qpi votlage.....maybe has somthing to do with it being at 200blck

fuck it...as long as it stays cool and stable...im keeping it there