chinny22 wrote on 2022-11-30, 11:44:

I say go for it, in the end only you will know if it works for you. […]
Show full quote

I say go for it, in the end only you will know if it works for you.

I miss understood you were installing 2 video cards. Assuming you would simply disable the GF6 in Win98 and FX in WinXP. otherwise it'll be very confusing driver wise as both cards share the same driver package.
That said I'm not even sure if you really need both video cards. Admittedly I didn't do much testing or troubleshooting but only game I remember not liking the GF6 was C&C Generals. Could well be all your games run fine?
As you have a backup Win9x PC I'd just play any games that don't like the GF6 on that and like you said get 999 fps in Quake 3

Also I never had any problems with HT enabled as Win98 just ignores it, if any inefficiency does exist the CPU is so overpowered for Win98 you don't notice anyway.

Worst case as I see it if the PC doesn't work out just forget about XP, chuck the FX card in and your still left with a crazy fast Win98 PC and as a worst case scenario it's a pretty good one.

Yeah about that dual GPU setup...

The PCI FX5 cards are horrible, and the AGP FX5 cards are overpriced and offer DX9 support as a "demo feature".
The PCI GF4 cards are MX only so horrible again, and the AGP GF4 Ti cards are overpriced and have DirectX 8 max.
The PCI GF6 or newer I didn't bother with, nor anything from ATI, as they break table fog/palettized textures.
Going the reverse way - strong W98 AGP card + fastest PCI DX9 GPU just flips the problem, there isn't a PCI card worth it for DX9.
The PT880 (Pro/Ultra) chipset offers AGP and PCIE (at x4, should be enough?) so this could be mitigated, but these rare dual/hybrid GPU motherboards are expensive, and god forbid the MBO dies, finding a replacement would be pain.
AGP this modern takes only 1.5V cards, so I dunno what options are regarding Voodoo cards if I ever wanted one in the build.

So unless I have some kind of "AGP switchboard" or a "dual AGP riser" or manually change AGP GPUs per Windows boot, a solid hybrid seems just out of reach...

A single AGP GPU if I had to choose for the system always has one drawback - FX5 is just a faster "DX9 games in DX8 mode" card than GF4Ti, GF4Ti is DX8 fixed so I dunno how it handles Glide wrappers but it has a long driver support for compatibility but slowest of the bunch, and GF6 wins with DX8 and DX9 but breaks some compatibility and I'm unsure if anything other than 6200 is officially supported and I can't go further than this series for DX9 WXP speeds.

What about an AGP GF4Ti/FX5xxx + PCI 9500GT?
It's slightly period incorrect, but much cheaper (compared to 2 builds or AGP+PCIE mobo) and covers this whole idea, Turbo98 + solid mid-life XP?
Except for this issue: Re: Best low-profile PCI graphics card?

Or AGP to PCI, PCI-E to PCI thingies I see around?

Yeh I don't know I'm afraid.
If it was me I'd just see how far you can get with just the AGP GF6. It may be your worrying about compatibility unnecessarily.

2mg wrote on 2022-11-29, 22:35:

In Win98?
If yes, what's wrong with PCI FX5xxx and W98, W98 was made for PCI and AGP cards afaik?

In any windows. Its a driver bug, and the AGP version is mostly immune to these problems.

2mg wrote on 2022-11-29, 22:35:

What I've read here, they're the fastest AND most compatible for W98, tho didn't see if AGP or PCI...

Its not bad, but depends on what you want to play. If you want to play games after 2000, then the FX series will be fine, otherwise i recommend the matrox g200/riva128/savage4/permedia2.
Its not the fastest, but it depends on the CPU again. You will need a pentium3 at least to unleash its potential, otherwise it will be a slideshow on old systems.

chinny22 wrote on 2022-12-01, 09:48:

Yeh I don't know I'm afraid.
If it was me I'd just see how far you can get with just the AGP GF6. It may be your worrying about compatibility unnecessarily.

