supercordo wrote on 2025-06-07, 17:00:If you edit old post I won't see it.
Ah ok, well here are the results with the CPU at 3570.
2000 - 19682
2001 - 16295
2003 - 5487
Some new scores with my P4 at 3570. I tightened up the memory timings in the bios a bit and did a slight overclock of the graphics card.
2000 - 20095
2001 - 16512
2003 - 5646
I don't think much more can be squeezed out of this particular setup, but this was fun to play around with for a bit.
pete8475 wrote on 2025-06-08, 05:56:Some new scores with my P4 at 3570. I tightened up the memory timings in the bios a bit and did a slight overclock of the graph […]
Some new scores with my P4 at 3570. I tightened up the memory timings in the bios a bit and did a slight overclock of the graphics card.
2000 - 20095
2001 - 16512
2003 - 5646I don't think much more can be squeezed out of this particular setup, but this was fun to play around with for a bit.
Updated!!
I think im going to fire up my Socket 423 again.
Question : Are Prescott's with 1M of L2 and Prescott's with 2MB of L2 counted separately (775 only thing), or does Prescott 2M counts as Cedar Mill ?
agent_x007 wrote on 2025-06-13, 15:52:Question : Are Prescott's with 1M of L2 and Prescott's with 2MB of L2 counted separately (775 only thing), or does Prescott 2M counts as Cedar Mill ?
We can go by what CPUz says.
^Honestly, I think it's not exactly fair to do both -1M and -2M versions under one banner (just because CPU-z says "Prescott" under codename).
Maybe if result separation isn't possible, consider at least adding [-1M] or [-2M] near submission's name to indicate which version was tested ?
agent_x007 wrote on 2025-06-13, 17:17:^Honestly, I think it's not exactly fair to do both -1M and -2M versions under one banner (just because CPU-z says "Prescott" under codename).
Maybe if result separation isn't possible, consider at least adding [-1M] or [-2M] near submission's name to indicate which version was tested ?
Cedar Mill's are 65nm. Prescotts are 90nm.
supercordo wrote on 2025-06-13, 17:41:Cedar Mill's are 65nm. Prescotts are 90nm.
Both 5800X and 5800X3D are 7nm.
What's your point ?
agent_x007 wrote on 2025-06-14, 17:54:supercordo wrote on 2025-06-13, 17:41:Cedar Mill's are 65nm. Prescotts are 90nm.
Both 5800X and 5800X3D are 7nm.
What's your point ?
And both 5800x and 5800X3D both use the Vermeer core. So whats your point?
supercordo wrote on 2025-06-14, 20:57:And both 5800x and 5800X3D both use the Vermeer core. So whats your point?
My point (as mentioned previously) is that grouping processors in one table because CPU-z name is unfair.
It gives users of Prescott-2M an unfair advantage when they compete with users of Prescott-1M core.
Clearly marking which core of Prescott is used, can solve this by showing what is fastest with 1M and 2M of L2 respectively (for example using "-2M" or "-1M" near users name).
Unless you don't care about this "minor" problem ?
At which point, I just give up.
Should I give up ?
agent_x007 wrote on 2025-06-14, 21:28:My point (as mentioned previously) is that grouping processors in one table because CPU-z name is unfair. It gives users of Pres […]
supercordo wrote on 2025-06-14, 20:57:And both 5800x and 5800X3D both use the Vermeer core. So whats your point?
My point (as mentioned previously) is that grouping processors in one table because CPU-z name is unfair.
It gives users of Prescott-2M an unfair advantage when they compete with users of Prescott-1M core.
Clearly marking which core of Prescott is used, can solve this by showing what is fastest with 1M and 2M of L2 respectively (for example using "-2M" or "-1M" near users name).Unless you don't care about this "minor" problem ?
At which point, I just give up.
Should I give up ?
Coming form the person who has the highest score with a $700+ Intel pentium Extreme Edition.
Its already a lot of work to update the scores when someone posts. Im not going to do even more work. It will stay the way it is with the scores sepperated by core name and socket.
Not sure how me having an Pentium Extreme should matter from results fairness point of view, but... OK ?
As for response : That's well within your right as OP for this list.
Hi,
I would like to ask a favor from Pentium 4 owners. 😀
Here you can find my benchmark that can run even on Win98 when at least Directx6 is installed (but of course it also works on Win2000/XP).
x86 microarchitecture benchmark (MandelX)
Please, run the benchmark and upload the results.
Pentium 4 results are completely missing so far and I would like to see how Pentium 4 compares to previous and next generations.
Thanks in advance !
Falcosoft wrote on 2025-06-14, 23:16:Hi, I would like to ask a favor from Pentium 4 owners. 😀 Here you can find my benchmark that can run even on Win98 when at least […]
Hi,
I would like to ask a favor from Pentium 4 owners. 😀
Here you can find my benchmark that can run even on Win98 when at least Directx6 is installed (but of course it also works on Win2000/XP).
x86 microarchitecture benchmark (MandelX)Please, run the benchmark and upload the results.
