Madshrimps Forum Madness

Madshrimps Forum Madness (https://www.madshrimps.be/vbulletin/)
-   Hardware Overclocking and Case Modding (https://www.madshrimps.be/vbulletin/f10/)
-   -   OC, yes, but what about OD? (https://www.madshrimps.be/vbulletin/f10/oc-yes-but-what-about-od-68763/)

JimmyJump 23rd December 2009 20:21

OC, yes, but what about OD?
 
The OD meaning Overdose, of course. I mean, how much is too much? Okay, I can hear you snigger in the back and shout "when it croaks, it's usually too much", but being rather inexperienced (although I have an amp that goes way up to 11) I would like to know a few things.

For example.

I just overclocked my CPU (i7-860) and RAM (Kingston HYX 1333MHz) through the automatic "CPU Level-Up" on my Asus P55 Gene mobo. The setting for the RAM was "Auto" so the mobo calculates voltage and clockspeed according to the CPU's speed. CPU has its stock cooler, by the way. The "Level-Up" mode has three settings: "i870@2.93MHz", "Crazy@3.06GHz" and "Crazy@3.36Ghz". I chose the latter.

The result is: CPU@3.36GHz/45°C (normally 2.8GHz/28°C) and RAM@1600MHz.

Everything runs smoothly (though I haven't tested this setting for longer periods, because of the stock cooler), but in Everest (Ultimate edition 5.30.1900) I saw that the RAM's DIMM voltage went up from 1.59v to 1.81v...

There's my problemo, muchachos: I don't mind speeds going through the roof, but when the voltage starts swinging towards the ceiling, I'm a tad reluctant to keep on tinkering...

I know that there's no pre-stamped frame for RAM overclocking, as the rims react all differently, even the ones from the same brand/speed, but does anybody know more or less what the limits are?

Kougar 23rd December 2009 21:32

That is one of the reasons I don't use "auto" overclocking programs. They are based upon general results the individual mobo manufacturer has observed and includes an extra buffer room to "ensure" stability for the worst CPUs. Your CPU and RAM would probably operate fine at voltages below what the program sets. Motherboards are particularly bad at detecting RAM voltages from what I've seen...

The only thing you can really do is learn precisely what point your CPU switches from unstable to stable at specific speeds/voltages. Same with your RAM. Learning what the upper limits are for the various components helps with keeping a frame of reference and knowing when you're pushing the line.

I haven't paid close attention to DDR3 (since Core i7 920's are limited to 1.65v RAM), but JEDEC spec and most manufacturers build the RAM to 1.5v spec. I think 1.8v is about the safe limit, but JMke/Massman would know better.

Kurgan 23rd December 2009 22:14

When I OC'd my Q6600 and Q9450 under watercooling I did alot of research on what it took other people to get to a predetermined level Q6600@3.6 and Q9450@3.8 and used a "generous" but not really high voltage,then I ran Prime95 until I was satisified the system was stable.

Once I was happy the system was stable, I'd bring the voltage down in low incriments until stability became an issue and raised it back up a tad.

jmke 23rd December 2009 22:27

I stopped ocing the PC I used daily, when it crashed just when you needed it most; that was back in the days of Athlon XP ;)
I value stability over speed, and been running my main rig at stock speeds, no regrets, 100% stability, have never experienced blue screen, hard freeze in this WinXP install which I have since 2004

other rigs here my have the occasional OC, but if its purpose is other than CPU/VGA tests, I won't overclock those components to the limit, to eliminate possible failure;)

JimmyJump 23rd December 2009 23:35

Quote:

Originally Posted by Kougar (Post 251160)
That is one of the reasons I don't use "auto" overclocking programs.

It's not a 'program' as such, the OC is set through the BIOS.

Quote:

Originally Posted by Kougar (Post 251160)
Motherboards are particularly bad at detecting RAM voltages from what I've seen...

Never detected any question-raising discrepancies the last, say, 5 years. before that, some readouts were indeed to take with a pinch of salt.

Quote:

Originally Posted by Kougar (Post 251160)
The only thing you can really do is learn precisely what point your CPU switches from unstable to stable at specific speeds/voltages. Same with your RAM. Learning what the upper limits are for the various components helps with keeping a frame of reference and knowing when you're pushing the line.

System runs stable. I just played a spot of "Oblivion" for two hours under the aformentioned settings. Only thing I have to watch is the CPU temperature, running the Intel stock cooler. But that seems to stay below 60°C even with a 20% overclock. I'll have an Artic Cooling (Freeze Xtreme Rev. 2) CPU fan coming my way in a couple. See what that one does.

Dimms don't get too hot either. Around 60°C also. Is a guess as I measured those after having finished playing and the PC settled again. With all 7 coolers at max speed (Antec 1200) the massive airflow cools-off everything to 'standard' temperatures pretty fast.

