Pentium 4 3.2@4.5 Stable. If I get the 3.6 do you think I'll get 5GHz stable?

[JeremyF50]

Hi guys,

First post here, so, hello!

I'm currently running a 3.2GHz Pentium 4 Cedar Mill, the 65nm core. This is the final stepping, so it's a 65W chip rather than 86W.

With it's default multiplier and an FSB of 1125mhz I've managed 4.5GHz prime95 stable for 8 hours and even fired up Crysis. Haven't played with the memory much yet, just set it to 5:4 for a 900mhz clock.

I was able to get it somewhat stable at 4.6-4.7 but I haven't played with it long enough yet, and 4.5 seems to be playing nicely.

I'm only using a crap 3rd party heatsink which is probably a little better than the stock Intel coolers. I'm using Enthusiast DDR21066 with an XFX Nforce 780i Sli board and I've got a Noctua NH-D14 in the cupboard to use later.

If I purchase the 3.6GHz Cedar Mill D0 chip which has a multi of 18 rather than 16, do you think I'll manage 5GHz with that?

I don't know what is limiting the current chip. Maybe it doesn't like the higher FSB, It's not cool enough, I don't know. 1112mhz is the FSB I'd need out of the 3.6GHz chip (for 5GHz), and this 3.2GHz one is currently stable with 1125 (4.5GHz). I know the board can handle these FSB speeds being designed for 1333 (1600 maybe?) for the Core 2 series.

By the way, with a 9800GX2 in this, it runs Crysis on High (XP/DX9) at 1080p very playable! wprime 32m was still 10 seconds slower than my dual Pentium 3s at 1.74GHz though

Thanks guys

 

[blotto]

Why on earth would you spend any money at all on new P4 chip? You could at the worst, drop in a cheap q6600 and play around with overclocking that.

 

[JeremyF50]

Basically been there and done that. Trying something I haven't before. I've still got Core 2 chips laying around that I use in this board too. I pickup a lot of older chips for free that are getting thrown out at work, like this 3.2GHz. The 3.6 is cheap and I'll probably buy it anyway, just wondering what to expect.

 

[masteryoda34]

Today I learned you can apparently still buy Pentium 4s.

OP: Don't buy a Pentium 4.

 

[Blain]

I would not spend anything on upgrading a P4 platform simply to eke out a few more Mhz that don't really matter.

But if you can't be talked down, buy a Presler rather than Cedar Mill core.
Presler 960's are selling pretty cheap.

 

[JeremyF50]

It's just for hobby sake, not my main machine. Something to play with, see what I can get it to do.

 

[Maximilian]

Basically been there and done that. Trying something I haven't before. I've still got Core 2 chips laying around that I use in this board too. I pickup a lot of older chips for free that are getting thrown out at work, like this 3.2GHz. The 3.6 is cheap and I'll probably buy it anyway, just wondering what to expect.

:thumbsup:

Ive done stuff like that, experimentation is fun!

My guess would be no, the 3.8ghz P4's were really at their limit from what i recall so i dont think any of them including the 3.6's will reach 5ghz (not without crazy liquid nitro cooling or whatever). Worth a shot though!

 

[JeremyF50]

The 3.8 was a 90nm prescott core (115w TDP) but the 65nm cedar mill core was 86, down to 65 with the D0 stepping.

This 3.2 is a Cedar Mill D0 like the 3.6 (fastest Cedar Mill Pentium 4) that I'm looking at.

Of course there is the 533mhz/512kb Celeron variant with a much higher multi - but I'll leave that one for another day.

 

[NTMBK]

Don't listen to the naysayers. You can have fun tinkering with old hardware just as much as you can tinkering with brand spanking new hardware- and a P4 is a hell of a lot cheaper to get hold of than a 3770k!

Come on, how many of you guys really tweak and overclock your PCs to get an imperceptibly better framerate in games? Or to render a scene that 5% faster? We do it because it's fun.

 

[JeremyF50]

Thanks Maximilian and NTMBK for the support, guys

Now does anybody care to speculate on if the 3.6 will manage the big 5 given the info in 1st post?

 

[Blain]

Now does anybody care to speculate on if the 3.6 will manage the big 5 given the info in 1st post?

Stable at 5GHz on the hardware in 1st post... no

 

[tweakboy]

Whatcha doing goofing off with old hardware ehhe.

 

[JeremyF50]

Stable at 5GHz on the hardware in 1st post... no

Ok, no worries. Care to elaborate why? Thanks for the response

 

[Magic Carpet]

JeremyF50,

Just get one and overclock it, then come back with results. Good fun

EDIT: I also have a D0 stepping Cedar Mill, it's a good chip actually. Power-wise it's not far from Conroe and runs super cool. Need a better mobo to try oc that. Still waiting.

