Overclocking Pentium 4 540 - no change?!

[trodas]

Since my Super Pi 1M results on 3.2GHz Pentium 4 540 are somewhat disappointing 45.172 - 44.157sec:
...

So then I tried to overclock it a little. It is 200x16 = 3200MHz by default. The only one CPU tweak the MSI PM8M3-V mobo offer is the - after I unclocked it in the bios - FSB. 210MHz FSB post, yet failed to boot into Windows 2000 SP4, claiming that ACPI fail. Lol. 208MHz worked, booted and run SuperPi and CPU-Z validations:
CPU-Z Validator 3.1
CPU-Z Validator 3.1

So, a 3328MHz compared to 3200MHz = I hoped to gain something. But the SuperPi result was disapointingly low - 44.078sec...!!!


So, all the overclocking get me from 44.157sec to 44.078sec :o
I know that this is just a little 104% overclock, but 4% from 44.157sec is 1,76628 sec - and that I hoped to gain. Instead I gained 0,079 sec.

1,76628
0,079

Hmmm. This is disproportional. Anyone can explain me this, or WTF is going on?!

[Bond007]

Honestly I wouldn't expect much from a small overclock on an old CPU. Nothing scales perfectly either. You almost always require a much greater increase in frequency % than what a specific program will increase in performance in %.

[Dead Things]

Netburst was a notoriously poor performer in SuperPi anyway. It takes kinda a lot to make much of a dent in performance. Having said that, the LGA775 P4's tended to have a fair bit of overclocking headroom - so I'm a bit surprised that you were able to get so little out of it at stock volts.

[trodas]

Bond007 -

Nothing scales perfectly either.

True. No point of arguing that the "overclock" is laughable and no point of perfect scaling either. However is not SuperPi only CPU - dependant app? With higher clock there almost automaticaly should be scalled up performance.
Not perfectly as the clock get higher, but at least remotely close reassembling the incrase in clock.
There should not be a 22,6 difference between increased Mhz and decreased time. It just does not make slight sense to me.


Dead Things -

Netburst was a notoriously poor performer in SuperPi anyway.

Quite obviously 3.3GHz and still cannot get under 44 sec? Come on...

It takes kinda a lot to make much of a dent in performance.

Yep, but... at least I would expect some corelation of the time and clock. This is like completely random...

Having said that, the LGA775 P4's tended to have a fair bit of overclocking headroom - so I'm a bit surprised that you were able to get so little out of it at stock volts.

The "mainboard" used did not allow increase of Vcore voltage. It allow only changes in ram and AGP voltages, witch must be first unlocked in ModBin, because with defalt bios these settings are invisible.

The increase is indeed laughably small. Rams was tested at 216MHz with tight timings some time ago and should not have a problem at 210MHz. The poor overclock might also have something to do with the AGP/PCI clocks being tied to the FSB, IIRC. Definitively there is no way to set them or lock them.

It was just a attempt to check out, how much I can overclock and how much this will change the SuperPi performance. For serious P4 overclocking it would need a good P4 LGA 775 mainboard. I was just playing there, during experimenting with another mobo that used to run stable at relative high overclocks and now it did not give a go at all...
So I sort of hoped that someone have similar experience or not increasing performance in SuperPi on the Presscot CPUs, so it can tell me that this is normal, because it just did not feel right to me.

[trodas]

LGA775 P4's tended to have a fair bit of overclocking headroom - so I'm a bit surprised that you were able to get so little out of it at stock volts

Confirmed that this is the board issue. By - another CPU. Friend send me a 2x Pentium 4 650 - a LGA 775 3.4GHz chip with 2MB L2. Except my confusion that I need to put some load on the chip to stay at 3.4GHz (it switch to 2.8GHz ASAP (x14) when there is no load... at x17 it should be, but only when there is some work to do...) I'm fairly happy with it. Now the problem witch overclocking:

Put a 208MHz FSB and got the ACPI bios failure BSOD during booting.
Put a 206MHz FSB and got it working fine:

3.5GHz is the most it can do:

CPU-Z Validator 3.1
trodas`s CPU Frequency score: 3500 mhz with a Pentium 4 3.4 GHz Prescott

Therefore by another independent CPU it is verified, that the FSB increases there could be only very slight It makes me wonder, if using Pentium 4 670 (3.8GHz chip) would even allow me any overclocking. Perhaps 2MHz FSB up...?

