I think I got a good OC Q6000 chip...

[Gerr]

I was lucky enough to get a G0 stepping of the Q6600 during Fry's Black Friday deal of $199 for this CPU and a crappy ECS mobo. Last night, I used Prime95 v25.5 to stress test my CPU for an hour. Two of the cores maxed out at 52*C and the other two at 48*C, and this is with the stock Intel HSF, but using Arctic Silver Ceramic thermal compound instead. Room temp was around 72*F.

For Xmas, I plan on buying a OC friendly mobo, the ASUS P5K-E is on the top of my list at the moment, and an Arctic Cooling Freezer 7 Pro HSF. Once I have both of these, my goal is to reach 9x333 for a 3.0Ghz CPU.

Only problem is my memory, which is 2x2GB of SmartDate DDR2-800 with 5-5-5-15 times, which I don't think will OC that well.

Think my goal is reasonable once I get my Xmas stuff?

 

[BonzaiDuck]

Geez-- that seems almost "worth it" -- to pay less for both the G0 and a crappy mobo than you would pay at NewEgg just for the processor.

The P5K's seem to be pretty good.

Get some good RAM. The prices are ridiculously low right now. You can get D9 chips on a pair of G.SKILLs 2 x 2GB for between $100 and $130, so the 2GB dual-channel kits would be just over half that. I've seen Crucial Ballistix 2GB kits for between $50 and $60, and you'd probably only need DDR2-800 Ballistix or Tracer modules:

Crucial Ballistix 2GB 2x1GB DDR2-800's

I've said it at least once before -- you can "go with" the low-end of the Intel processor line, and know what you're getting. You can trim costs on a good mobo by buying "standard" instead of "deluxe," if you're willing to forego extra features.

But I always spend more than I need to for RAM, to allow for over-clocking flexibility -- either at high speed and not-so-loose latencies, or at lower speed and very tight latencies.

 

[Nurn]

I wouldn't worry too much about your ram. Even if you overclock your Q6600 to 9x333, your ram will be underclocked (assuming your divider is set at 1:1). You may very well find that your ram when running at DDR2-667 speeds (2x333) can be tightened up to 4-4-4-12. Even if you continue to run it at 5-5-5-15, you really won't see much of a difference in performance.

You don't have to worry about temperatures on your Q6600 cores, until they hit 70 degrees Celsius after a couple of hours on load test. 9x333 is a pretty mild overclock, you may not even need to bump up your CPU voltage from the default, in which case your temps should not go up all that much.

Sounds like you got a pretty good deal.

Edit.... actually the temps you report sound almost too good to be true. Be careful, if you use Speedfan to report temps, they can be out by 15 degrees Celsius. Use Coretemp.

 

[BonzaiDuck]

Most people here would confirm that 9x333 to 3Ghz for a Q6600, either B3 or G0 stepping, is a piece of cake -- a done-deal -- absolutely, positively, certainly -- YES!!

Even with the B3, you should be able to "get there" at 0.05V lower VCore than the maximum spec and 0.07V less than the so-called "default" of 1.372V. It can only be better with the G0, and specifics have been quoted all over the place in these forums for the last several months from time the G0 first became available.

As I said -- I have a B3 and can get it to 3.2 without problems of heat or failure at <= 1.418V. With the Crucials, I can tweak the latencies with 1:1 CPU-DDR ratio at only DDR2-712 to get bandwidth benchmarks that I find very nice, if not stunning. Second-best with benchmark result of 9,400 MB/s "Read" uses 2T command-rate and 2.175V memory voltage; I can get almost 9,800 MB/s "Read" with 1T and 2.2V, which is the warranty-recommended voltage for those Crucials. Of course, I'm using Ballistix DDR2-1000's.

The 800's may not be so "downwardly elastic" with the latencies, but they'll still be good. I'm experimenting with some Tracer 800's this week to see if I can get CAS=3 and tRAS <=8 -- I'll let you know.

 

[BonzaiDuck]

Not dissing you Nurn, but compare your statement "tightened to 4,4,4,12" and my experience of 3,3,3,6,1T, tRC=9 1:1 at 700 Mhz.

We need to emphasize that there's less need for the Viagra-Big-Bon-Iva of Mega-spurtz if you increase the bandwidth through lower latencies. With the latencies, you're reducing N of clock-cycles per memory op; with increased frequency, you're increasing the number of clock-cycles-per-second. Fewer clock cycles corresponds to greater memory bandwidth (and speed) no less than higher megahertz means packing more clock-cycles into a unit of time. You pack more clock-cycles into a unit of time if each mem op takes fewer clock-cycles.

