Finally an Intel system on the test bench that is faster than the Socket 754 Sempron 2800+ ! 😁

The 2004 platform Intel 925XE with a Pentium 4 670 CPU.
The Prescott 600 series were released winter 2004 - 2005 and the Pentium 4 670 is the fastest non extreme edition CPU the i925XE chipset supports.
I will only post overclocked scores for this system.
If someone wants to see the non overclocked scores just say so and I will post them.

The system is running @ 14*315 = 4400 mhz 1.45v
Superpi was 1s faster using 19 * 237 = 4500 mhz which also takes 1.45v to be Prime 95 stable.
The system is capable of running 14*321 mhz = 4500 mhz but that takes 1.5v to be Prime 95 stable.

The motherboard Abit AA8XE (Not the Fatality edition)
Is the motherboard small or the tower cooler large? A bonus point for the one that can name the cooler!

This is the first Intel system on the test bench that comes close to the AMD K8 CPUs performance in 3dmark 2001

3dmark 2003

3dmark 2005

3dmark 2006

PCmark 2005

SiSoft Sandra 2011 CPU Arithmetic + Multimedia

SiSoft Sandra 2001 Memory bandwidth and memory latency

SuperPi 1m 4500 mhz. The memory speed is detected incorrectly in CPU-Z. The system is using a 2:3 divider not a 1:2 divider.

And finally Youtube 720P without help from the video card.

I could use this system as my every day system.
When surfing the net it is impossible to notice that this system is almost 10 years old.

The next platform on the test bench will be Intel 975X. The last of the pre Core 2 Duo Intel platforms.
I will start on the other end of the spectrum of Intels single core NetBurst CPUs with a slow 2.8ghz Prescott 520/521.

Skyscraper wrote:

... A bonus point for the one that can name the cooler! ...

Danamics LMX Superleggera

Never heard of it. Liquid metal cooling is as exotic as they come, like the soviet Alpha submarines! 😎

Careful if you ever break and open the pipes, sodium spontaneously explodes in contact with water.

TELVM wrote:

Danamics LMX Superleggera

You get a gold star!

I like that the reviewing site got a box for the 3.5 inch PSU that drives the pump.

This is how my PSU looks like. Safety first!!
I got the cooler in a store clearance for ~40 euro a couple of years ago. New in a sealed box.
I do not think the company that made it exists any more so I am pretty sure I would get my money back with an interest if I would sell it.

The motherboard at its storage space with the cooler mounted.
The board is dirty, has scratches and bent caps but works perfectly. Its a dumpster find 😀
The Abit AA8XE is one of few Abit boards I actually like.

^ That poor Panny MBZ at the left! Help him! 🤣

TELVM wrote:

^ That poor Panny MBZ at the left! Help him! 🤣

To get that cap strait I would need to re-solder it 😀

The next platform on the test bench is the Intel 975X.
The i975X is a platform from late 2005 and it is the last chipset Intel released before the Core 2 Duo release.
The i925XE would have been a "better" platform to test single core NetBurst CPUs on but I do not want to dismount the large tower cooler 😜.
The overclocking options and performance with single core NetBurst CPUs are practically identical between my i925XE and i975X boards.

I found an Intel price list from early summer 2005. Its not complete but the P4 670 I tested earlier is on the list.
I have a few of the other CPUs on the list so I will do some tests to see what people got for their money back in 2005.

Intel Prices 06-12-2005

Pentium D 840 (3.2 GHz) -> $530;
Pentium D 830 (3.0 GHz) -> $316;
Pentium D 820 (2.8 GHz) -> $241;
Pentium 4 670 (3.8 GHz) -> $851.

Pentium 4 571 (3.8 GHz) -> $637;
Pentium 4 561 (3.6 GHz) -> $417;
Pentium 4 551 (3.4 GHz) -> $278;
Pentium 4 541 (3.2 GHz) -> $218;
Pentium 4 531 (3.0 GHz) -> $178;
Pentium 4 521 (2.8 GHz) -> $163.

The P4 670 is the most expensive CPU on the list.
Lets see how the cheapest CPU on the list compare.
Is the $700 price difference justifiable.

System spec

Asus P5W DH Deluxe i975X
Pentium 4 521 2.8 ghz
2*1gb DDR2 (some old Crucial Tracers)
Video Card Geforce 7900 GTX

The Asus P5W DH Deluxe

I thought that I would use the Zalman cooler with the single core NetBurst CPUs because it is easy to mount and dismount.

Not that easy...