For overclocking LGA 775 chips - I got a collection now, lol:

...is necessary to use some respectable board, like Asus P5E64 WS Evolution one:
ASUS P5E64 WS Evolution(X48) mini Review

That, combined with good cooling and perhaps even IHS removal, will allow some serious overclock
The MSI PM8M3-V is really a low-end mobo, happy to even work at such speeds, IMHO.

[trodas]

Looks like the minimum gain in SuperPi is somewhat typical for P4:

Because I see absolutely minimal gain from Pentium 4 540 at 3.2GHz 1MB L2: 48.45.890 to 48.04.047 with Pentium 4 650 at 3.4GHz 2MB L2 :( Suxx!



Twice the L2 and 200MHz more for what? 41sec in 48min times?! WTF!

[sswilson]

It's been an awfully long time, but IIRC super pi performance back in the P4 days was strongly influenced by memory and bus speeds as well cpu clock speeds. Any chance your OC has negatively affected either of those?

[b1lk1]

DDR2 and memory speeds are definitely the problem. I remember slapping P4's around with my 3200+ back in the day running OCZ VX ram. I had that stuff running 277MHz CAS2, with 4.1V of course, LOL!

[trodas]

sswilson -

super pi performance back in the P4 days was strongly influenced by memory and bus speeds as well cpu clock speeds

That is why it make me wonder, how the performance gains are so minimal. But it is consistent in these minimalistic gains when O/C:
P4 540 - 128MHz more (200x16 to 208x16) = 0.079sec gained at SuperPi 1M test
P4 650 - 200MHz more over 540 (200x17) + L2 (44.157 to 41.297) = 3sec gain at SuperPi 1M test
P4 650 - 200MHz more over 540 (200x17) + L2 (48min 45.890 to 48min 04.047) = 41sec gain at SuperPi 32M test

Looks like the 1M test is way more affected by the ram itself that the 32M test, so therefore there is more noticable (%) gain. But when overclocking the P4 650, the results will be likely similary minimal gains on performance...

Any chance your OC has negatively affected either of those?

Zero. All the "overclocking" was about increasing the FSB, no relaxed timings, nothing like that. At the 2-3-2-7 settings are these rams tested up to 216MHz, so... not need to worry at 206 or 208MHz



b1lk1 -

DDR2 and memory speeds are definitely the problem.

Yep, I wish I can use DDR2. Sadly, only DDR(1) are supported by this mobo.

I remember slapping P4's around with my 3200+ back in the day running OCZ VX ram.

That is quite possible, as P4 is definitively not effective per clock at all. I wonder, how the P4 670 CPU could looks by the MHz per seconds effectivity on SuperPi A good glimpse of that is shown up there:
Superpi 32m Benchmark Competition

I had that stuff running 277MHz CAS2, with 4.1V of course, LOL!

Cas2 at 277MHz? What chips you used, lol?
I looking for some ram for my S939 rig, about 2x1024 so I can have it for normal usage, when I make it work... and of course I would like if it can be clocked faster that 200MHz default... In fact, I having some settings problems and if you check there:
Superpi 1m Benchmark Competition
I go to the 4th place by running rams at only 200MHz and 2,5-3-3-7, witch is nothing exactly fast...

[trodas]

sswilson -

Any chance your OC has negatively affected either of those?

So, instead of just proclaiming, that the little so called "OC" did not negatively affected performance of chipset/ram, I made a MaxxRAM test:



64.7 points before (200x17), 70.6 points after (206x17) - I see a notable increase of performance! FSB overclocking rulez :D