Overclocking Tips - Refresher!

[Sunin]

I know we have a whole seperate forum for overclocking here at [H], but i wanted to share my experience in the last 3 weeks with 3 different Q6600's.

1. No two CPUs are alike, ever.. it seems
2. Memtest first to ensure memory is stable or you won't OC worth crap
3. Sweet spot with a Q6600 with DDR26400 memory is 3.6Ghz since memory runs 1:1
4. Lower the multiplier to 6, 7, or 8 and max your FSB, and run prime for 15 min and then up your FSB and try again.
5. Once you determine your max FSB back it down to a speed the mobo is rated to run at such as 333 for 1333 fSB mobo's and try the CPU multiplier at 8x and then 9x again using Prime 95 to test each core for 15min.
6. Once you have a stable OC, say 333x9 = 3Ghz @ stock voltage then the next thing to do is edge it up a notch until you notice it errors. Increase voltage 1 or 2 levels which adds approximately .02v to the CPU, and try again.
7. Use everest or some tool to read the vdroop, many mobo manufacturers have their own tools, its important to know that say 1.35v in bios is actually 1.3v to the CPU...
8. Use Coretemp to monitor temps, if you exceed 60c you need to either back down the voltage and OC or remount the HSF or get a better cooling solution.
9. In general I shoot for no more than 55c at night, maybe a hair more, so that when I look back through my daily temp logs I see no core ever exceeding 60c.
10. In general I don't ever push more than 1.45v through the core. We want these machines to fold for many a year and overvolting 24/7 is rough on them at 100% loads. Also to note, 1.2275v =153watts out of the entire system, 1.385v = 193watts, the difference of a few mhz or so in OC is not worth it in the long run.
11. A High vid as read by Coretemp DOES NOT MEAN A BAD OVERCLOCKER! I had one of the 3 have a vid of 1.35 and with just a nudge above that it was stable at 3.2Ghz and temps don't even hit 50c. Not a great OC but it was liveable considering one of the other processors with a vid of 1.2275 won't go over 3.15Ghz. Now I need to revisit that processor knowing what I know now.
12. If you hit a wall try more voltage, within reason. Try a different multiplier as long as you don't exceed your max stable FSB.
13. Generally I do not overclock memory because the combination of High FSB and High memory clocks only nets you about 1-3% boost and for the heat and issues that ram and HIGH FSB's can cause it just is not worth it.
14. Once your satisified withe overclock I run prime 95 for 1 hr using SFF's the top choice and then load up dual SMPs. Dual SMPs I've noticed run about 3c cooler then the Prime 95 SFF. Make sure you get Prime 95 version 24 which tests all cores at once with only running a single exe.
15. Finally whatever the max is you may want to back it down 100mhz just to give yourself a buffer for when the temps get warm in your house or when the Heat Sink gets all gunked up with dust. Regular maintenance is a must to keep a folding machine in peak operation. I take each machine down once a month and clean it. It takes less than 30 minutes and will help ensure a stable folding farm.

In the end here is what you need to pay attention to: Keep the volts reasonable, Keep temps low, and provide plenty of airflow through the case and you will have a nice PPD machine that will last!

I will be hosting the tools on my website shortly, but I made a USB drive with all of my F@H items (SMP, FahMon, Affinity Changer), All Drivers vid/mobo/etc, and another directory for Everest, Prime95 and CoreTemp. This way within 5 minutes I can pop everything on the new computers desktop and be working toward the end OC and stable SMPs

I realize alot of this is a rehash, but figured it might help a few.

 

[aldamon]

I've found that #8 is not true at all, especially with quads. See #1 for why.

 

[Sunin]

Honestly #8 is 100% true if you have a good cooling solution. Sometimes you need to lap a processor but that is way beyond this guide. I have used a Zalman 9500, which performs slightly less 5c less than the Thermalright Ultra 120 Extreme(TRUE). If you can get the TRUE, do it! I've never had to exceed 60c to get the cpu's to oc well and yes maybe I'm a hair conservative but in general my rule of thumb is never over 60c. Its a good guide, but yes processors do vary and there maybe times that you have to exceed it, but in general it works.

 

[aldamon]

Sunin, I'm sorry man, I know you're trying to help, but the 60C limit is typically for the external temperature sensor. Your tip says to keep it under 60C in CoreTemp. That's just not correct, especially with quads. And I have a TRUE. Thanks.

 

[Sunin]

Ok what are you getting at? That 60c is too high? I'm confused a bit...

The sensors on motherboards are hardly ever accurate and the only real way to monitor temps is via Coretemp. Core temp also shows a max core temp of 120c, but that is just insane...

I can show you 3 Q6600's temp logs, since I keep my temp logs running, that will show mine never raise above 60c and I would not consider my OC's to be bad.