Yes, but see a post from a user here, saying 6000 are unstable in W98.

Geri wrote on 2022-12-01, 10:15:

In any windows. Its a driver bug, and the AGP version is mostly immune to these problems. […]
Show full quote

In any windows. Its a driver bug, and the AGP version is mostly immune to these problems.

...

Its not bad, but depends on what you want to play. If you want to play games after 2000, then the FX series will be fine, otherwise i recommend the matrox g200/riva128/savage4/permedia2.
Its not the fastest, but it depends on the CPU again. You will need a pentium3 at least to unleash its potential, otherwise it will be a slideshow on old systems.

About PCI driver bug, does it affect only FX5 PCI series, or do GF4MX PCI and GF6 PCI series?
Because I could fit an GF7/8/9 PCI GPU for a bit of DX9 performance, and GF4/FX5/GF6 into AGP for DX7/8/early 9.

The second part is already done in another build, plus these cards are useless for this W98 + P4 + fast W98 GPU (+ WXP) hybrid build.

2mg wrote on 2022-12-01, 16:55:

chinny22 wrote on 2022-12-01, 09:48:

Yeh I don't know I'm afraid.
If it was me I'd just see how far you can get with just the AGP GF6. It may be your worrying about compatibility unnecessarily.

Yes, but see a post from a user here, saying 6000 are unstable in W98.

ideal? no but you already know that but I wouldn't call them unstable. I mean people are running GF7 with modified drivers so it's doable. Will say this is for AGP though, not sure if PCI or PCIe changes things.

Will say in my case no drivers would recognize my 6800 Ultra AGP but I also had 4GB ram. After applying the Rlowe RAM patch that and a bunch of other separate hardware issues went away.

Geri wrote on 2022-12-01, 10:15:

In any windows. Its a driver bug, and the AGP version is mostly immune to these problems.

Is this FX series only issue?
I'd add a GF4 440 MX PCI for some Win98 compatibility, or reverse it and put 6200 PCI for some XP compatibility.

2mg wrote on 2022-12-01, 16:55:

About PCI driver bug, does it affect only FX5 PCI series, or do GF4MX PCI and GF6 PCI series?
Because I could fit an GF7/8/9 PCI GPU for a bit of DX9 performance, and GF4/FX5/GF6 into AGP for DX7/8/early 9.

The only modern nvidia pci card i have is an FX5200 pci. I dont have personal experience with other gen besides this.

Old GeForce PCI cards have horrendous performance in certain games. Like anything on Quake engines. Framerate barely changes with settings. GeForce 8+ cards are better, but don't have 16-bit dithering and lack Win9x support, so they are also out of the question too.

So for PCI video card, your only good option would be something from ATi. Don't buy Radeon 7000 PCI though, these cards are just weak (TNT2 performance).

The Serpent Rider wrote on 2022-12-03, 00:33:

Old GeForce PCI cards have horrendous performance in certain games. Like anything on Quake engines. Framerate barely changes with settings. GeForce 8+ cards are better, but don't have 16-bit dithering and lack Win9x support, so they are also out of the question too.

So for PCI video card, your only good option would be something from ATi. Don't buy Radeon 7000 PCI though, these cards are just weak (TNT2 performance).

Yeah, saw that vid, I mean, 9500gt and up (except in Far Cry) actually go above 60FPS - that gave me an idea to plop in one of those "high end PCI" GPUs and use them in XP, since they do have drivers for XP and can run at least early DX9 titles, while I put GF4Ti or FX5 into the AGP slot for that sweet 98 triple digit FPS.

Basically, I can have the early XP machine and a turbo 98 machine, which compliment one another as a transition between P3 and Core2, plus troubleshooting from one OS for the other if something breaks.
I can reverse it with a good DX9 AGP in XP and a W98 PCI GF4MX/FX5500 (btw where are the 5700LE PCI) too, but I can make a better XP system anyway, so that's kinda moot, also a poster here said PCI FX5s are crap driver/stability wise.

