4GHz and beyond?

[chiew]

Hello,
this is my first post here at Anandtech. I like reading the reviews at the main site, and I've read some posts here and people seem friendly and helpful.

My current rig is the following:
Antec P180B v1.1 case
Intel C2D E4300 @ 2.7GHz (9*300), stock voltages
Scythe Ninja Rev B CPU cooler
Gigabyte GA-965P-DS3 w/ 40mm fan on northbridge
2x1Gb Crucial Ballistix 8500 1066mhz RAM (@ 600mhz 3-3-3-12)
Antec truepower trio 650W
BFG 7800GT OC
Sound Blaster Audigy 2ZS

This is my first build and my first time OCing, and unfortunately I seem to have reached a wall of some sorts in regards to the OCing. I cannot get above 300mhz bus speed, even if I increase any combination of core, MCH and FSB voltages (I've tried up to 1.4, +0.2 and +0.2V respectively). I've followed a couple guides and turned off all of the things I'm supposed to (EIST, etc). I've also been disappointed with the northbridge on my DS3, since it is so hot, it requires the 40mm fan to be on, and that thing is pretty whiny and loud. I built my computer based on spending about $500 for the first five major components on the list (case, cpu, mobo, ram and psu), and succeeded, however I'm a bit disappointed that I have not been able to break into the 3GHz range cooly and quietly at the amount of money I've spent.

I was hoping that someone could shed some light on the following possible build to replace my current rig:

Intel C2D E6850 w/ Thermalright Ultra120 Extreme cooler
Gigabyte P35-DS3R (w/ hopefully no additional NB cooling)
same RAM
Silverstone SST-DA750 PSU
Silverstone TJ09 case

The hopes of this build are to avoid active cooling on the northbrige, and to reach 4Ghz quietly and cooly on air (planning maybe on using Nexus 120mms in the case since they are quiet). Instead of taking the cheaper route this time, I'd rather have something that will last awhile. That is the reason I chose the new PSU, as it has all power on one rail and should therefore be able to handle any large current draws future GPUs may have (as well as having the new 8pin GPU power dongle).

My biggest questions right now are the following:
1) will the P35 board allow bus speed of at least 450mhz? The E6850 has a multi of 9x, so that will yield 9*450 = 4050 = 4.05ghz hopefully. I have heard of a couple people breaking 4Ghz, but I was just wondering how likely it is. Such a bus speed will mean I'd be running 1800mhz FSB (450 quad pumped) on a board meant for 1333mhz FSB, which is huge. I am currently runnin 1200mhz FSB on my current 965P, which is only supposed to handle 1066mhz FSB, but 1333 to 1800 is a much larger difference than 1066 to 1200.
2) Will I be able to cool such a CPU on air cooling? And will the Nortbridge be able to handle 1800mhz without active cooling? I don't want to resort to water cooling as it is expensive and potentially messy (The desktop will have to be moved at least once a year).

Sorry for the long post, but looking forward to hearing from you guys .

 

[firewolfsm]

I don't think an upgrade is worth it for such a small increase in performance, you already have a Core 2 Duo, why buy another one? Also, nothing's wrong with your case or powersupply, I don't see why you'd want new ones.

It might be better to just buy the motherboard, that should run cool, offer more features, and might get you a higher overclock on the same chip as well.

Also, even the 965P DS3 can handle 450MHz FSB, it's not the motherboard that's holding you back.

 

[SerpentRoyal]

That MB runs HOT, very HOT! Use 1:1 memory divider and 5-5-5-15-2T timing. Bump Vcore to 1.45. You may want to bump on Vnorthbridge to 1.45.

Step up to the Abit IP35-E/IP35 if you want a cool n quiet overclocking board. Easy 488MHz. IP35-E is only $98 AR at MWAVE. There's not that much real-world difference between E4300 @ 3.2GHz and E6850 @ 3.6GHz. Not sure if you can achieve 4.0GHz with this baby.

Save your $ for the new 45nm chips. These CPUs should be able to break the 4GHz barrier without the need of fancy cooling. Abit IP35s support these chips.