But it sure looks nice once mounted

3dmark 2001 @ 2800 mhz. The P4 670 scored 27000 points @ 3800 mhz stock settings as a comparison

SuperPi 1m @ 2800 mhz. The P4 670 did 1m in 33s @ 3800 mhz stock settings as a comparison

So far the price difference seem justified.
Lets do some overclocking.

SuperPi 1M 4200 mhz. It beats the P4 670 clocked @ stock speed but cant match it when clocked to 4500 mhz.

3dmark 2001 4200 mhz

3dmark 2003 4200 mhz

3dmark 2005 4200 mhz

3dmark 2006 4200 mhz

PCmark 2005 4200 mhz

SiSoft Sandra 2011 CPU Arithmetic and Multimedia 2800 and 4200 mhz

SiSoft Sandra 2011 Memory speed and memory latency 2800 and 4200 mhz

Youtube 720P without help from the video card 4200 mhz. It is not as fast as the P4 670 @ 4400 but it is fast enough

For $163 the Pentium 4 521 must have been great value back in the summer of 2005 but only if it was overclocked.

Nice round, tried any modern games on it yet?

nforce4max wrote:

Nice round, tried any modern games on it yet?

I think I will wait with modern game tests until I test Pentium D and Athlon X2 CPUs
Although I do install Need for Speed Underground 2 on all systems to see that they work.
So far they all run the game flawlessly using max settings @ 1280*1024 😜

Using a fast single core CPU every game made before year 2008 usually work fine. When it comes to later games they often only run with all options at the minimum setting if at all.
I do not like to run games using "Low" settings 😀

From testing I did this spring I know that World Of Warcraft still runs OK on a Pentium 4 3.06@3450 mhz with a Radeon X1950 PRO AGP
Starcraft 2 also runs but not good enough for competitive online play.
Diablo 3 runs but just barely on the same machine.

Any plans on trying out any of the Cedar Mill chips? Those have been mostly forgotten since they were pretty quickly overshadowed by the Core2, but they were kinda the 'last hurrah' for the single-core P4...

Skyscraper wrote:

... The Asus P5W DH Deluxe […]
Show full quote

... The Asus P5W DH Deluxe

Notice how this 2005 mobo is beginning to look 'modern': Solid caps and DPI shielded inductors have made their appearance, and the VRM mosfets have got a heatsink to cool them down.

Old Thrashbarg wrote:

Any plans on trying out any of the Cedar Mill chips? Those have been mostly forgotten since they were pretty quickly overshadowed by the Core2, but they were kinda the 'last hurrah' for the single-core P4...

I do not own any Cedar Mill CPUs but I am keeping my eyes open.
There are unlimited numbers of 500 and 600 series Prescotts on Swedish "Ebay", not many 6x1 Cedar Mills.

TELVM wrote:

Notice how this 2005 mobo is beginning to look 'modern': Solid caps and DPI shielded inductors have made their appearance, and the VRM mosfets have got a heatsink to cool them down.

I have to admit that this edition of the motherboard is from early 2006.
But yes the i975X boards that were released late 2005 were no different.
Only half of the mosfets got a heat pipe though. F**K the rest!
This board has hold together great 😀 exept the W-LAN that comes and goes*.
*as in: works perfectly or the device does not even exist.

TELVM wrote:

Notice how this 2005 mobo is beginning to look 'modern': Solid caps and DPI shielded inductors have made their appearance, and the VRM mosfets have got a heatsink to cool them down.

..and thus the era of crap hardware slowly came to an end. Ironically they start doing this after CPUs started to run more efficiently. Had Socket 478 and Socket A boards used polymer caps and heatsinks, they likely would have been much more reliable.

Polys had been around way before 2005~6 in quality server and workstation boards:

Image from Capacitor Lab

But only after the 'plague' scare in the first half of the past decade did they bother to put polys in consumer boards.

Provided the formula you steal is the correct one 😁 and decent cooling that protects them from murderous heat, good lytics can be very reliable.

Its time for another update.

I noticed how the 500 series Prescott seemed notably slower in 3dmark 2001 than the 600 series Prescott.
Lets see how the Prescott 570 stand up against the 670.

System spec

Asus P5W DH Deluxe i975X
Pentium 4 570 3.8 ghz
2*1gb DDR2 (some old Crucial Tracers)
Video Card Geforce 7900 GTX

3dmark 2001 stock settings. The P4 670 got 27000 points with the same memory speed and timings as a comparison.

Superpi 1m stock settings. The P4 670 did 1m in 33s with the same memory speed and timings as a comparison.