The warning lights on the mobo, both for the RAM and CPU stay green all the time as well (orange for serious overclocking, red for going bonkers). I just lowered the OC settings to "Crazy@3.06GHz", which lowers the RAM speed to 1460MHz@1.7v

Quote:

Originally Posted by Kougar (Post 251160)
I haven't paid close attention to DDR3 (since Core i7 920's are limited to 1.65v RAM), but JEDEC spec and most manufacturers build the RAM to 1.5v spec. I think 1.8v is about the safe limit, but JMke/Massman would know better.

I don't understand what you're saying here, Kougar. i920 is LGA 1366 with DDR3 Triple Channel, which has higher voltages. It's the LGA 1156 DDR3 Dual Channel that is limited to 1.65v.

And a lot of RAM manufacturers (like Kingston) are nowadays giving lower JEDEC specs than what the RAM can actually take, just to give overclockers the illusion of huge overhead. With the result that some RAM can be overclocked with as much as 1000MHz... :naughty:


Quote:

Originally Posted by Kurgan (Post 251166)
When I OC'd my Q6600 and Q9450 under watercooling I did alot of research on what it took other people to get to a predetermined level Q6600@3.6 and Q9450@3.8 and used a "generous" but not really high voltage,then I ran Prime95 until I was satisified the system was stable.

Once I was happy the system was stable, I'd bring the voltage down in low incriments until stability became an issue and raised it back up a tad.

Thanks Kurgan. I just wonder why the CPU's are e-mail linked?

Quote:

Originally Posted by jmke (Post 251169)
I stopped ocing the PC I used daily, when it crashed just when you needed it most; that was back in the days of Athlon XP ;)
I value stability over speed, and been running my main rig at stock speeds, no regrets, 100% stability, have never experienced blue screen, hard freeze in this WinXP install which I have since 2004

other rigs here my have the occasional OC, but if its purpose is other than CPU/VGA tests, I won't overclock those components to the limit, to eliminate possible failure;)

This isn't my daily rig, it's a system I put together for gaming and doing the occasional Photoshop stuff. My daily rig is a E8400/Asus P5K-E based set-up. My reason for having a game-specific machine, is for the same reasons you mention :)

Kougar 24th December 2009 03:41

Program or BIOS, either way they use predetermined settings programmed by the manufacturer's own testing of a "sample size" of processors, and I presume also RAM. They then program settings suitable for even the worst OCing processors, which means if you have an average to good OCing chip it's more voltage than you actually need. Neither the BIOS / software is smart enough yet to auto-OC on settings specific to your own hardware.

Also if you overclock it's generally recommended to disable Speedstep, C1E, and other power saving features. You can leave them on but as you overclock higher you'll risk creating instability. These power saving features modify both the frequency and voltage of the hardware and aren't designed for the CPU to be operating faster than it was programmed for.

Quote:

I don't understand what you're saying here, Kougar. i920 is LGA 1366 with DDR3 Triple Channel, which has higher voltages. It's the LGA 1156 DDR3 Dual Channel that is limited to 1.65v.
I am not aware of any RAM voltage limitations for LGA1156. LGA1366 processors have RAM voltage limitations because of the QPI bus. Intel warns that there cannot be a 0.5v difference between QPI and RAM voltages. So they warn against anything above 1.65v. This is why all triple channel memory kits you see on the shelf do not exceed 1.65v in ratings.

Triple channel kits don't require additional voltage by default, it just depends what you run the settings at. For me 1.65v is plenty for 3 x 2GB @ DDR3-1600 7-7-7-16 Again this depends on the make/model RAM you get as they are all binned differently. For the adventerous some sites have shown that users can raise the QPI bus voltage to safely raise the RAM voltages beyond 1.65, as long as the difference is kept below 0.5v.

I don't follow what ya mean, but JEDEC specs themselves don't change. RAM that adhere's to JEDEC's specification must be built to run at one of their specification settings. http://en.wikipedia.org/wiki/DDR3_SDRAM

JEDEC only endorses voltages of 1.5v, so if memory manufacturers want to sell overclocking RAM, all they need is the right kind of chips and voltages way above 1.5v, then they can sell the memory at whatever rating they please. ;)

Quote:

Originally Posted by jmke (Post 251169)
I stopped ocing the PC I used daily, when it crashed just when you needed it most; that was back in the days of Athlon XP ;)
I value stability over speed, and been running my main rig at stock speeds, no regrets, 100% stability, have never experienced blue screen, hard freeze in this WinXP install which I have since 2004

other rigs here my have the occasional OC, but if its purpose is other than CPU/VGA tests, I won't overclock those components to the limit, to eliminate possible failure;)

Hence my comments awhile back on another thread about personally only being interested in stable overclocks... if it's not 100% stable I don't care and would never use it.