$50 to $70 is not too much to pay for that sort of difference. tCL of 3 versus 4, tRCD of 3 versus 4 -- yield big increases in bandwidth, and lowering tRAS accordingly means possibility of dropping tRC = tRP + tRAS for more enormous bandwidth improvements.

 

[emfiend]

I don't know much about the ECS board, but given the OC-ability of the Q6600, just setting the multi to 9x333 is worth a shot with your current system. I wouldn't be surprised if it works. The worst that will probably happen given your stock results is that the board wont post.

 

[Nurn]

If you've got $50 to $70 to spend, and you're looking for your best bang for the buck value, you'd be better of with buying a decent CPU cooler (e.g. Thermalright 120 or ultima90), to allow you to get 2 or 3 hundred more mHz out of your CPU. Tightening up your memory timings can possibly yield real-world performance gains of 2 or 3%, but bumping up your CPU from say 9x333 to 9x367 will get you a whole lot more benefit.

 

[jeffw2767602]

If I were you I would Ebay that ECS Mobo and hopefully get $30 or $40 out of it to put towards a Thermalright Ultra-120 Extreme and a Scythe Sflex 1600RPM fan. Then you could ALMOST certainly hit 3.6GHz at 9x400 on that ASUS mobo you are looking at and you wouldn't need to replace the ram since your FSB would be 400MHz. Most G0s can hit 3.6 with about a 10% bump in voltage or less (though results vary, obviously). I've seen some G0s able to hit 3.6 on stock voltage, but that is very very rare. However, if you are only shooting for 3.0GHz you could probably do that with the stock cooler, and you may even be able to decrease the voltage. Your ECS mobo and RAM would probably be okay at 3GHz as well, but I dont have any experience with ECS so I dont know for sure.

 

[Gerr]

I used Core Temp to monitor my CPU temps, but only ran it for 60 min, probably could have run it longer. Also remember I used a better thermal compound, and have found the ceramic version of the AS line is the best to date, getting really good temps from in on all uses.

I already have the SmartData memory, just got it, so not looking to buy more. I could probably tighten it up at 667 to 4-4-4-12 and that would be good enough for me. I could always upgrade my memory later.

My goal is 3.0Ghz(9x333) at stock voltages, but might bump it higher if I can, but start to get nervous when I push it too far. Once I am stable at 3.0, might try for 3.2, but that's probably as far as I go. I like round numbers, and at 3.0, that's a good OC, plus with Speed Step, it reduces it to 2.0, nice round numbers...

 

[Gerr]

I also have researched the HSF's and found that the AC Freezer 7 Pro is only just behind the top guys in cooling preformance, but at less than half the price. Since I am not pushing my OC to far, it should be more then good enough for me.

 

[Gerr]

Jeff, I thought most mobo's have problems reaching the 400 FSB mark? 3.6Ghz would be nice, but that's one hell of an OC, not sure I would feel safe at that level. Ya the performance would be nice, but peace of mind is very underrated.

 

[BonzaiDuck]

Originally posted by: Nurn
If you've got $50 to $70 to spend, and you're looking for your best bang for the buck value, you'd be better of with buying a decent CPU cooler (e.g. Thermalright 120 or ultima90), to allow you to get 2 or 3 hundred more mHz out of your CPU. Tightening up your memory timings can possibly yield real-world performance gains of 2 or 3%, but bumping up your CPU from say 9x333 to 9x367 will get you a whole lot more benefit.

1,000 MB/s is not a 2 to 3% "real-world" performance gain, even if the benchmark result is "synthetic."

It's simple integer math.

Two or more years ago, THG plugged the OCZ-DDR-500 EL Gold modules because you could OC them 25% to the spec, but the latencies were at best 3,4,4,7. Conclusions from that article and some others suggested that pushing the FSB yielded gains in excess of tightening the latencies. Then, a year or so later, they reversed that position.

But those were DDR1 days. Here, we've got SPD and EPP defaults for memory with tRAS as high as 15 to 18. RAS has very little impact on bandwidth by itself, but it limits the bank-cycle-time. So if you can cut tCL, tRCD and tRAS -- with tRAS down from 15 to a range between 6 and 9, you can reduce tRC to between 9 and 12 -- down from some defaults I've seen of 21.