The cores on Quads from intel's own sight says:

http://processorfinder.intel.com/details.aspx?sspec=slacr

Max Thermal Spec: 71c

Thermal Specification: The thermal specification shown is the maximum case temperature at the maximum Thermal Design Power (TDP) value for that processor. It is measured at the geometric center on the topside of the processor integrated heat spreader. For processors without integrated heat spreaders such as mobile processors, the thermal specification is referred to as the junction temperature (Tj). The maximum junction temperature is defined by an activation of the processor Intel® Thermal Monitor. The Intel Thermal Monitor’s automatic mode is used to indicate that the maximum TJ has been reached.

So I guess what am I not getting?

 

[aldamon]

The only reading in CoreTemp that matters is "xC to Tjunction remaining." As long as you enable that option, and it doesn't say 0C to Tjunction remaining, the temp is fine. Your 60C number is the one for the inaccurate sensors, which is why there's a cushion built in. Just look at what you posted:

The thermal specification shown is the maximum case temperature at the maximum Thermal Design Power (TDP) value for that processor. It is measured at the geometric center on the topside of the processor integrated heat spreader.

That would be TCase.

 

[Sunin]

Hmm, I'l have to look into that. I guess than what you are saying is on some processors 60c may be to hot and on others 60c maybe to low. I'll check tonight.

 

[aldamon]

I guess than what you are saying

I give up.

 

[SmokeRngs]

I'll reply in regards to some of this later tonight, probably when I get home, but part of it has to do with the temps and the rest with other overclocking things.

 

[Sunin]

LOL anyone else get what is he trying to get at? Other than you should check it a different way in coretemp? Doesn't really say any which way why a general rule of thumb of 60c is inappropriate. Our own forums here in the OC section has a sticky that states the guys core temps goal was 65c... so I am obviously curious and am trying to understand you aldamon.

Thank aldamon I am trying to understand what you are getting at trust me...

 

[HighYield]

Ok guys, how high can you go with a Q6600? I thought that as long as you were under 70C max you would be ok. I know cooler is always better, but do you really have to worry about it if you peak in the mid 60s?

 

[Kendrak]

I think under 70 is in spec......

However It comes down to lifespan.

Running a proc in the mid 60's and 70's is going to shorten that life span.

I retire a few systems, but almost as many fry, short out, or some other WTF?! way. Mostly cuz I push em [H]ard.

 

[aldamon]

I think under 70 is in spec......

70C is the spec for TCase, or the temperature in the center of the top of the IHS. Intel clarified this in their thermal specification, which Sunin's provided. CoreTemp measures Tjunction (the temperature inside the cores), which is obviously going to be hotter than the top of the IHS.

A tip that tells someone to keep a Q6600 under 60C in CoreTemp doesn't make any sense because it's telling people to keep their TJunction temp under 60C, which is ~48C TCase (what Intel is using for their thermal spec). There's a lot more TJunction headroom than 60C on a Q6600. In fact, no guesswork is needed. CoreTemp tells you the exact "xC to Tjunction remaining."

People get confused because they don't understand what CoreTemp is telling them in relation to thermal specs. Most people don't have CoreTemp. They use the software that came with their motherboard to see temps. Intel's specs are for Joe Six Pack.

 

[Sunin]

I didn't get a chance to check coretemp I was too busy building a cooling solution for my mini server room.

I understand what Aldamon is saying but on the same page I've pushed 1.5v through a Q6600 once to see if I could get it to stabilize at 4ghz. Temps never reached 52c by the cores. Sooo I'm very weary of going much higher I mean yeah I'm well below TJunction, but I'm also getting to the uncomfortable voltage area for 24/7 use...

Each to their own and i will read things via coretemp tonight with Tjunction, but from the sound of it I am ultra low on the temps side. /shrug, guess that is why I have never ever burnt out a cpu in my 20 years of overclocking. Now could I have eeked out more of an overclock in the past, most definitely.

Thanks Aldamon that last explanation was alot clearer it gave me a feeling for where you were going with it and I will try to use TJunction in coretemp in the future without pushing too much voltage.

What would everyone say is the max voltage for 24/7 folding? I get nervous around 1.45 to 1.5v what are your guys feelings?

 

[Xilikon]

1.5v should be the good max for 24/7. As for temps, what matter is the delta to tJunction which matter. In CoreTemp, if you look into preferences, you will see a option to show that delta. Try to keep it over 20-30C to be safe.

Anyway, about shortening the life of processors, what is the difference between 5 years vs 10 ? You would have upgraded by then anyway.

 

[Sunin]

1.5v should be the good max for 24/7. As for temps, what matter is the delta to tJunction which matter. In CoreTemp, if you look into preferences, you will see a option to show that delta. Try to keep it over 20-30C to be safe.

Anyway, about shortening the life of processors, what is the difference between 5 years vs 10 ? You would have upgraded by then anyway.

