VOGONS


Optimal Pentium 4 setup

Topic actions

Reply 40 of 55, by myne

User metadata
Rank Oldbie
Rank
Oldbie

Wiki is really all there is left out there.
I did look for a datasheet.

Things I built:
Mechwarrior 2 installer for Windows 10/11 Re: A comprehensive guide to install and play MechWarrior 2 on new versions on Windows.
Dos+Windows 3.11 auto-install iso template (for vmware)
Script to backup Win9x\ME drivers from a working install

Reply 41 of 55, by Ren225

User metadata
Rank Newbie
Rank
Newbie
VivienM wrote on 2024-08-24, 12:52:
Ren225 wrote on 2024-08-24, 06:11:

My goal is specifically a P4 because it's slower than the Duo / Quad chips while still offering 64-bit support for newer OSs and applications.

I just finished running some benchmarks in Windows XP, and while I'm not surprised that it's slower, I'm still surprised at how MUCH slower it is. Even the basic E6600 is at least 2 - 3x faster, depending on the application. I'll probably install Vista onto a second drive and run some more tests, just to see.

Your benchmarks confirm what everybody discovered back in the summer of 2006 - the P4 was a lousy, inefficient design barely kept afloat by Intel's superior transistors.

Yep, absolutely! The P4 isn't great, and it was never an argument trying to make it look good, other than it's what I'm going with for my own esoteric needs, so I may as well see how far it can be pushed to make it run as quickly as possible. If I'm really hurting for LGA775 performance, I can always throw in the E6600 and get that 2-3x uplift. I'm sure even something like the low end E4200 would be faster, hehe 😀

The great thing about all this retro gear is that it's ultimately just for fun now, unlike 15-20 years ago where I'd be chucking the P4 as quickly as I could get away with, which is how I ended up with the Q6600 in the first place.

Getting back to the main DDR2 vs DDR3 comparison topic, it turns out that the P4 does prefer DDR2 as it offer slightly more performance likely due to faster timing more than anything else. But the E6600 took a slight performance hit even when running at 1066Mhz 5-5-5-15, compared to DDR3 running at 1066Mhz 6-6-6-16. The DDR2 sticks sure got pretty toasty at that speed though, but everything was stable even after 45+ minutes in game. Maybe with really high end DDR3 timing can be brought down further, but I found that DDR2 at 800Mhz 4-4-4-11 gave the P4 best performance out of everything, at least as far as it was able to be pushed and still POST.

As far as OS: Win XP > Win 7 > Win Vista, as far as performance. But I'm sure it's nothing new to anyone here haha

And with that, my system is where I need it to be as a weird P4 powered Windows XP / Windows 7 "dual drive / dual boot" monstrosity. Thanks to everyone for their input and advice 😀

Attachments

  • Old computer DDR2.PNG
    Filename
    Old computer DDR2.PNG
    File size
    74.25 KiB
    Views
    478 views
    File comment
    LGA775 DDR2
    File license
    Public domain
  • Old computer DDR3.PNG
    Filename
    Old computer DDR3.PNG
    File size
    70.45 KiB
    Views
    478 views
    File comment
    LGA775 DDR3
    File license
    Public domain

Reply 42 of 55, by agent_x007

User metadata
Rank Oldbie
Rank
Oldbie

There is quite a bit more to memory performance on 775 than just frequency and timings.
But to get those extra features you need mid range later boards at least (like straps settings, "Performance Level" boosters).
Also, FSB speed is limiting factor on DDR3, since 1066MT/s FSB allows only 1066MT/s as max.
800MT/s FSB on Pentium 4s locks you to 800MT/s max on memory.
Try to get a lower stock clock 800MT/s FSB P4, and raise FSB to the same level as E6600.
Sure, you OC the CPU, but both Pentium 4/Pentium D need all FSB speed they can get.
Also, if FSB speed or memory speeds were the only things needed to reach Core 2 performance on Pentium Ds, we would have a WHOLE different conversation now 😁