With the correct overclock settings, I've never experienced any crashes that were not the result of a badly written program, outdated version, or driver. This PC is typically left running 24/7 under full loads for 1-3 weeks between reboots, so 99.9% stability is not good enough. :)

Jimmy, for OC stability testing I tend to stick with LinX... 25 runs at 100% of your RAM on a 64bit OS generally is the best I've seen for checking stability.... it doesn't test memory well, as I've seen poeple pass linx runs but get crashes or Prime errors because of unstable RAM. Because overlocking the RAM achieves so little in the way of benefits I don't even bother OCing the RAM, I just change the CPU multiplier to keep the RAM at or below model specifications.

JimmyJump 24th December 2009 09:06

Quote:

Originally Posted by Kougar (Post 251178)
I am not aware of any RAM voltage limitations for LGA1156. LGA1366 processors have RAM voltage limitations because of the QPI bus. Intel warns that there cannot be a 0.5v difference between QPI and RAM voltages. So they warn against anything above 1.65v. This is why all triple channel memory kits you see on the shelf do not exceed 1.65v in ratings.

Well, maybe I remembered wrong then. I thought LGA1156 had those limitations because of the PCI lanes being imbedded on the CPU...

Quote:

Originally Posted by Kougar (Post 251178)

I don't follow what ya mean, but JEDEC specs themselves don't change. RAM that adhere's to JEDEC's specification must be built to run at one of their specification settings. http://en.wikipedia.org/wiki/DDR3_SDRAM

JEDEC only endorses voltages of 1.5v, so if memory manufacturers want to sell overclocking RAM, all they need is the right kind of chips and voltages way above 1.5v, then they can sell the memory at whatever rating they please. ;)

I've got some Kingston HyperX DDR2 PC8500 RAM with a JEDEC specification of 1.5v/800MHz... That RAM was designed to run at 1066MHz, but because at the time there were no mainboards supporting that speed, Kingston changed the specifications to those I mentioned...

I'm not trying to sell this as fact (in other words: it's my own speculation), but I think most 'big' RAM manufacturers (OCZ, Kingston, G-Skill, Corsair, you name it) test their RAM at much higher speeds than the specifications they sell it under, just to be on the safe side with OC over-enthusiasts. Which would amount to nothing, in my eyes, as a die hard OC-er is always searching for the limits...



Quote:

Originally Posted by Kougar (Post 251178)
Jimmy, for OC stability testing I tend to stick with LinX... 25 runs at 100% of your RAM on a 64bit OS generally is the best I've seen for checking stability.... it doesn't test memory well, as I've seen poeple pass linx runs but get crashes or Prime errors because of unstable RAM. Because overlocking the RAM achieves so little in the way of benefits I don't even bother OCing the RAM, I just change the CPU multiplier to keep the RAM at or below model specifications.

Well, I just reverted back to Win7 32-bits, because of lack in software support. Okay, so most 32-bits software runs perfectly smooth on a 64-bits OS, but I think I'll wait nevertheless.

For the rest, thank you very much for the interesting and useful advise, Kougar. I'll check some of the stuff you mentioned. May even bother you guys with some results too ;)

leeghoofd 24th December 2009 10:27

does the biops readout any other voltages Jimmy ? Usually they are too much for the required speeds... especially dangerous are eg QPI and RAM voltages voltages like Kougar already mentioned.

If you can readout the voltages, I would set them a tad lower and retest and so on till you find the lowest stable voltage...

Did it also increase Vcore ? I would start experimenting a bit mate to find a good combination between extra speed, voltage requirements, noise and heat output....

Kougar 24th December 2009 12:11

Quote:

Well, maybe I remembered wrong then. I thought LGA1156 had those limitations because of the PCI lanes being imbedded on the CPU...
Not that I am aware of. IIRC PCIe runs at 1.5v, but QPI runs at 1.1v.... that is 0.4v of difference right there. ;)

PCIe built into Lynnfield means you have to use more voltage when overclocking, because you are also overclocking the PCIe controllers.

I agree completely with Leeghoof... if you are worried about voltages yet want to OC more, you need to manually test how low you can go on the various voltages.

Massman 24th December 2009 12:59

Quote:

Originally Posted by Kougar (Post 251178)
I am not aware of any RAM voltage limitations for LGA1156. LGA1366 processors have RAM voltage limitations because of the QPI bus. Intel warns that there cannot be a 0.5v difference between QPI and RAM voltages. So they warn against anything above 1.65v. This is why all triple channel memory kits you see on the shelf do not exceed 1.65v in ratings.

The voltage limitation for the LGA1366 platform was set in place to limit the difference between the voltage provided for the memory and the voltage provided for the integrated memory controller. The QPI bus is the connection between IOH and CPU die; it doesn't have anything to do with the memory.