Exactly as long as it is not 2 years

 

[aldamon]

As for temps, what matter is the delta to tJunction which matter. In CoreTemp, if you look into preferences, you will see a option to show that delta. Try to keep it over 20-30C to be safe.

Thank you.

 

[Axdrenalin]

Good post guys, and thanks for the refresher. I'm a bit wary when I clock my quads, and maybe its just me wanting to be on the side of caution. Of course, the temps I see aren't what I would call great either.

Example: On one setup, I use a GB965P-DS3 with a Q6600 and a Scythe Mine cooler. The temps were ridiculous at 3Ghz (334x9), and would frequently hit mid to high 60's, which I wasn't real comfortable with. This same cooler ran straight 49C on a e6750 at 3Ghz, so I know the cooler was decent. I lapped the CPU a bit (not as much as I should, but still working on that) and I get temps around 60-61 under two SMP's right now.

My main rig uses the exact same specs, MB, CPU, RAM, and HSF - all at 334x9. I get max temps on it of 55C and I've never touched the CPU or HSF.

My third Q6600 sits on a BadAxe2 MB with a Zalman 9500. I've tried this cooler on two different Quads, and it doesn't cool near as well as the Scythe Mine. As long as I keep it horzontal it does okay averaging 55-57C, but when I jack the components into my Antec 900 case the temps go up some because of the weight of the HSF. I've mounted the bracket on it through the heatpipes this time to see if that improves things a bit.

BTW - the ram I'm currently using is Crucial Ballistix 6400 DDR2800

I'd love to get 3.6 on these CPU's and run the RAM 1:1, but if I'm seeing these temps at 3.0Ghz what kinds of temps will I see then? Advice, casual comments and suggestions are always appreciated, especially if it will help me achieve better output. For the first time since I started in February, someone has shown up on my threat list, right Sunin?

 

[Sunin]

First I use MX-2 by ARCTIC Cooling
I also use Thermalright Ultra120 Extreme's (TRUE)

If the processors show signs of high temps I lap them. I bought 10 sheets of 8 different grades and will sandpaper the damn things flat to ensure good temps!

 

[Imitation]

First I use MX-2 by ARCTIC Silver
I also use Thermalright Ultra120 Extreme's (TRUE)

If the processors show signs of high temps I lap them. I bought 10 sheets of 8 different grades and will sandpaper the damn things flat to ensure good temps!

Actually MX-2 is made by arctic cooling, arctic silver is not the same company.

 

[Axdrenalin]

I also use Thermalright Ultra120 Extreme's (TRUE)

Thanks for the tip. Just ordered two of them from Directron @ $55 each, plus I got a couple Scythe S-Flex 1600rpm 120mm fans to go with them. Hopefully these will make a big difference in the temps I'm seeing.

 

[Sunin]

I use Yate Loons High flow fan that does 78.6cfm with the True. I configure 2 fans on the true and can provide pictures, but it is a push pull config and you can definitely feel the air moving!

 

[SmokeRngs]

Remember that for voltages for the CPU, what you set in the BIOS will be different than what is actually being fed to the CPU and the voltage will change on idle vs load. I get pretty bad vdroop on my Gigabyte boards and I don't think my Abit IP35-e is much better. My 965P-DS3 has a core voltage of 1.525 or so set in BIOS but after vdroop and all cores under full load, the voltage is reading around 1.4-1.41. On my P35-DS3R I have about the same set in the BIOS and under full load after vdroop it reads 1.41-1.42. It's the after vdroop voltage you need to worry about as that is the voltage which is actually going to the processor. The Q6600 in the 965P-DS3 is running at 3.51 which seems to be the max for it and the Q6600 in the P35-DS3R is running 3.6. Both processors run those speeds 100% stably.

Since after lapping the TRUE (which dropped the temp of the hottest core by 13C) I think I might see about getting a bit more than 3.6 out of the processor. I now have about 7-8C to play with compared to the watercooling I had on it before which leaves me a little more room to play with voltages. I generally will not worry about temps until they stay steadily above 65C.

My main quad@3.6 has a TRUE on it, my second quad@3.51 has an Ultima90 on it and the third quad@3.6 has a Tuniq Tower on it. The TRUE gives the best temps with the Ultima90 not far behind and the Tuniq Tower bringing up the rear. However, I believe I can drop the temps on the Tuniq Tower when I get around to replacing the stock fan. That fan doesn't seem to push hardly any air even at the highest setting. I inadvertently dropped the temps about 5C on that heatsink when I put an 80mm fan inside the case to blow on some other components with some bleed over onto the the Tuniq. I was rather surprised at this.

Just remember, good cooling is your friend as well as thorough stress testing. I do not consider a hour or so of prime to be thorough. That's barely getting anything done at that point and I've seen systems seem perfectly stable at 1 hour of testing go to hell later on down the road because a little more prime testing would have shown instability in the processor.