Reply 43 of 55, by Ren225

User metadata
Rank Newbie
Rank
Newbie
agent_x007 wrote on 2024-08-25, 21:15:
There is quite a bit more to memory performance on 775 than just frequency and timings. But to get those extra features you nee […]
Show full quote

There is quite a bit more to memory performance on 775 than just frequency and timings.
But to get those extra features you need mid range later boards at least (like straps settings, "Performance Level" boosters).
Also, FSB speed is limiting factor on DDR3, since 1066MT/s FSB allows only 1066MT/s as max.
800MT/s FSB on Pentium 4s locks you to 800MT/s max on memory.
Try to get a lower stock clock 800MT/s FSB P4, and raise FSB to the same level as E6600.
Sure, you OC the CPU, but both Pentium 4/Pentium D need all FSB speed they can get.
Also, if FSB speed or memory speeds were the only things needed to reach Core 2 performance on Pentium Ds, we would have a WHOLE different conversation now 😁

Thanks for the tip! 😀

Reply 44 of 55, by The Serpent Rider

User metadata
Rank l33t++
Rank
l33t++
Shponglefan wrote on 2024-08-24, 20:11:

Is there any difference between an 865 versus 875 chipset under Windows 98?

875 has some memory performance tweaks on paper, but some manufacturers implemented them on 865 anyway.

I must be some kind of standard: the anonymous gangbanger of the 21st century.

Reply 45 of 55, by Ren225

User metadata
Rank Newbie
Rank
Newbie

I decided to do some more experimentation on people's advice and try out a low end Core 2 Duo. I snagged a cheap E6300 off e-bay, underclocked it to 1.2Ghz (200Mhz FSB and 6x multiplier), and reran my little series of benchmarks.
The results, unsurprisingly, are actually kind of inline with what I was getting with the P3 630 @ 3GHz: Some things are a little slower, others a little faster.
I'll probably stick with this setup for now, if only for the heat and power draw improvement.

And it's also interesting seeing benchmarks of the P3-S @ 1.4Ghz, and how the E6300 @ 1.2Ghz is not THAT much faster. If I didn't need 64-bit and SSE2, I'd definitely be considering the P3 over the P4 and newer 😀

Attachments

Reply 46 of 55, by PcBytes

User metadata
Rank l33t
Rank
l33t
Trashbytes wrote on 2024-08-22, 06:14:
IIRC some late model 775 P4 boards did support early DDR2 but that was due to the Nothbridge supporting it and not the CPU since […]
Show full quote
H3nrik V! wrote on 2024-08-22, 05:32:

The real fun part is that both boards seem to also support Core2 CPUs, so if the P4 doesn't do it for you, you can tinker a little with some more power 😀

But yeah, Pentium4 wasn't DDR3 territory, actually, I think it was more like DDR1 - at least Northwood arcitechture ..

IIRC some late model 775 P4 boards did support early DDR2 but that was due to the Nothbridge supporting it and not the CPU since it wasn't till Nehalem that Intel moved the Memory Controller the CPU, AMD IIRC had it on CPU since the K8 (2003).

Some late model 478 boards also supported DDR2, usually it was the ones with 915 and later chipsets, dont quote me on this as it was a weird time with late 478 boards. (I think Asrock and Biostar made 915 478 boards with DDR2, not that I would touch a Biostar board)

-Derp Northbridge not ICH

ASRock's implementation of the 915 chipset on 478 didn't feature DDR2. At least the one I own - P4Dual-915GL - has DDR1 slots only, but does add a nice PCI-E slot, making it easier to choose GPUs for it.

945 was where they went with DDR2 on the 478 - P4i945 was the name IIRC.