 

[kingfetty]

1: Yes the P35 will let you hit 450mhz ez, I have an Abit IP35 and it has no issues at all, there are reports of this board hitting 490+ fsb.

2: Chances are you won't pull it off with air cooling but in order to give you the best chance here are a few thoughts:

a: Go with a Core 2 Duo rather than a Core 2 quad, as the quad produces much more heat and requires more voltage to keep stable.

b: Make sure you get one of the newer G0 stepping models (1333FSB), even though they have a lower multiplier, they have a higher thermal specification (another 10C) and require less power.

c: The E6850 may give you the best shot as it has a 9x multiplier and you would only need to push the FSB to 445 to hit 4ghz.

d: Look into everything to get extra heat dissipation (arctic silver, good airflow case, Thermalright Ultra-120 eXtreme, keep ambient temp low, etc.)

 

[firewolfsm]

All P35 boards run pretty cool, no matter who made it, it's still the same chipset.

 

[chiew]

hi guys,
thanks for the posts .
i have read about penryn, and i know that it will likely yield higher speeds, be faster per clock, and also have 50% more L2 cache. it seems like it is worth waiting for. it is also in the same TDP range as current core 2 duos (65W), yet i think i read that the new materials they are using will be generally more efficient and therefore cooler. it looks like penryn will be coming out later this year, and will be out in larger volumes early next year. i'd rather not take on an early adopter approach, since after awhlie prices settle and the bugs get worked out. if i do wait until early next year, i might as well wait a bit more and get the nehalem. i'm basing this build off it being my senior year in college, and thus it might be easier to build my next computer wherever i end up, instead of having to ship a beast to my next location.

i guess one question i have is:
how old are the 945P, 965P, 975X, as in when did they come out?
what is the likelihood of a P35 or X38 successor before the mandatory new chipset comes out to support nehalem?

 

[chiew]

additionally one of the reasons i wanted to get a better board/processor combo is i'm running ddr2 8500, which is rated at up to 1066mhz.
right now since my bus is 300mhz, i'm running my ram at 600 (since i linked the two)
if i were able to get to 450mhz bus, i would be running my ram at 900, much closer to what they were supposed to get. a jump from 600mhz ram to 900mhz ram will yield some advantage, no?
just a little peev of mine that my ram is getting bottlenecked

 

[bryanW1995]

you can alter the cpu/memory ratio in bios. at 300, you should run 1:1.66. You might even be able to get away with the mythical 1:2, though I haven't gotten my 1066 ballistix over 1161 iirc. Make sure to set timings at 5-5-5-15 2t if you try for 1200. That 40mm fan sounds pretty annoying, a p35 ds3r or one of the abits will definitely be an upgrade and allow you to jump into penryn next year for minimal cost.

 

[chiew]

thanks bryan,
i have noticed that you can use something other than the 2 multiplier for the ram, and i was tempted to do 2.66 since that would give me ram at 800mhz (800mhz is the ram "limit" of my mobo). however, i have heard that running the ram at faster than 1:1 does not provide any benefit in arenas outside of the pi calculator. if the ram is running faster than the bus, doesn't it all jam up at the bus (as in the bus is the slowest worker in the assembly line)?

 

[aigomorla]

Originally posted by: kingfetty
1: Yes the P35 will let you hit 450mhz ez, I have an Abit IP35 and it has no issues at all, there are reports of this board hitting 490+ fsb.

2: Chances are you won't pull it off with air cooling but in order to give you the best chance here are a few thoughts:

a: Go with a Core 2 Duo rather than a Core 2 quad, as the quad produces much more heat and requires more voltage to keep stable.

b: Make sure you get one of the newer G0 stepping models (1333FSB), even though they have a lower multiplier, they have a higher thermal specification (another 10C) and require less power.

c: The E6850 may give you the best shot as it has a 9x multiplier and you would only need to push the FSB to 445 to hit 4ghz.

d: Look into everything to get extra heat dissipation (arctic silver, good airflow case, Thermalright Ultra-120 eXtreme, keep ambient temp low, etc.)

wow, with someone with such little post, i applaud your wise comments.

I second this statement all the way, however, MX-1 is better then AS.