I know that some manufacturers have name the IMC voltage "QPI Voltage", but that naming is incorrect as there's no such thing as a QPI voltage. QPI is the data bus between CPU and IOH; voltage for the bus are provided by the transmitter (Vioh) and receiver (vUNC). The voltage for the uncore part of the CPU, which is referred to by Intel as "non-core", is equal to the VTT voltage.

As for LGA1156, there should be limitation set in place and in fact lower than the 1.65V we see now. I've explained this in the original Core i5 mainboard article here on Madshrimps:

Quote:

The question I want to bring forward here is as follows: Given the big fuss Intel made regarding the maximum memory voltage when releasing the first Core i7 series, why didn't they do the same with the new generation of Core i5 processors? If exceeding 1.65V on a Core i7 platform is, according to Intel, indeed not within safety ranges, how can they not tell anything about the new safety ranges on the Core i5 platform, which has a lowed stock VTT voltage to begin with and for which the Intel data sheets clearly indicate that 1.65V Vdimm is the absolute maximum?

~ http://www.madshrimps.be/?action=get...&articID= 960
Sadly enough, I never received any further information about this :(.

JimmyJump 24th December 2009 13:24

Quote:

Originally Posted by leeghoofd (Post 251193)
does the biops readout any other voltages Jimmy ? Usually they are too much for the required speeds... especially dangerous are eg QPI and RAM voltages voltages like Kougar already mentioned.

If you can readout the voltages, I would set them a tad lower and retest and so on till you find the lowest stable voltage...

Yes there are. let me check on those and I'll get back to you. Am only using one PC atm... (well, two, but watching TV on the other one through an USB DVB-T, which is too sensitive and drops out when surfing and stuff)

Quote:

Originally Posted by leeghoofd (Post 251193)
Did it also increase Vcore ? I would start experimenting a bit mate to find a good combination between extra speed, voltage requirements, noise and heat output....

Yes it did increase V-Core. I think I'm gonna wait till I have my new CPU cooler to have another go at experimenting. especially considering the technical mumbo-jumbo Massman threw in my general direction. That was almost scary dude. I think I'm gonna do a bit of reading on those technical aspects, just to really know what exactly I'm doing here... :)

leeghoofd 24th December 2009 13:47

yep I would run everything stock till you get a good CPU cooler, then start lookign for a bit more ooomph. But i would strongly suggest to stay way from them AUTO settings. Eg my Rampage Gene II would set 2volts for my low voltage rams. If I didn't check it I would have smoked them for sure...

And yes Massman is scary I second that, far too much technical info in that little head...

Kougar 24th December 2009 21:22

Hm, I thought the uncore voltage could not be adjusted, but going by what you are saying if I adjust the "QPI voltage" (what Gigabyte unfortunately labels it) I am in fact changing the VTT voltage/uncore voltage? Or does changing the VTT not affect the uncore voltage as I had thought?

So Intel does mention a 1.65v limitation in the data sheets for LGA1156? Then to pose a question, what about the PCIe controllers, those would still be running at 1.5v, which is 0.4v higher than the VTT on the same die. Why would VTT/RAM voltage be an issue but PCIe wouldn't? IMC's just that sensitive? Anandtech mentioned something about amplitude, but was as much as I remember...

Massman 24th December 2009 21:42

Quote:

Originally Posted by Kougar (Post 251279)
Hm, I thought the uncore voltage could not be adjusted, but going by what you are saying if I adjust the "QPI voltage" (what Gigabyte unfortunately labels it) I am in fact changing the VTT voltage/uncore voltage? Or does changing the VTT not affect the uncore voltage as I had thought?

So Intel does mention a 1.65v limitation in the data sheets for LGA1156? Then to pose a question, what about the PCIe controllers, those would still be running at 1.5v, which is 0.4v higher than the VTT on the same die. Why would VTT/RAM voltage be an issue but PCIe wouldn't? IMC's just that sensitive? Anandtech mentioned something about amplitude, but was as much as I remember...

You are right about the PCH voltage: this one has to be adjusted as well when you're increasing the voltage of the VTT. There's not that much documentation on this (well, not when I was doing research), but it's assumed that keeping the VTT and PCH within 0.4V is within safe range.

I've experienced this problem, by the way, when I was testing the MSI P55-GD80:

Quote:

- If you plan to use 1.6V VTT, increase the PCH 1.05V to 1.3V. I've asked MSI to make this happen automatically, but this is the key to make it happen. The reason is really simple: PCH is integrated on uncore die, so ...

~ http://hwbot.org/forum/showpost.php?...4&postcount=26
If I'm not mistaken, the entire non-core die makes usage of the VTT plane; this includes PCH and IMC.


All times are GMT +1. The time now is 07:09.

Powered by vBulletin® - Copyright ©2000 - 2024, Jelsoft Enterprises Ltd.
Content Relevant URLs by vBSEO