"Enter at your own peril, past the bolted door..."
Main PC: i5 3470, GB B75M-D3H, 16GB RAM, 2x1TB
98SE : P3 650, Soyo SY-6BA+IV, 384MB RAM, 80GB

Reply 47 of 55, by Trashbytes

User metadata
Rank Oldbie
Rank
Oldbie
PcBytes wrote on 2024-09-07, 09:52:
Trashbytes wrote on 2024-08-22, 06:14:
IIRC some late model 775 P4 boards did support early DDR2 but that was due to the Nothbridge supporting it and not the CPU since […]
Show full quote
H3nrik V! wrote on 2024-08-22, 05:32:

The real fun part is that both boards seem to also support Core2 CPUs, so if the P4 doesn't do it for you, you can tinker a little with some more power 😀

But yeah, Pentium4 wasn't DDR3 territory, actually, I think it was more like DDR1 - at least Northwood arcitechture ..

IIRC some late model 775 P4 boards did support early DDR2 but that was due to the Nothbridge supporting it and not the CPU since it wasn't till Nehalem that Intel moved the Memory Controller the CPU, AMD IIRC had it on CPU since the K8 (2003).

Some late model 478 boards also supported DDR2, usually it was the ones with 915 and later chipsets, dont quote me on this as it was a weird time with late 478 boards. (I think Asrock and Biostar made 915 478 boards with DDR2, not that I would touch a Biostar board)

-Derp Northbridge not ICH

ASRock's implementation of the 915 chipset on 478 didn't feature DDR2. At least the one I own - P4Dual-915GL - has DDR1 slots only, but does add a nice PCI-E slot, making it easier to choose GPUs for it.

945 was where they went with DDR2 on the 478 - P4i945 was the name IIRC.

Yeah, was a weird time with usually Asrock doing blackmagic shenanigans with chipsets, slots and ram types on different board models. I really miss this part of computing when fabs were free to try different things and come up with hardware combinations that were far from standard.

Reply 48 of 55, by Trashbytes

User metadata
Rank Oldbie
Rank
Oldbie
Ren225 wrote on 2024-09-07, 05:15:
I decided to do some more experimentation on people's advice and try out a low end Core 2 Duo. I snagged a cheap E6300 off e-bay […]
Show full quote

I decided to do some more experimentation on people's advice and try out a low end Core 2 Duo. I snagged a cheap E6300 off e-bay, underclocked it to 1.2Ghz (200Mhz FSB and 6x multiplier), and reran my little series of benchmarks.
The results, unsurprisingly, are actually kind of inline with what I was getting with the P3 630 @ 3GHz: Some things are a little slower, others a little faster.
I'll probably stick with this setup for now, if only for the heat and power draw improvement.

And it's also interesting seeing benchmarks of the P3-S @ 1.4Ghz, and how the E6300 @ 1.2Ghz is not THAT much faster. If I didn't need 64-bit and SSE2, I'd definitely be considering the P3 over the P4 and newer 😀

well the Core2 stuff not being much faster than a P3 at the same clocks is understandable . .Core2 is pretty much a Pentium M (Which is a P3) with some of the better Netburst improvements and a dose of steroids. Its actually amusing that in the end they had to go back to move forwards and that they finally understood speed cannot overcome inefficiencies in power draw, heat and IPC. (Netburst wasn't totally terrible it had some great core improvements, its LOOOOOOOONG pipeline was what killed it)

Sadly Intel has forgotten this and their latest 14 gen CPUs are once again falling into the faster must be better fallacy again, which along with their LGA1700 sockets being terribly designed is causing them to fall behind AMD once again.

Reply 49 of 55, by PcBytes

User metadata
Rank l33t
Rank
l33t

"Those who forget their history are condemned to repeat it."

Literally Intel nowadays.

"Enter at your own peril, past the bolted door..."
Main PC: i5 3470, GB B75M-D3H, 16GB RAM, 2x1TB
98SE : P3 650, Soyo SY-6BA+IV, 384MB RAM, 80GB

Reply 50 of 55, by VivienM

User metadata
Rank Oldbie
Rank
Oldbie
PcBytes wrote on 2024-09-07, 13:21:

"Those who forget their history are condemned to repeat it."

Literally Intel nowadays.

Except it's much, much worse today. In the depths of the HotBurst darkness, Intel still had their traditional edge in transistors. Sure, those nice small transistors were wasted on poor inefficient designs, but that manufacturing ability was there. When they fixed the inefficient designs with C2, their combination of good designs + best transistors kept AMD off-side for a decade of dutiful tick-tock improvements.