Originally posted by: firewolfsm
All P35 boards run pretty cool, no matter who made it, it's still the same chipset.

And P35 chipsets run hot if you amp the voltage which is required for higher FSB. So i dont know where your getting P35's run cool.

 

[bryanW1995]

yeah, I was originally gunning for 4 ghz but I now realize that was just a pipe dream. I'm currently at 3.68 at 1.57vcore and can't run orthos for 5 min. I'm a little scared to bump vcore up, even if it is only for long enough to run orthos 12+ hrs to prove system stability...

 

[chiew]

bryan, could that be because of your multi (ie if you had a 9x you might be able to hit 4ghz?)?

btw that leads me to another question (haha im full of questions)
is it that the clock speed requires boosts to vcore, or that the fsb requires the boosts to vcore?

for ex:
E4300 (9x multi) 9x300 = 2700mhz
E4400 (10x multi) 10x300 = 3000mhz
would they require the same voltages (meaning that the FSB was the factor in determining how much voltage to give the core), or would the E4400 require more voltage (meaning that the clcok was the factor determining how much voltage is needed).

 

[bryanW1995]

all things equal it is easier to get a higher oc with a higher multiplier, but you still need enough voltage to keep the chip stable. For example, I cheaped out and got the e6750 instead of e6850 b/c I wanted to save $$ for a future penryn upgrade and I figured that I would be able to hit 8x500 on my abit ip 35 pro just as easily as I could hit 9x 445. Unfortunately, I have been unable to get over 7x468 or 8x 460. I can't even post above 7x468. I'm currently running orthos at 8x460 at 1.60v with temps on both cores at 63c.

To answer your question, you don't need to change vcore to up the fsb unless you get too high over your stock speed. You need to up the nb voltage to up fsb over a certain point, however.

 

[SerpentRoyal]

Originally posted by: bryanW1995
all things equal it is easier to get a higher oc with a higher multiplier, but you still need enough voltage to keep the chip stable. For example, I cheaped out and got the e6750 instead of e6850 b/c I wanted to save $$ for a future penryn upgrade and I figured that I would be able to hit 8x500 on my abit ip 35 pro just as easily as I could hit 9x 445. Unfortunately, I have been unable to get over 7x468 or 8x 460. I can't even post above 7x468. I'm currently running orthos at 8x460 at 1.60v with temps on both cores at 63c.

To answer your question, you don't need to change vcore to up the fsb unless you get too high over your stock speed. You need to up the nb voltage to up fsb over a certain point, however.

Your problem is the chip. I have a lowly E6320 @ 488MHz with only 1.5Vcore. Most of these E6xx0 are good up to 3.5GHz on air. Very few are stable at +3.7GHz on air.

 

[SerpentRoyal]

Originally posted by: chiew
bryan, could that be because of your multi (ie if you had a 9x you might be able to hit 4ghz?)?

btw that leads me to another question (haha im full of questions)
is it that the clock speed requires boosts to vcore, or that the fsb requires the boosts to vcore?

for ex:
E4300 (9x multi) 9x300 = 2700mhz
E4400 (10x multi) 10x300 = 3000mhz
would they require the same voltages (meaning that the FSB was the factor in determining how much voltage to give the core), or would the E4400 require more voltage (meaning that the clcok was the factor determining how much voltage is needed).

All a big gamble when you overclock a CPU. 50% of these chips should be able to hit 3.0GHz at default Vcore. A higher multi CPU will result in less stress on the MB and RAM.

 

[Diogenes2]

chiew,

What BIOS are you running ?

I tried some later BIOS's on my DS3 ( F7, F11 ) and lost my overclock ..

With F4, all is good ..

 

[bryanW1995]

Originally posted by: SerpentRoyal

Originally posted by: bryanW1995
all things equal it is easier to get a higher oc with a higher multiplier, but you still need enough voltage to keep the chip stable. For example, I cheaped out and got the e6750 instead of e6850 b/c I wanted to save $$ for a future penryn upgrade and I figured that I would be able to hit 8x500 on my abit ip 35 pro just as easily as I could hit 9x 445. Unfortunately, I have been unable to get over 7x468 or 8x 460. I can't even post above 7x468. I'm currently running orthos at 8x460 at 1.60v with temps on both cores at 63c.