SuperPi 1m 4300 mhz

3dmark 2001 4300 mhz

3dmark 2003 4300 mhz

3dmark 2005 4300 mhz

3dmark 2006 4300 mhz

PCmark 2005 4300 mhz

SiSoft Sandra 2001 CPU Arithmetic and Multimedia 2800 3800 4200 4300 mhz

SiSoft Sandra Memory speed and latency 2800 3800 4200 4300 mhz. Not having access to a low multiplier hurts this CPU here.

Since I tested the P4 670 on a board with i925XE chipset I wanted to be sure that the Prescott 500 series lousy performance in 3dmark 2001 compared to the 600 series isnt some issue with the i975X chipset.
I found a P4 630 and did a quick test run.

3dmark 2001 P4 630 4200 mhz. Much better than the P4 521 @ 4200 mhz even though the 630 use lower fsb and memory speed.

SuperPi 1m P4 630 4200 mhz. The same result as the P4 521 @ 4200 mhz

With NetBurst CPUs 3dmark 2001 prefers large cache size vs low cache latency!!!
Memory speed does not have the same impact as with K8 CPUs.
FSB impact the score much more than memory speed.

I wonder how much the scores will change if you benched with a board running DDR3, I know the boards are getting rare but from my test with core 2 laptops running ddr3 there is a noticeable difference compared to anything from the time that used ddr2. The reason why the scores are fsb dependent is due to the chipset and memory controller only being able to work with the cpu over the fsb. The bus its self isn't the issue and even at 200mhz (800qdr) the bus has roughly 6.4GB/s.

nforce4max wrote:

I wonder how much the scores will change if you benched with a board running DDR3, I know the boards are getting rare but from my test with core 2 laptops running ddr3 there is a noticeable difference compared to anything from the time that used ddr2. The reason why the scores are fsb dependent is due to the chipset and memory controller only being able to work with the cpu over the fsb. The bus its self isn't the issue and even at 200mhz (800qdr) the bus has roughly 6.4GB/s.

I think I can get these CPUs all the memory bandwidth they need with DDR2.
The P4 570 overclocked tests were run with 227 mhz FSB and 750 mhz memory speed and it performed much worse than the P4 521 @ 300 mhz fsb and 600 mhz memory speed.
The P4 570 was only somewhat saved by the fact that it was running 100 mhz faster otherwise it would have been beaten in every test.

With the Pentium D dual core CPUs the situation should be a bit different but I think high memory speed still will be of limited use without high fsb.
Most high multiplier Pentium D CPUs have speedstep just as the 600 series Prescott so they have access to a low multiplier.
The exeption beeing the strange Pentium D 805. Default speed is 20 x 133 = 2666 mhz and no lower multiplier is available.

The next CPU on the test bench will be the Pentium D 805!
It will be interesting to see just how much the low fsb will impact the performance compared to Pentium D CPUs running @ 800 mhz bus speed
It will also be interesting to see in which tests the dual cores really will outshine the single cores.

I have messed with two 805s and they nearly ruined a board 🤣 (overclocking) and yes they are faster. Try the 9xx series as they are closer to c2d performance than the 8xx.

nforce4max wrote:

I have messed with two 805s and they nearly ruined a board 🤣 (overclocking) and yes they are faster. Try the 9xx series as they are closer to c2d performance than the 8xx.

I am finally done with the Pentium D 8xx testing.
No motherboards were destroyed benchmarking these CPUs
The Pentium D 9xx CPUs will be next.

I tested 3 Pentium D 8xx CPUs

Pentium D 805 20*133 = 2666 mhz ~Price late 2005 $163
Pentium D 820 14*200 = 2800 mhz ~Price summer 2005 $241
Pentium D 840 16*200 = 3200 mhz ~Price summer 2005 $530

I tested the 805 and the 820 @ 3800 mhz because it makes it easy to compare their performace against a single core Pentium 4 570 3.8 ghz.
The 805 was tested with low fsb and decent memory speed and the 820 tested with high fsb and low memory speed so it is also intresting to see how these two low end dual core CPUs compare against each other.

I will not post screenshots for all test results with these two Pentium D CPUs and the P4 570 @ 3.8. It would just be too many screenshots to upload.
Feel free to request a screenshot for any benchmark if you feel something smells fishy 😁

Platform: i975X, Memory: 2x1gb ddr2, Video card: Geforce 7900 GTX.

All tests were run with these settings.

Pentium 4 570, 3.8 ghz 1.35v, 19*200, memory speed 301 mhz 3 3 3 8
Pentium D 805, 3.8 ghz 1.45v, 20*190, memory speed 317 mhz 3 3 3 8
Pentium D 820, 3.8 ghz 1.45v, 14*271, memory speed 271 mhz 3 3 3 6

SuperPi 1m

Identical scores! All three CPUs finished the 1m test in 34 seconds.