 

[Imitation]

Just remember, good cooling is your friend as well as thorough stress testing. I do not consider a hour or so of prime to be thorough. That's barely getting anything done at that point and I've seen systems seem perfectly stable at 1 hour of testing go to hell later on down the road because a little more prime testing would have shown instability in the processor.

+1 smoke, cooling is a big deal for overclocking stability and one thing I decided not to skimp on a while back. I feel like the stock heatsinks are fine, but they're called "stock" for a reason. They work great when you're running stock, but quickly run out of headroom.

And while i've heard some people(not here) doubt the importance of prime stability, they weren't folders. IMO, the worst thing you can do is get greedy and push your system over the edge and start dumping EUE's back to stanford. Take the 20-50mhz speed hit, get the system stable and everyone will be happy.

 

[BillR]

Remember that for voltages for the CPU, what you set in the BIOS will be different than what is actually being fed to the CPU and the voltage will change on idle vs load. I get pretty bad vdroop on my Gigabyte boards and I don't think my Abit IP35-e is much better. My 965P-DS3 has a core voltage of 1.525 or so set in BIOS but after vdroop and all cores under full load, the voltage is reading around 1.4-1.41. On my P35-DS3R I have about the same set in the BIOS and under full load after vdroop it reads 1.41-1.42. It's the after vdroop voltage you need to worry about as that is the voltage which is actually going to the processor. The Q6600 in the 965P-DS3 is running at 3.51 which seems to be the max for it and the Q6600 in the P35-DS3R is running 3.6. Both processors run those speeds 100% stably.

Since after lapping the TRUE (which dropped the temp of the hottest core by 13C) I think I might see about getting a bit more than 3.6 out of the processor. I now have about 7-8C to play with compared to the watercooling I had on it before which leaves me a little more room to play with voltages. I generally will not worry about temps until they stay steadily above 65C.

My main quad@3.6 has a TRUE on it, my second quad@3.51 has an Ultima90 on it and the third quad@3.6 has a Tuniq Tower on it. The TRUE gives the best temps with the Ultima90 not far behind and the Tuniq Tower bringing up the rear. However, I believe I can drop the temps on the Tuniq Tower when I get around to replacing the stock fan. That fan doesn't seem to push hardly any air even at the highest setting. I inadvertently dropped the temps about 5C on that heatsink when I put an 80mm fan inside the case to blow on some other components with some bleed over onto the the Tuniq. I was rather surprised at this.

Just remember, good cooling is your friend as well as thorough stress testing. I do not consider a hour or so of prime to be thorough. That's barely getting anything done at that point and I've seen systems seem perfectly stable at 1 hour of testing go to hell later on down the road because a little more prime testing would have shown instability in the processor.

+1 smoke, cooling is a big deal for overclocking stability and one thing I decided not to skimp on a while back. I feel like the stock heatsinks are fine, but they're called "stock" for a reason. They work great when you're running stock, but quickly run out of headroom.

And while i've heard some people(not here) doubt the importance of prime stability, they weren't folders. IMO, the worst thing you can do is get greedy and push your system over the edge and start dumping EUE's back to stanford. Take the 20-50mhz speed hit, get the system stable and everyone will be happy.

You both make good points on the importance of cooling. However, you both left out one of the big factors few keep in mind.

CASE COOLING!!!!!

I don’t own a case anymore that doesn’t use a minimum of 3 120mm case fans. If you really want cool duct the hot air from they HSF straight out the top or back of the case.

All my Quads are all in the high 40s low 50sc as far as the difference (delta if you must) and none run above the low 50’s c on a per core basis.

Now, giddyup and fold damnit

 

[Imitation]

Very good point bill. The 120mm fans are key here. Having the ability to load 2 or 3 into a case will def keep it cooler and quieter. I flat out refuse to consider a case if it doesn't handle 120mm fans too.

 

[Wharf_Rat]

I dont really even look at temps anymore. I buy good heatsinks and try to cool the case as best I can.

When the thing starts blue screening or random reboots and more voltage on the CPU and/or MCH doesn't stop it... I stay at the last know stable.

 

[Sunin]

Anyone experience this in their "Vdroop"?

Bios set to 1.525v
Volts at idle 1.36v
Volts under Prime95 1.46v

I have an overall vdroop of .625, but generally vdroop results in the voltage dropping under load and as you can see my volts actually increase .1v under load.

Really weird eh?

I also did a bit of overclocking and eeked out each of my Core Duo/Quads to 3.6ghz stable. So I added another good GHZ of crunching power to my already existent Farm. I also am going to play with my one machine that defaulted to 3.6ghz on default volts. I have a feeling that puppy has 4Ghz in it.