To answer your question, you don't need to change vcore to up the fsb unless you get too high over your stock speed. You need to up the nb voltage to up fsb over a certain point, however.

Your problem is the chip. I have a lowly E6320 @ 488MHz with only 1.5Vcore. Most of these E6xx0 are good up to 3.5GHz on air. Very few are stable at +3.7GHz on air.

whoops, was just writing a long post and failed orthos after 2min of 1.54v x 3.60. Went down to 3.55 at 1.52v. Ran stable all night at 1.64, first for an hour on orthos and then rest of the night with seti on both cores. temps were ok when I came in this am. Aimogorla warned me about running vcore over 1.55 v on air and over 1.6 on anything but phase, but I didn't get the warning until this am. Fortunately everything is a-ok. At least I know that chip will run over 3.7! I think I'll shoot for 3.5-3.55 stable 24/7 with vcore of 1.5 or lower. I'm not disappointed any more about this chip or the mobo, I haven't heard of too many that got theirs over 3.7 even at higher volts. Now I can focus on memory then lowering all voltages...

 

[Nemesis 1]

Originally posted by: chiew
hi guys,
thanks for the posts .
i have read about penryn, and i know that it will likely yield higher speeds, be faster per clock, and also have 50% more L2 cache. it seems like it is worth waiting for. it is also in the same TDP range as current core 2 duos (65W), yet i think i read that the new materials they are using will be generally more efficient and therefore cooler. it looks like penryn will be coming out later this year, and will be out in larger volumes early next year. i'd rather not take on an early adopter approach, since after awhlie prices settle and the bugs get worked out. if i do wait until early next year, i might as well wait a bit more and get the nehalem. i'm basing this build off it being my senior year in college, and thus it might be easier to build my next computer wherever i end up, instead of having to ship a beast to my next location.

i guess one question i have is:
how old are the 945P, 965P, 975X, as in when did they come out?
what is the likelihood of a P35 or X38 successor before the mandatory new chipset comes out to support nehalem?

Well chiew I like your reasoning. About the early adaptors . But heres my view on that.

My wife has C2D I don't. I want X38 chipset and penryn . Even tho Nehalem is right around the corner. My wife will buy Nehalem I will wait for Nehalem C(late adaptor) that should use 32nm(eary adapter) 3D gates.

So with intel early adaptor depends on what your talking about . The processs or the arch. I am going for the mature Arch(New Process) . Wereas my wife is going after the early arch on mature process.

Tick tock works for us.

 

[Arkaign]

4Ghz is not realistic with current yields and air cooling. Check back with Penryn and other updates as they come.

Btw, C2D/C2Q @ 3Ghz and above is ridiculously fast for all current apps/games. Even with 8800GTX SLI, you're still video card limited for the most part.

Achieving 4Ghz might be a nice bragging point, but 24x7 stability @ 3.5Ghz is the superior option at this time.

 

[Nemesis 1]

This is what I want . Water cooled . 2x/4xcores running at 4ghz+(penryn) This should also have 2xFSB.

I don't care what the dam thing cost. I am going to preorder from intel. By the Time its out well have low latency DDR3 running at FSB of 2200+. Got to have this.


http://www.xtremesystems.org/f...howthread.php?t=153683

 

[chiew]

hello guys,
sorry for the late reply. i did not check this thread this morning since i have been working.

serpentroyal: so what you are getting at is that the E4300 doesn't perform as well as the E6320 because of the binning that intel is doing? perhaps they caught on to many people overclocking their E4*** and got angry . in any case i thought that the E4300 would do as well as the E6300 (esp since 9x multi compared to 7x), and since i read somewhere that 2mb L2 cache cores run cooler than 4mb L2 cache cores.

Diogenes2: I am running BIOS F12. Good point on the bios update having an effect on the OC, I will try to check on that. I had a feeling that it might be the bios, since i haven't seen many people posting OCs on the F12. Is there a place that I can find old BIOSes? Unfortunately I don't have a floppy disk drive, so when I updated my BIOS i did gigabyte's online (in windows) update.