3dmark 2001

The 820 scored the best result with 28659 points
In second place we find the 805 with 28024 points
The P4 570 scored a bit lower than the dual cores with 25956 points

3dmark 2001 as shown in earlier tests prefer high FSB over high memory speed.
3dmark 2001 does also benefit from dual core CPUs but there isnt a huge difference in scores.

3dmark 2003

The 820 scored highest with 22482 points
In second place we find the 805 with 22388 points
In last place we wind the P4 570 with 21495 points

3dmark 2003 dosnt care much about fsb or memory speed so its not a huge surprise that the two dual core CPUs score almost equal.
3dmark 2003 will be GPU limited what ever we do but it seems to prefer dual core CPUs over Single Core CPUs.

3dmark 2005

The 820 scored highest with 9930 points
In second place we find the 805 with 9813 points
The P4 570 with 8560 points is last.. again...

I think we are starting to see a trend.

3dmark 2006

The 820 scored highest with 6058 points
In second place we find the 805 with 6047 points
In la... third place The P4 570 with 5259 points

The newer the benchmark the more the dual core CPUs advantage over the single core CPU grows.

PCmark 2005

1: 820. PCmark score 6602 points. CPU: 6311, Memory: 4938, Graphics: 8938, HDD: 6606
2: 805. PCmark score 6490 points. CPU: 6252, Memory: 4419, Graphics: 8724, HDD: 6597
3: 570. PCmark score 5382 points. CPU: 4861, Memory: 4554, Graphics: 8725, HDD: 6640

The 820 wins the memory test by a large margin although it has the slowest memory. Again we see the importance of FSB.
The Pentium 4 570 wins the hdd test and get the about the same graphics and memory scores as the 805 but it is dead last in the CPU test.

SiSoft Sandra 2011 CPU Artihmetic and Multimedia
The Pentium D 820 (red) scores equal to the Pentium D 805 (Orange) and the Pentium 4 570 (green) simply suck.

SiSoft Sandra 2011 Memory speed and latency
In these two graphs the 820 is red, the 570 is orange and the 805 is green (notice that the 805 and 570 have switched color).
Here we again can see that memory speed in mhz dosnt give much benefit if the FSB is too low.

To sum it up.
Dual core is better than single core and memory speed gives little benefit if the front side bus is to low to make use of the memory speed.

I will post the results for the Pentium D 840 @ 4000 mhz in a separate post.

Its time for another update.
Im more or less done with all the Pentium D testing.

I diddnt post the results for the Pentium D 840 in the last post since I wanted to compare it against the Pentium D 9xx CPUs
I tested the Pentium D 935 and 945. I will post results for the 935 @ stock 3.2 ghz to compare against the Pentium D 840 @ stock 3.2 ghz.
I will also post results and sceenshots for the Pentium D 945 @ 4700 mhz to compare against the Pentium D 840 @ 4000 mhz.

I had a scare yesterday.
The Pentium D does not benefit much from memory speed greater than the FSB.
The FSB is quadrupled so a fsb of 200 in practice equals 800 mhz.
Its the same with the memory. 2 channels of DDR memory clocked @ 200 mhz in practice equals 800 mhz.
This explains why there is little benefit clocking the memory higher than the FSB.
To get the best performance I needed memory that can do 3 2 2 4 2 @ 333 fsb 1:1 so the Pendtium D can run @ 14*333 = 4666 mhz with optimal memory performance.

I have some Micron D9 fat body modules that can handle the voltage needed for the timings I wanted and I did a quick test run with those timings @ 266 mhz fsb
I did use 2.4V memory voltage since I know that is what it would take to get the modules stable @ 333 mhz fsb with these timings.
All was good, the system stable and there were unicorns and rainbows everywhere.
Now I needed to change CPU to a Pentium D 915 since it was the only one that would run @ 14x multiplier.
All Pentium D 9xx should be able to run at this multiplier but its a known issue with my motherboard that this dosnt work exept for the 915 since it has 14x as stock multiplier.

Before changing the CPU I loaded "bios defaults" in the bios setup. My motherboard did not like this at all.
Instant power off was the result when I hit "save and exit" and after that the board would not post. I guess the high memory voltage played a part in this.
Nothing helped until I hot flashed a new bios. The board seems fine but that made me give up the idea of 4666 mhz with 666 mhz memory and optimal timings.
The Pentium D 945 was benched at 4700 mhz with lower fsb and less than optimal timings but this should not be more than ~1% slower than the settings I wanted to use.