Is 9500gt in WinXP S478 period correct, no it's not (tho XP did last until early 2010s!), but my other options are:

- Either repurpose the s478 I have into a full turbo GF4/FX5 98 system (with XP limited to DX8 only) or make another similar turbo 98 build, but then my current one has no actual purpose.

- Get one of those hacky dual AGP/PCIE mobos, but this is tedious to source, not cheap, and seems that neither the AGP nor PCIE are the best implementations on those mobos. Some are 478, some 775, some have W98 drivers, some don't. AFAIK 99% of don't support Core2 (are 775 Prescotts any better in any way than 478?)

- Physically swap GPUs in the same PC each time I want XP or 98 (maybe add a AGP riser so not to wear out the one on the mobo), impractical and wears things.

- I have no experience (and it's expensive) to get those AGP2PCI which were made for Voodoos anyway.

- Plop FX5700 or higher into the AGP and be done with it - except high end FXs seem obscenely overpriced, loud, and generally meh, weak DX9, least compatible W98 drivers (aside from 6000 series, but they really break some legacy thingies ala palleted textures).

Addendum question:

I see some 775 boards with AGP, or AGP+PCIe (usually VIA PT880), with W98 drivers, but most are still limited to pre-Core2 CPUs, tho some do support Preslers.

How does a Northwood/Prescott 478 compare to 775 one, at the same speed, any benefits or just a different pinout?
Are Cedar Mills worth it? Or skip them and go for Preslers?

2mg wrote on 2022-12-04, 03:49:

Addendum question: […]
Show full quote

Addendum question:

I see some 775 boards with AGP, or AGP+PCIe (usually VIA PT880), with W98 drivers, but most are still limited to pre-Core2 CPUs, tho some do support Preslers.

How does a Northwood/Prescott 478 compare to 775 one, at the same speed, any benefits or just a different pinout?
Are Cedar Mills worth it? Or skip them and go for Preslers?

4CoreDual-SATA2 R2 supports Core2 CPUs along with Quad Core2 CPUs along with said AGP and PCIe, as for comparisons between 478 P4s and 775 P4s they are much the same other than higher bus speeds on the 775 boards which meant they could lower the multipliers thus it gave them a larger range of CPU speeds to work with. Accordingly the higher bus speeds and faster DDR2 meant they did have a slightly higher performance bracket but shit its still a P4 so even at the higher speed the performance increase wasn't huge.

Cedar Mill being the final single core revision of the Netburst architecture means they really are the only option if you want to overclock the snot out of them without melting the VRM or needing beefy cooling. If you are looking for dual core then my suggestion is to just jump to the Core2 based Pentiums ..I cant recommend the Pentium D series ..they got the nick name Pentium Disaster for a reason.

TrashPanda wrote on 2022-12-04, 05:46:

4CoreDual-SATA2 R2 supports Core2 CPUs along with Quad Core2 CPUs along with said AGP and PCIe

Offtopic, but is it smart getting a mobo that doesn't have W98 drivers for a W98 build?
I see people running some of these interesting non-W98 ASrocks, but most of them have no W98 drivers, and I see that some users say it's a mixed bag...

comparisons between 478 P4s and 775 P4s they are much the same other than higher bus speeds on the 775 boards which meant they could lower the multipliers thus it gave them a larger range of CPU speeds to work with.
Accordingly the higher bus speeds and faster DDR2 meant they did have a slightly higher performance bracket but shit its still a P4 so even at the higher speed the performance increase wasn't huge.

So a 775 Prescott is just a 478 Prescott ported to a new socket, no CPU architecture optimizations, aka they will perform very similar (and the only boost coming from faster DDR and FSB)?

Cedar Mill being the final single core revision of the Netburst architecture means they really are the only option if you want to overclock the snot out of them without melting the VRM or needing beefy cooling. If you are looking for dual core then my suggestion is to just jump to the Core2 based Pentiums ..I cant recommend the Pentium D series ..they got the nick name Pentium Disaster for a reason.