So again -- N of clock-cycles-per-op is just as important as N of clock-cycles-per-second.

If you don't believe it, I can show you my 3DMark06 scores as well as my Sisoft and Everest memory-bandwidth benchies.

 

[BonzaiDuck]

:thumbsup: for the TR-Ultra-120-Extreme and Ultima-90 though . . . . .

Definitely need good cooling, even for modest over-clocks if you want to insure less heat-stress to the processor.

 

[Gerr]

Everything I have read on the AC Freezer 7 Pro says it's not a cheap cooler. It's a low/mid priced cooler that preforms close to the high end ones. Best bang for the buck IMO. Yes a Ultra-120 would be better, but the performance difference between the 2 coolers doesn't warrant the price difference.

 

[BonzaiDuck]

Freezer PRo is "OK." I can't remember if it was among the 15+ coolers compared to the TR U-120-Ex last May.

But I did benchtests on several factors:

-- Diamond thermal paste
-- Ducting improvements
-- choice of coolers

I could measure the improvement with the better thermal paste. I could measure the improvements separately for the ducting.

And I could confirm Anand's result for the U-120-Ex, while showing a 5 to 6C improvement over the original Tr-Ultra-120 at same room ambient measured to within 0.3F degrees -- and separately from the other two factors.

Even if I was "cautiously over-clocking," the choice of the cooler, the paste and a bunch of foam-board and plexiglas meant a difference of between 8 and 10C degrees. So for over-clocks between 25 and 33%, the core temperatures at load were equal to those for all-stock settings.

My opinion: for the difference of $20 or $30, it's worth it to get a high-end air-cooler. No dissing the Freezer-Pro, though. . . . . . IT'S OKAAYYYY.

 

[Nurn]

Bonzai, you obviously know a lot more about memory than I do. I always figured that tighter latencies didn't really give a big boost to performance. I would definitely be interested in seeing your 3dmark06 numbers for different latencies with all other things being equal. You may persuade me to tweak my own ram, as I have pretty much ignored it during my overclocking adventures (my ram is 2 x 2 GB Mushkin DDR2-800 rated for 4-4-4-12, coupled with a Q6600 that I run at 9x333 with a 5:6 divider.

 

[myocardia]

With the C2D architecture, memory read and write speed doesn't make much difference, whether it's from lower latencies or higher speed, at least when using real-world apps. Sure, it's measurable, it just isn't really worth spending any amount of time tweaking, unless your primary use for that system is SuperPi or Sisoft Sandra.

 

[aigomorla]

Bonzai,

The Q6600 at frys on black friday was 199 /w ECS board.

LOL... now where can you find a brand spanking new in BOX Q6600 for 199?


Yeah i picked one up too. I couldnt resist the price. And its going into either my NAS, or my new project.

Building a Budget Watercooling system around a CM690 this XMAS, which should take that sucker to 3.6 easily, and a 8800GT in the same loop.



But yes 333x9 is very possible on these G0's And i also got a G0 as well.

The board is sitting in my pile of junk box. I have about 3 other ECS boards sitting in there. Maybe i'll post them on egay. :T


As for the Board, i recomend the Gigabyte P35-DS4. Its a pretty bad ass board, and im starting to like Gigabyte more then ASUS.

The ABIT IP35-PRO also might be a good recomendation as well.

If you can still find it, the best board ive played with was the Gigabyte P35-DQ6. And many DQ6 owners will tell you that board rocks.

Originally posted by: Gerr
Jeff, I thought most mobo's have problems reaching the 400 FSB mark? 3.6Ghz would be nice, but that's one hell of an OC, not sure I would feel safe at that level. Ya the performance would be nice, but peace of mind is very underrated.

I havent seen 1 P35 board which had issues passing 400fsb.

And whats wrong with 3.6ghz? Im running 3.73 loaded all the time @ 1.45Vcore with load temps = to people's Idle temps. :T

As long as your conservative on your voltage, and you keep those temps in check, i dont think theres anything wrong with 3.6ghz. Most G0's can do that with around 1.38-1.42Vcore :T

Originally posted by: jeffw2767602
I've seen some G0s able to hit 3.6 on stock voltage, but that is very very rare.

Link please...

I think you misread the CPU-Z version and it was an old one which reported everything over 1.3Vcore as 1.195 or something around there.

The new CPU-Z fixes all that and displays correct vcore.