Pentium 4 results are completely missing so far and I would like to see how Pentium 4 compares to previous and next generations.
Thanks in advance !
How do we upload them If our P4s aren't on the internet?
Here is a Prescott 3.4ghz 1m cache
Dothan Burger wrote on 2025-06-14, 23:45:Falcosoft wrote on 2025-06-14, 23:16:Hi, I would like to ask a favor from Pentium 4 owners. 😀 Here you can find my benchmark that can run even on Win98 when at least […]
Hi,
I would like to ask a favor from Pentium 4 owners. 😀
Here you can find my benchmark that can run even on Win98 when at least Directx6 is installed (but of course it also works on Win2000/XP).
x86 microarchitecture benchmark (MandelX)Please, run the benchmark and upload the results.
Pentium 4 results are completely missing so far and I would like to see how Pentium 4 compares to previous and next generations.
Thanks in advance !How do we upload them If our P4s aren't on the internet?
Here is a Prescott 3.4ghz 1m cache
It's not a problem, thank you for the result!
With your prior permission I uploaded the result. Tell me if you have problem with this:
http://falcosoft.hu/mandelx_benchmark_results.php
@Edit:
BTW, It's interesting that the SSE result is virtually the same as the SSE2.
The SSE2 time result is normal in the sense that it is about half of the FPU result (2 vs. 1 pixel per round) but the SSE time result should be about half of the SSE2 ( 4 vs. 2 pixels per round).
It seems the problem is the double amount of used 'CMOV' instructions. CMOV was introduced with the P6 (Pentium Pro) and it is generally faster than traditional branches. But it seems not on the Pentium 4.
No other microarchitectures are affected by this so far.
Hello,
To my surprise I obtained the attached scores for 3dmark00 and 01 and a 4coredual-vsta…
I was just testing it but from the chart it seems significant.
4Coredual-vsta
Pentium 4 540 FSB 214Mhz
1024mb ddr1 Dual
GeForce 6800LE unlocked default clocks
Win2000 sp4
3dMark01: 17172
3dMark00: 16375
Falcosoft wrote on 2025-06-15, 04:13:It's not a problem, thank you for the result! With your prior permission I uploaded the result. Tell me if you have problem with […]
Dothan Burger wrote on 2025-06-14, 23:45:Falcosoft wrote on 2025-06-14, 23:16:Hi, I would like to ask a favor from Pentium 4 owners. 😀 Here you can find my benchmark that can run even on Win98 when at least […]
Hi,
I would like to ask a favor from Pentium 4 owners. 😀
Here you can find my benchmark that can run even on Win98 when at least Directx6 is installed (but of course it also works on Win2000/XP).
x86 microarchitecture benchmark (MandelX)Please, run the benchmark and upload the results.
Pentium 4 results are completely missing so far and I would like to see how Pentium 4 compares to previous and next generations.
Thanks in advance !How do we upload them If our P4s aren't on the internet?
Here is a Prescott 3.4ghz 1m cache
It's not a problem, thank you for the result!
With your prior permission I uploaded the result. Tell me if you have problem with this:
http://falcosoft.hu/mandelx_benchmark_results.php
Yes you can use it.
Scores updated
According to Agner Fog's instruction tables the CMOV instructions were further slowed down on Prescott compared to earlier Willamette/Northwood P4 CPUs.
In case of Willamette/Northwood the latency is 6 clocks (compared to others it is already very high) while on Prescott it is 9.5 clocks!
This can explain the poor SSE results but it could be very interesting to see a Willamette/Northwood result to confirm/refute the theory.
If someone has a Willamette/Northwood please, help to investigate this issue further.
In case you cannot run Windows there is already a DOS version of the benchmark, too:
https://falcosoft.hu/manbench.zip
The Windows version can still be downloaded from here:
https://falcosoft.hu/otesz/mandelx.zip
If you cannot or do not want to upload your result sharing it here is completely OK.
Thanks in advance!
PS: if you do not want to spam this topic with the results you can share it here (it's a dedicated topic)
x86 microarchitecture benchmark (MandelX)
I'm sorry disturbing this topic but I feel the end result of the above problem is somewhat relevant.
To summarize @pete8475 was so kind that he run the benchmark on his Northwood P4 and shared the results:
Re: x86 microarchitecture benchmark (MandelX)
But the results were not that much better, at least not the improvement I expected regarding SSE performance.
But later someone uploaded a Willamette 1.6 GHz result. And Look at the Willamette 😀
The Willamette 1 GHz normalized SSE result is almost 2x of Northwood/Prescott ones...