I know Cedar Mill has lowered TDP, but is it any more optimized compared to Prescott/Prescott-2m?
Is actually Prescott-2m 775 any better than Prescott 775?
Same question goes for Pentium D, is Presler noticeably better than Smithfield, and is Pentium D better than Cedar Mill?

I know I'll use only 1 core in W98, but generally I want to know how they compare (Presler vs Smith, Pentium D vs Cedar, Cedar vs Prescott/P-2m) for future stuff.

2mg wrote on 2022-12-04, 09:37:

Offtopic, but is it smart getting a mobo that doesn't have W98 drivers for a W98 build? I see people running some of these inter […]
Show full quote

TrashPanda wrote on 2022-12-04, 05:46:

4CoreDual-SATA2 R2 supports Core2 CPUs along with Quad Core2 CPUs along with said AGP and PCIe

Offtopic, but is it smart getting a mobo that doesn't have W98 drivers for a W98 build?
I see people running some of these interesting non-W98 ASrocks, but most of them have no W98 drivers, and I see that some users say it's a mixed bag...

comparisons between 478 P4s and 775 P4s they are much the same other than higher bus speeds on the 775 boards which meant they could lower the multipliers thus it gave them a larger range of CPU speeds to work with.
Accordingly the higher bus speeds and faster DDR2 meant they did have a slightly higher performance bracket but shit its still a P4 so even at the higher speed the performance increase wasn't huge.

So a 775 Prescott is just a 478 Prescott ported to a new socket, no CPU architecture optimizations, aka they will perform very similar (and the only boost coming from faster DDR and FSB)?

Cedar Mill being the final single core revision of the Netburst architecture means they really are the only option if you want to overclock the snot out of them without melting the VRM or needing beefy cooling. If you are looking for dual core then my suggestion is to just jump to the Core2 based Pentiums ..I cant recommend the Pentium D series ..they got the nick name Pentium Disaster for a reason.

I know Cedar Mill has lowered TDP, but is it any more optimized compared to Prescott/Prescott-2m?
Is actually Prescott-2m 775 any better than Prescott 775?
Same question goes for Pentium D, is Presler noticeably better than Smithfield, and is Pentium D better than Cedar Mill?

I know I'll use only 1 core in W98, but generally I want to know how they compare (Presler vs Smith, Pentium D vs Cedar, Cedar vs Prescott/P-2m) for future stuff.

Other than a smaller process node and lower TDP they are exactly the same, Cedar Mill is die shrunk Extreme Edition Presscott-2m as such they perform the same but CM produces less heat and uses less power clock for clock which does make it perfect for overclocking. (CM does have a multiplier Lock but unless you have absurd amounts of cash to burn you wont be buying the EE 3.73Ghz CPU)

Is Presshot 2m better than Presshot ? yes, the larger cache on the 2m models increased performance quite a bit for cache dependent tasks. The original Presshot range was also a mess since it was around that time that x64 was added to the CPU core and not all models supported it, there were also identical CPUs but with different TDP ratings which confused the mess even further.

Pentium D is a disaster ..avoid these CPUs .. better yet forget they ever existed, but since you asked, Smithfield runs hotter and is a power hog compared to Pressler, think of Smithfield as two Presshot CPUs on the one die and consider Pressler as two Cedar Mill CPUs on one die. Now you might be thinking well wouldn't that make Pressler better than CM .. no, Pentium D runs HOT and will punish the VRM until it throttles itself before you can ever get the performance out of both CPUs, this also means overclocking them is difficult.

That said ...if you like a challenge grab a Pentium D 955/965 Extreme Edition which has the unlock multiplier, this alone makes them easier to overlock or underclock to avoid the heat issue. Just realise that Pentium D was pretty much a power hog no matter what you were trying to do with it, not abnormal to see them hitting 130Watts+ power draw just for the CPU, they were also outclassed in every way by the Core2 Pentium CPUs.