 

[jeffw2767602]

aigo there are a couple on xtremesystems forums, check the intel section. i think there are a few in the official q6600 g0 overclocking thread. mine can reach 3.33GHz or so at stock voltage (VID 1.30v) so i would definately say it is plausible that there are a few great q6600s that can do it.

 

[jeffw2767602]

Originally posted by: Gerr
Jeff, I thought most mobo's have problems reaching the 400 FSB mark? 3.6Ghz would be nice, but that's one hell of an OC, not sure I would feel safe at that level. Ya the performance would be nice, but peace of mind is very underrated.

Most p35s can do 450+ with quads. my p35-dq6 craps out at around 480mhz. 400mhz is a piece of cake and i highly doubt it will hurt longevity. i have my g0 at 3.6ghz with 1.44v 24/7. and i know how you feel about peace of mind. i need this comp to last a little while due to limited funds. women and college are expensive.

 

[aigomorla]

Originally posted by: jeffw2767602
aigo there are a couple on xtremesystems forums, check the intel section. i think there are a few in the official q6600 g0 overclocking thread. mine can reach 3.33GHz or so at stock voltage (VID 1.30v) so i would definately say it is plausible that there are a few great q6600s that can do it.

my ES can do 333x9 on stock voltage.

400x9 on stock is almost impossible. You'd need a QX9650 to pull that off.

Ive bee looking on the overclocking thread and no one has it at stock voltage.

So can you link it to me?


dont mean to instigate with you jeff, cuz curious on where you saw that 3.6ghz at stock voltage.

 

[BonzaiDuck]

It took ME a while to come up to speed on timings. I think it was 2003 when a friend had got some Mushkin DDR-433's and was boasting about the stock timings (2,2,2,5). He didn't know how or why it made a difference: his local computer shop sold him on the value of it. Before that, I just did software programming, database, and built "stock" computers.

One good place to look is a web-site called "Lost Circuits" __ probably lostcircuits.com. Adrian Rojak's BIOS guide has lots of info. I stumbled onto another guide -- don't have the link available at the moment -- for (mostly) 680i OC'ing.

But it only stands to reason, per the trade-off I mentioned, between clock-cycles per op and clock-cycles-per-second.

Don't have my screenies available, but the 3DMark06 went from about 10,500 to 11,465 with the "shareware" version at 1280x1024.

I always had this idea that pushing the CPU speed was "just part of it." The bottlenecks are all "lower on the storage-pyramid." More L2 cache may provide an opportunity for better performance -- and especially if you can get the memory to work faster -- either way you do it.

 

[BonzaiDuck]

[Forgot to hit the Quote button -- that was for NURN.]

 

[jeffw2767602]

Originally posted by: aigomorla

Originally posted by: jeffw2767602
aigo there are a couple on xtremesystems forums, check the intel section. i think there are a few in the official q6600 g0 overclocking thread. mine can reach 3.33GHz or so at stock voltage (VID 1.30v) so i would definately say it is plausible that there are a few great q6600s that can do it.

my ES can do 333x9 on stock voltage.

400x9 on stock is almost impossible. You'd need a QX9650 to pull that off.

Ive bee looking on the overclocking thread and no one has it at stock voltage.

So can you link it to me?


dont mean to instigate with you jeff, cuz curious on where you saw that 3.6ghz at stock voltage.

i found it, its in the "retail q6600 g0 overclocking thread", its a week 37b chip that aparently is an awesome overclocker...ill link it in a few minutes when i find it again.

 

[jeffw2767602]

http://www.xtremesystems.org/f...d.php?t=152603&page=67

ineedaname is the poster that claimed his g0 did 3.6 stock. aparently that chip will do 4ghz with under 1.5v. however upon looking at his later posts it appears that 3.6 on stock wasnt entirely prime stable. he however can hit 3.6 with only 1.3v which is still amazing. i wish mine could hit 4ghz!! my g0 seems to scale weird. i can get to 3.33ghz at stock volts (1.3v) but i lose it somewhere between 3.33 and 3.4. then it takes another .14v to get to 3.6 (1.44v) and it takes 1.488v to get to 3744mhz.

Originally Posted by ineedaname
I just got my q6600 g0 batch code is 37B

So far I've done 4ghz @ 1.42v on AIR. Haven't done full stability test yet but it should b doable at below 1.45v.

It can do 3.6ghz at stock volts.

I got a Xeon 3320 also but haven't tested it yet. That one is b3 batch 17A engineering sample.

I'll get some screenies tmr cuz i'm way 2 tired right now. About 2 fall asleep in my chair.