Nemesis 1: looks like you and your wife will be able to enjoy each other's computers . With Intel's new tick tock strategy, it makes it hard to define what is early adoption. I think part of the reason my Pentium 4 3Ghz machine lasted me awhile was that Intel didn't bring out anything new in the years I had it (I bought it in 2004). Come to think of it, our family was running a Pentium 4 1.7Ghz in 2001! That means that Intel's been using P4 for at least 5 years. No chance of them using the same architecture these days. Guess you gotta just buckle down with one architecture and fab, or spend lots of money (which is probably what Intel hopes) keeping up with their tick tock.

As for Nehalem, it sounds really nice with the architecture update. I won't be using their integrated GPU chip, since I heard it will be mid-low end. I also hear they're getting rid of the FSB in Nehalem, which is kinda strange. Well it all depends on what's gonna get me to bite. And I already see the merits of waiting for things to mature, since not long ago they were selling X6800 at $1000 and now they're selling the E6850 at $290 (albeit with locked multi). Not long ago my 7800GT OC was $300, and now its going on ebay for $120. Things are always improving, so its hard to decide when to jump in. But gotta jump in with no regrets. Good and bad sides to Moore's law I guess.

At this point, quad core sounds cool, but would probably be overkill for me. I also like to wait for things to be usable, and it sounds like most programs won't be multithreaded for quad core use in awhile (since the average user is probably just buying into dual cores). I also read that Intel will be bringing an equivalent of "hyperthreading" back, so that will give each core the ability to process two threads. Interesting. The most CPU intensive things I do are gaming and also photoshop work. I like being a semi-enthusiast, but it comes with a pricetag

 

[SerpentRoyal]

There's some speed binning with the lower-priced CPUs. Same core architechture. E6xx0 chips are soldered to the heat spreader. E4xx0 chips use thermal paste. Less cache. Less efficient heat transfer. So they tend to balance each other.

 

[VirtualLarry]

Originally posted by: chiew
thanks bryan,
i have noticed that you can use something other than the 2 multiplier for the ram, and i was tempted to do 2.66 since that would give me ram at 800mhz (800mhz is the ram "limit" of my mobo). however, i have heard that running the ram at faster than 1:1 does not provide any benefit in arenas outside of the pi calculator. if the ram is running faster than the bus, doesn't it all jam up at the bus (as in the bus is the slowest worker in the assembly line)?

There is some definate truth to that statement, and is likely the reason why AMD64 CPUs have 1:1 as the fastest RAM speed, with slower numbers being divisors.

I've pondered why Intel chipsets have 1:1 as the slowest RAM speed, and have multipliers for faster speeds. It seems useless to me, unless marketing is their key goal.

There is only very marginal utility in having RAM running faster than CPU FSB. Other hardware devices can access RAM at nearly the same time as the CPU, such as the video card. So if you use a video card that can use system RAM (such as NVidia's TurboCache feature, or ATI's HyperMemory), then you may see more system efficiency by running RAM faster than FSB. Other than that, no.

If the system was running multiple CPUs, with seperate FSBs to the chipset, then there might also be some benefit. But with a single shared FSB, the FSB is the bottleneck and not the RAM. So again, no benefit to running RAM faster than FSB.

 

[VirtualLarry]

Originally posted by: SerpentRoyal
There's some speed binning with the lower-priced CPUs. Same core architechture. E6xx0 chips are soldered to the heat spreader. E4xx0 chips use thermal paste. Less cache. Less efficient heat transfer. So they tend to balance each other.

Where is there proof about this heat spreader issue? Because AFAIK, all of their CPUs are mfg/assembled the same way. It would cost them more $$$ to do certain classes of CPUs seperately. I just don't see it.

 

[SerpentRoyal]

Based on my past builds, the average overclock speed for E4xx0 is 3.25GHz. E6xx0 is 3.4GHz. Perhaps the soldered heat spreader was largerly responsible for the 100MHz increase.

How would I know this? Because I've removed the heat spreader from these chips. All E4xx0 use a grey thermal paste. All E6xx0 use low temperature solder (no thermal paste).

Dunno if you can GOOG this info.