If you want a laugh . .grab a Celeron D which is the garbage tier model of these CPUs .. its hilarious that Intel managed to sell any of the Celeron D lineup.

EDIT - I should add that there were two rare Presshot 2m CPUs that had Intel VT-x the models were the 662 and 672, I have yet to find any confirmation that CM had this function and all the Intel ARK pages for CM CPUs list that function as not enabled, so if you need virtualization for some odd reason then the Pentium 4 HT 662 and 672 are the only models with it.

TrashPanda wrote on 2022-12-04, 11:14:

Other than a smaller process node and lower TDP they are exactly the same, Cedar Mill is die shrunk Extreme Edition Presscott-2m as such they perform the same but CM produces less heat and uses less power clock for clock which does make it perfect for overclocking. (CM does have a multiplier Lock but unless you have absurd amounts of cash to burn you wont be buying the EE 3.73Ghz CPU)

"Other than a smaller process node and lower TDP they are exactly the same" - CM vs Prescott-2m, or 775 Prescott vs 478 Prescott, or CM vs 775 Prescott?

"Cedar Mill is die shrunk Extreme Edition Presscott-2m as such they perform the same" - CM vs EE Prescott, or CM vs Prescott-2m?

I really need an X > Y or X = Y here starting with 478 Prescott thru Gallatin and 775s and Pentium Ds and so on if you're bored 😁

Is Presshot 2m better than Presshot ? yes, the larger cache on the 2m models increased performance quite a bit for cache dependent tasks. The original Presshot range was also a mess since it was around that time that x64 was added to the CPU core and not all models supported it, there were also identical CPUs but with different TDP ratings which confused the mess even further.

With x64 being that EM64T, at least on Asrock page it has than name?
About identical CPUs, you mean those SKU naming like SLES7 (made it up)?

Pentium D is a disaster ..avoid these CPUs .. better yet forget they ever existed, but since you asked, Smithfield runs hotter and is a power hog compared to Pressler, think of Smithfield as two Presshot CPUs on the one die and consider Pressler as two Cedar Mill CPUs on one die. Now you might be thinking well wouldn't that make Pressler better than CM .. no, Pentium D runs HOT and will punish the VRM until it throttles itself before you can ever get the performance out of both CPUs, this also means overclocking them is difficult.

So if I disable one core in D, do I still get more cache, and is that D core any better than a solo Cedar Mill/Pres-2m?
For single core, is a P4EE (Gallatin/Prescott-2m) any better than Pentium (D) EE (Smith/Presler) if you disable a core?

That said ...if you like a challenge grab a Pentium D 955/965 Extreme Edition which has the unlock multiplier, this alone makes them easier to overlock or underclock to avoid the heat issue. Just realise that Pentium D was pretty much a power hog no matter what you were trying to do with it, not abnormal to see them hitting 130Watts+ power draw just for the CPU, they were also outclassed in every way by the Core2 Pentium CPUs.

If you want a laugh . .grab a Celeron D which is the garbage tier model of these CPUs .. its hilarious that Intel managed to sell any of the Celeron D lineup.

Yeah I'm just perusing my options for a fast W98, I doubt I need any Extreme CPUs or Pentium Ds, but still, if a single core (+cache +fsb +optimisations) in any of them presents a boost compared to solo CPU solutions, I'd like to hear about it.

EDIT - I should add that there were two rare Presshot 2m CPUs that had Intel VT-x the models were the 662 and 672, I have yet to find any confirmation that CM had this function and all the Intel ARK pages for CM CPUs list that function as not enabled, so if you need virtualization for some odd reason then the Pentium 4 HT 662 and 672 are the only models with it.

I doubt I'll be virtualizing anything, I do like the idea of a fast 775 CPU tho to emulate DOSBox if need be, or for laughs, virtualize pre-98 systems.
So I doubt I'd even need VT-x to virtualize anything pre-98, tho I'm used doing that with CPUs from this decade, dunno what would happen with 775 CPU doing VMs.
Isn't Wikipedia trustworthy enough for this VT-x info since ARK is a mess sometimes?