Asus P5W DH i975X
2*1 gb DDR2 memory
Geforce 7900gtx
Thermalright Ultra-120 with a low rpm fan.

Pentium D 840 @ 3200 mhz memory: 666 mhz 3 4 3 9
Pentium D 935 @ 3200 mhz memory: 800 mhz 4 4 4 12
Pentium D 840 @ 4000 mhz memory: 1000 mhz 5 5 5 15
Pentium D 945 @ 4700 mhz memory: 555 mhz 3 3 3 6

Super Pi 1M

Pentium D 840 @ 3.2 did the 1m test in 39 seconds
Pentium D 935 @ 3.2 also finished the 1m test in 39 seconds
Pentium D 840@ 4000 managed to finish the test in 31 seconds

The Pentium D 945 @ 4700 scored the best SuperPi 1m time so far

3dmark 2001

Pentium D 840 got 25447 Points
Pentium D 935 got 27003 Points
Pentium D 840 @ 4000 managed to get 30195 points

The Pentium D 945 @ 4700 scored 35421 points

3dmark 2003

Pentium D 840 got 21610 Points
Pentium D 935 got 22053 Points
Pentium D 840 @ 4000 managed to get 22887 points

The Pentium D 945 @ 4700 scored 24049 points

3dmark 2005

Pentium D 840 got 8831 Points
Pentium D 935 got 8951 Points
Pentium D 840 @ 4000 managed to get 10419 points

The Pentium D 945 @ 4700 scored 11173 points

3dmark 2006

Pentium D 840 got 5822 Points
Pentium D 935 got 5821 Points
Pentium D 840 @ 4000 managed to get 6117 points

The Pentium D 945 @ 4700 scored 6291 points

PCmark 2005

Pentium D 840 got 5954 Points. CPU: 5304, Memory: 4101, Graphics: 8488, HDD: 6570.
Pentium D 935 got 5899 Points CPU: 5325, Memory: 4174, Graphics: 8436, HDD: 6601.
Pentium D 840 @ 4000 managed to get 6905 points

The Pentium D 945 @ 4700 scored 7502 points

SiSoft Sandra CPU Arithmetic and Multimedia

The Pentium D 945 @ 4700 at the top followed by the 840 @ 4000. Both the CPUs @ 3200 mhz score simular.

SiSoft Sandra Memory tests

The Pentium D 945 @ 4700 has the best memory bandwidth although it uses the lowest memory speed setting.
The Pentum D 840 @ 4000 also has decent memory bandwidth but the two stock slocked CPUs are a bit behind.
The stock clocked Pentium D 840 with its "out of sync" 666 mhz memory is the big loser when it comes to latency.

This is the Pentium D 840 @ 4000 playing Youtube 720P.

The Pentium D 8xx is not that much slower than the Pentium D 9xx but it does not reach the same clock frequencies.
At 4700 mhz I am feeling confident that the Pentium D 945 will be able to handle some more or less modern games.
A good 24/7 speed with worse cooling would be ~4.4 ghz. All 4 Pentium D 9xx CPUs i tried could reach this speed with only ~1.3V.
To reach 4.5 - 4.6 ghz 1.4v - 1.45v was needed and to reach 4.7 ~1.5v was needed.
With 1.6v I am sure I could have reached 4.8+ but that would have been a little too hot for my taste.

I will now test some socket 939 setups to see how they perform compared to the single and dual core NetBurst CPUs
When I am done with that I think I will do a gaming test shoot out between the best performing setups.

So it takes a 4.7GHz Pentium D to match a 2.4/533 Pentium M in 3DMark01. 🤣
PM 745 @ 2.4 on i915GM with 2GB dual-channel DDR2-533 and GeForce GTX260 = 37102 (win XP) and 26490 (win7)

Can't wait to see your S939 results!

Standard Def Steve wrote:

So it takes a 4.7GHz Pentium D to match a 2.4/533 Pentium M in 3DMark01. 🤣
PM 745 @ 2.4 on i915GM with 2GB dual-channel DDR2-533 and GeForce GTX260 = 37102 (win XP) and 26490 (win7)

Can't wait to see your S939 results!

I just had to do one more quick run with with the Pentium D setup. Slower than a Pentium M is not acceptable 😉
I changed the video card to a Geforce 8800 GTS to see if 3dmark 2001 would gain or if its totally bottlenecked by the CPU.
The score is not that much higher but I think with an even faster video card the Pentium D(eaccelerator) would at least break 40k 😀