Today, their manufacturing has been left behind by the 'smartphone economy'. All the new Intel laptop chips are largely fabbed by TSMC. The power hungriness of the recent desktop chips is not "oh we're being idiots and designing something inefficient because marketing wants high-GHz", it's "we need to guzzle this much power to get roughly comparable performance because our transistors are so much behind the times compared to AMD's TSMC-made chips".

Funny thing is - Intel was the big beneficiary of the previous industry shift. When PCs started driving the semiconductor industry, x86/x64 eventually came to eat almost everything else, and even what it didn't eat (e.g. IBM mainframes) migrated to PC-inspired underpinnings. And now smartphones drive the semiconductor industry, and the winners in PCs (Apple Silicon, AMD, etc) are those who have latched themselves to the smartphone economy. Andy Grove must be spinning in his grave.

Reply 51 of 55, by Ren225

User metadata
Rank Newbie
Rank
Newbie
Trashbytes wrote on 2024-09-07, 10:11:
Ren225 wrote on 2024-09-07, 05:15:
I decided to do some more experimentation on people's advice and try out a low end Core 2 Duo. I snagged a cheap E6300 off e-bay […]
Show full quote

I decided to do some more experimentation on people's advice and try out a low end Core 2 Duo. I snagged a cheap E6300 off e-bay, underclocked it to 1.2Ghz (200Mhz FSB and 6x multiplier), and reran my little series of benchmarks.
The results, unsurprisingly, are actually kind of inline with what I was getting with the P3 630 @ 3GHz: Some things are a little slower, others a little faster.
I'll probably stick with this setup for now, if only for the heat and power draw improvement.

And it's also interesting seeing benchmarks of the P3-S @ 1.4Ghz, and how the E6300 @ 1.2Ghz is not THAT much faster. If I didn't need 64-bit and SSE2, I'd definitely be considering the P3 over the P4 and newer 😀

well the Core2 stuff not being much faster than a P3 at the same clocks is understandable . .Core2 is pretty much a Pentium M (Which is a P3) with some of the better Netburst improvements and a dose of steroids.

That's interesting, I can see why some people choose to run a dual P3-S setup. Not only does it have the coolness factor, but probably still usable for 2000's games and applications considering the architecture. Still, a Core 2 Duo is probably cheaper to obtain with longer legs haha.

Reply 52 of 55, by fosterwj03

User metadata
Rank Oldbie
Rank
Oldbie

I always find the Core 2 to P4 comparison discussions interesting. I have no love for the P4 architecture, but I can say as an EE with a digital design background that the P4 architecture was way ahead of its time (about 20 years).

Intel likely realized that, with speed being king, they would have to extend the pipelines to maintain stable operations at, what was then, extremely high clock speeds. It was a matter of transistor-to-transistor latency that process improvements could never overcome. Larger stages could take too long to complete their operations within a single, fast clock tick. At a certain clock speed, the next tick occurs before the last instruction completes. You would need to make a smaller stage (either via breaking up a complex stage or creating a more efficient stage) to increase speed. Intel opted to break up the complex stages (vice efficacy) likely knowing that they would reach that point eventually anyway. The P4, as a result, performed poorly at initial clock speeds, but ramped up quickly to 3 GHz + speeds in relatively short time.

Still, the P4 stage design was really meant for 5GHz + speeds that the manufacturing processes of the time couldn't support. Excessive, for the time, power draw was simply a function of process size, clock speed, transistor density, transistor usage, and voltage. You can see all of this at play today with current designs from both Intel's and AMD's 5 GHz plus chips. Both companies employ much longer pipelines than superscaler chips of the late 90s, and transistor density has pushed power consumption much higher than we felt comfortable with in those days. As I said in another thread, we don't bat an eye at 20-stage pipelines and 100W+ chips these days, but we hate that it happened 20 years ago.

