In games I think the P4 EE will be very strong.

When you talk about having samples, do you mean engineering samples or just examples of each core?

Of the chips you mentioned, the Cedar Mill 6x1 chips are likely to perform best, due to cache size and manufacturing process. You can probably find comparison on CPU-World.com

Have you thought about comparing performance of socket 423 P4 chips as well? They are all Willamette, but it might be good to see how the chipsets compare for the same core. The move to socket 468 was just a package shrink, with pins closer together allowing a smaller package.

PhilsComputerLab wrote:

In games I think the P4 EE will be very strong.

You're probably right, the P4 EE Gallatin chips have L3 cache, might make a difference clock for clock.

debs3759 wrote:

When you talk about having samples, do you mean engineering samples or just examples of each core?

Of the chips you mentioned, the Cedar Mill 6x1 chips are likely to perform best, due to cache size and manufacturing process. You can probably find comparison on CPU-World.com

Have you thought about comparing performance of socket 423 P4 chips as well? They are all Willamette, but it might be good to see how the chipsets compare for the same core. The move to socket 468 was just a package shrink, with pins closer together allowing a smaller package.

Well. I kinda called the CPUs samples despite i don't have any Engineering samples. I have a few Socket 423 CPUs and a working Socket 423 system, but my aim is comparing clock per clock perfomance. This is why i'm using the same chipset for all CPUs. The testbed will be a based on the Intel 865G with dual channel DDR when my Socket 423 buid uses the older Intel 850 with RAMBUS. I currently jave about 120 P4 CPUs from an 1.5 GHz Willamette to a 3.8 GHz P4 670

Pentium 4's at the same clock comparison...

If you get/have all CPU's with x16/x17 multiplier, it is doable.
Why those ?
Because x16/x17 are min/max multipliers for Gallatin (PGA 478) 😀

Now :
If you ditch Gallatin : You can go as low as x14 (lowest multiplier for Prescott).
If you ditch Willamette : You can go to 800MHz FSB and 3GHz+ clock speeds without major problems (depending on cooling you got 😉).
I don't know how low FSB can be set on that LGA 775 MB, but it can be "tricky" to get those newer boards to post on 400MHz FSB.
I recommend overclocking Willamette as far as FSB will go and set those settings for all other CPUs.

Performance wise, it should be something like this (under normal conditions) :
Cedar Mill = Prescott-2M > Gallatin > Prescott = Northwood (more/less - depends on program) >> Willamette.
In your case, it is possible for Gallatin to be faster than Prescott-2M/Cedar Mill, because of larger cache (2,5MB [L2+L3] vs. 2MB [L2]), no SSE3 programs and (probably) low FSB speeds.

Also :
1) I would ditch Win XP x64 testing.
It's really pointless since there are only two CPU's cores that can do it, and they are almost exacly the same (Cedar Mill is a Prescott-2M, that's why I'm NOT counting it in that two I mentioned).
Also : Programs you listed are not x64 native, so there will be nothing to gain here from 64-bit.
Last thing to 64-bit :
Using Win XP x64 is not necessary, because where Win XP x86 won't be used, Win 7 x64 SP1 will be.
2) Few test programs to think about :
Quake III Arena (timedemo "demo001"), SuperPI XS mod (1M), Half Life 2 : Lost Coast (?), F.E.A.R. (build-in test), check out also "Cache & Memory benchmark" from AIDA64 or few ScienceMark 2.0 tests.

agent_x007 wrote:

Pentium 4's at the same clock comparison... […]
Show full quote

Pentium 4's at the same clock comparison...

If you get/have all CPU's with x16/x17 multiplier, it is doable.
Why those ?
Because x16/x17 are min/max multipliers for Gallatin (PGA 478) 😀

Now :
If you ditch Gallatin : You can go as low as x14 (lowest multiplier for Prescott).
If you ditch Willamette : You can go to 800MHz FSB and 3GHz+ clock speeds without major problems (depending on cooling you got 😉).
I don't know how low FSB can be set on that LGA 775 MB, but it can be "tricky" to get those newer boards to post on 400MHz FSB.
I recommend overclocking Willamette as far as FSB will go and set those settings for all other CPUs.

Performance wise, it should be something like this (under normal conditions) :
Cedar Mill = Prescott-2M > Gallatin > Prescott = Northwood (more/less - depends on program) >> Willamette.
In your case, it is possible for Gallatin to be faster than Prescott-2M/Cedar Mill, because of larger cache (2,5MB [L2+L3] vs. 2MB [L2]), no SSE3 programs and (probably) low FSB speeds.

Also :
1) I would ditch Win XP x64 testing.
It's really pointless since there are only two CPU's cores that can do it, and they are almost exacly the same (Cedar Mill is a Prescott-2M, that's why I'm NOT counting it in that two I mentioned).
Also : Programs you listed are not x64 native, so there will be nothing to gain here from 64-bit.
Last thing to 64-bit :
Using Win XP x64 is not necessary, because where Win XP x86 won't be used, Win 7 x64 SP1 will be.
2) Few test programs to think about :
Quake III Arena (timedemo "demo001"), SuperPI XS mod (1M), Half Life 2 : Lost Coast (?), F.E.A.R. (build-in test), check out also "Cache & Memory benchmark" from AIDA64 or few ScienceMark 2.0 tests.

I know my old Gigabyte GA-P43T-ES3G could underclock the FSB low as 400 mhz. The LGA 775 motherboard i'm planning to use is the AsRock 775i65G since is practially identical to the Socket 478 sister, the P4i65G since the only differenfce is the socket. I need to get that mobo to see how low can do the FSB, i know the P4i65GV can even go below 400 MHz FSB (100), just low as 66 (266 MHz FSB) yet is not documented, need to check the P4i65G lowest FSB as well. P4i65G officially supports 400 MHz FSB.

About the multi, I have all x16 multi P4s except for the Willamette 1.6, Galaltin 3.2 EE and P4 641, P4 541
I have all x17 except for the Northwood 3.4 GHz, 1.7A (HTless), 3.4 Ghz Galatin EE and P4 651, P4 551

Sadly, i still to find a reasoneable priced Gallatin since the Chinese ebay stores used to have some of them, but they went out of stock and the only 3.2 GHz SL7AA P4s are abount 50-60 € + 50-150 € shipping (they used to be about 40 € with free shipping), there are an SL7CH Gallatin for 107 € +5 € shipped, but still too much for me. I even saw a Pentium Extreme Edition 965 much cheaper than that (39 €)
http://www.ebay.com/itm/INTEL-Pentium-4-Extre … S8AAOSwQaJXQ7tB

About the 64 bit tests, i'm still not sure to do or not, also there are 64 bit Prescott-1M (P4 5x1 models)

Imho, the best way to test P4 is the "several step" method.
Use 3.0/3.06 Ghz to test all P4s except Willamettes and EEs.
If you except Cedar Mills - you can use 1 frequency for all P4s - 2.8 GHz.
Then just compare WIllamette 2.0 GHz with Northwood 2.0A.