 

[Tuna-Fish]

Ok, no worries. Care to elaborate why? Thanks for the response

IIRC, 65nm P4:s tended to be limited by heat when OCing. I suspect that a beefier cooling solution will give you more frequency, but higher multipliers won't.

How are your temps presently?

 

[YBS1]

It's been so long I could have forgotten, but I don't recall anyone hitting 5GHz stable on air cooled Pentium 4's. I'm wanting to say mine was stable at ~4.6 but that was on water. Mine was a Northwood though, so YMMV.

 

[JeremyF50]

@Magic Carpet: Have got one on the way :biggrin:

@Tuna-Fish: Idling in the 30s and Prime95 in the 80s - Not a great heatsink though. Will install the Noctua NH-D14 when I install the 3.6GHz chip. It's being posted next week to me.

@YBS1: A 4.6GHz Northwood would be godlike. People were lucky to get those things to 4ghz, they were practically at their limit using 130nm lithography. Although Northwood was better clock for clock than Prescott (and thus Cedar Mill) at many tasks, the Prescott performance scaled better with higher clocks, so they're probably about even in the high 4.X area. In some cases it's the opposite though, because the Northwood outscales the Prescott in a few areas.

Link in case you're interested: http://www.sudhian.com/content/?p=1468

 

[YBS1]

Like I said, it's been so long I could be just full of crap. That number sounds right in my head, but I can't find any old screenshots just quickly browsing through my misc junk folder.

 

[YBS1]

LOL I think I just had a case of time really distorting my memory. Did a post search on here, 3.6GHz seems more likely. haha

 

[JeremyF50]

LOL I think I just had a case of time really distorting my memory. Did a post search on here, 3.6GHz seems more likely. haha

Haha yeah. Northwood was quite a while ago!

Ok so I found out my memory divider was set to high (5:4) so I lowered that to 3:2 and 4.6Ghz was fine in prime95 for 30 minutes, moved on to 4.7Ghz and got bored after 10 minutes so now I'm trying for 4.8GHz with 2:1 and had do give the northbridge a slight bump in voltage too.

Silicon is probably degrading as I type this but meh - free CPU I saved from the bin at work

Prime95 for 9 minutes so far.

 

[VirtualLarry]

superpi at those speeds, for the lolz?

 

[Ketchup]

Bets are the same chip (Cedar Mill) that shipped with a higher clock would yield the same, or maybe even less overclock. Since it's the same core, I don't see a chip Intel decided to label higher on that particular day would give you higher than the stellar overclock you have already experienced (aside from a cases of impressive luck).

 

[JeremyF50]

Bets are the same chip (Cedar Mill) that shipped with a higher clock would yield the same, or maybe even less overclock. Since it's the same core, I don't see a chip Intel decided to label higher on that particular day would give you higher than the stellar overclock you have already experienced (aside from a cases of impressive luck).

You're likely correct.

Larry: I'm not familiar with superpi, i am however with wprime. Is it multithreaded? I haven't optimized the memory in this yet. I'll post a superpi number shortly for you.

EDIT: It appears it's not. There's some HyperPi front end for it to run multiple instances though? I guess superpi won't use the hyperthreading on this CPU.

 

[JeremyF50]

Didn't realise the Super PI benchmark would take so long lol. Stopped it after Loop 2 finished.

45s initial value finished
2m 59s loop 1 finished
5m 33s loop 2 finished
perhaps this gives you some indication, I don't know.


Strangely, I just got 168.6 seconds in wprime 32m. Last time I ran it (I believe it was 4.5GHz) it was 68 seconds, 10 seconds slower than my OC'd dual Tualatins. I assume it's because of the change in memory divider. I might have a play with that next.

EDIT: Ok I couldn't get it to boot with 1:1 but 5:4 worked. got 88.25 seconds on wprime 32m. I guess the 68 second time was due to 1:1 at 4.5GHz due to more memory bandwidth. If it's memory bandwidth intensive, stuffs me how my dual Tualatins got such a good score. I suspect the best result I'd get with Super PI would be at the highest clock I can remain at 1:1 with.

 

[pyjujiop]

You have more patience than I do, sir. I love tinkering around with old hardware, but the one line of CPUs I will not touch are anything NetBurst. I fix a lot of computers for a lot of friends, but if they ask me to work on a P4, I tell them to get a new computer. Fortunately, most of them have by now.