BTW, I suspect that a Core 2 Duo manufactured on a 90 nm process running at 3GHz would have consumed 100W+ if one could run that fast. Just saying that history needs a little perspective, too.

Reply 53 of 55, by VivienM

User metadata
Rank Oldbie
Rank
Oldbie
fosterwj03 wrote on 2024-09-08, 14:13:

You can see all of this at play today with current designs from both Intel's and AMD's 5 GHz plus chips. Both companies employ much longer pipelines than superscaler chips of the late 90s, and transistor density has pushed power consumption much higher than we felt comfortable with in those days. As I said in another thread, we don't bat an eye at 20-stage pipelines and 100W+ chips these days, but we hate that it happened 20 years ago.

But I think there are two big differences:
1) today's power guzzlers are only marketed towards enthusiasts. If you go and get your non-K, etc chip, it's supposed to run on a 65W TDP. And Dell OptiPlexes are not running the crazy beefy cooling systems that the P4/BTX era was going towards.
2) today's power guzzlers have dramatically lower idle power consumption. That, really, started with the 45nm C2D/C2Qs which dropped something like 30W over the early Conroes. And most computers spend most of their time idling...
(and, to a lesser extent, 3) cases, cooling systems, etc used on the higher-power enthusiast systems have adapted to cool that level of heat output quietly. The entire industry was not prepared for the increasing heat output of early 2000s systems and that reflected itself in increasing noise. The Apple/PPC side just as badly as the x86 world.)

So, for example, if I put an i7-14700 in a home server (which will be super-idle most of the time), it will guzzle substantially less power than if I put a Preshot 3.6GHz in my home server.

And if you fill up a computer lab full of OptiPlexes with i5-14500s, that will guzzle substantially less power than a computer lab full of Preshot 3.6GHzs.

Reply 54 of 55, by fosterwj03

User metadata
Rank Oldbie
Rank
Oldbie

It would be an interesting thought experiment to contemplate how Intel would have changed their plans if they knew in advance that their 90 nm process (shoot, even their 65 nm process) wouldn't scale effectively past 4 GHz. Would they have delayed the Pentium 4 design by 10 or 15 years and put their energies into a Banias-style design instead? How would things be different today without the lessons learned from the Pentium 4 (Hyperthreading, improved decode logic, improved branch prediction algorithms, advanced power saving states, etc)?

I bet Intel would have pushed the silicon to extreme TDPs in that case as well, just like they do today.

Reply 55 of 55, by Trashbytes

User metadata
Rank Oldbie
Rank
Oldbie
fosterwj03 wrote on 2024-09-08, 17:36:

It would be an interesting thought experiment to contemplate how Intel would have changed their plans if they knew in advance that their 90 nm process (shoot, even their 65 nm process) wouldn't scale effectively past 4 GHz. Would they have delayed the Pentium 4 design by 10 or 15 years and put their energies into a Banias-style design instead? How would things be different today without the lessons learned from the Pentium 4 (Hyperthreading, improved decode logic, improved branch prediction algorithms, advanced power saving states, etc)?

I bet Intel would have pushed the silicon to extreme TDPs in that case as well, just like they do today.

Would they be Intel if they didnt ? I mean for years that was their shtick and while they stopped that for a while when they had zero competition they went right back to it once AMD started whipping their asses. Only issue now is their tech just isn't there to handle that kind of punishment and so you get issues where their 13th and 14th gen chips are failing in large quantities. I still believe this is due to large amounts of electron migration taking place in the core causing irreversible damage a lot of this is due to the crazy amounts of voltage they are dumping into the core to run at 5Ghz+ speeds.

These failures are not just due to their transistor tech either their socket design is pretty woeful at allowing clean surface contact with the IHS. Some blame also lies with the 3rd party fabs and their habit of overclocking the BIOS right out of the box so the CPU starts with out of spec turbo and power settings just so the consumer can see them sexy 5Ghz numbers. (Yes ASUS Im staring at you you filthy scammers)

All in all Intel has a lot of work to do to get back to where they were with their Core CPUs and its possible with their current management they have hit their own Bulldozer era of manglement.