Cedar Mill is die shrunk Extreme Edition 2m 3.73 Presshot which only came in 775, you can ignore the Gallatin models. If you are looking for a performance chart then 775 versions will be better than anything from 478, the 2m 775 versions are better than the normal 512k cache 775 versions and CM sits at the top due to it being the final revision @ 65nm and built on the 90nm 2m EE Presshot. (2m just means 2Mb cache)

If you disable one CPU then you only get the cache from the active CPU, all the Pentium D is is two stupidly inefficient CPU dies glued together on one substrate, think of it as SMP on one socket ...which it essentially is. The EE Presshot @ 3.73Ghz was the best you could get in unlocked single core P4 CPUs, it will still kill a non EE Pentium D with one core disabled, and if you get a Pentium D EE 965 and disable one core then you have what amounts to a 65nm 2m 3.73 EE Presshot, kinda defeats the purpose of buying a 965 honestly. But it should run a bit cooler than the 90nm 3.73 single core version, again if you have money to burn then have at it.

I'm ignoring Gallatin Extreme Editions ..they were VRM killers, at 130nm they ran stupidly hot and consumed crazy amounts of power ..they also only had 512k of L2 cache and their performance sucked...they were the pinnacle of Intel's absurd push for clock speeds over IPC at the cost of TDP and Power. You can see when Intel realized that their 10Ghz goal could never be met with the P4 Uarc and they switched to trying to get its heat and power use down as their next two revisions both shrunk the process node from 130nm to 90nm and then down to 65nm.

Nope for single core the 2m 3.73 EE Presshot or the CM 661 are the best you can get in the Pentium 4 line up.

Wikipedia gets its info from ARK and links back to ARK, and the 775 ARK stuff is pretty dang accurate still.

I mean if you want some weird features there is always Socket 771 Xeons which can be modded to run on socket 775, however im not 100% sure any Pentium 4 Xeons made it to socket 771, I would have to check this.

TrashPanda wrote on 2022-12-04, 13:10:

(2m just means 2Mb cache)

HOW DID I NOT SEE THIS...

all the Pentium D is is two stupidly inefficient CPU dies glued together on one substrate

Isn't Presler actually 2 "cores" in one die vs Smith's two actual separate CPUs just glued to each other (and some wires)?

The EE Presshot @ 3.73Ghz was the best you could get in unlocked single core P4 CPUs

Wait, Gallatin has L3 cache, doesn't it help?

Pentium D EE 965 and disable one core then you have what amounts to a 65nm 2m 3.73 EE Presshot, kinda defeats the purpose of buying a 965 honestly. But it should run a bit cooler than the 90nm 3.73 single core version, again if you have money to burn then have at it.

Well, it would serve XP better, but yeah I get it, I'm here for W98 firstly.

they were the pinnacle of Intel's absurd push for clock speeds over IPC at the cost of TDP and Power

Wait, this is 2020s thinking!

CM 661 are the best you can get in the Pentium 4 line up.

As in the "smartest", I see Prescott 1m and 2M that go to 3.8, so cooling over +200mhz (that can also be gained back with OC)?

Xeons which can be modded to run on socket 775

Saw similar about Mobile Pentiums 479 with that 478 adapter, but nah, I'm good with "consumer" grade stuff.

2mg wrote on 2022-12-04, 18:07:

HOW DID I NOT SEE THIS... […]
Show full quote

TrashPanda wrote on 2022-12-04, 13:10:

(2m just means 2Mb cache)

HOW DID I NOT SEE THIS...

all the Pentium D is is two stupidly inefficient CPU dies glued together on one substrate

Isn't Presler actually 2 "cores" in one die vs Smith's two actual separate CPUs just glued to each other (and some wires)?

The EE Presshot @ 3.73Ghz was the best you could get in unlocked single core P4 CPUs

Wait, Gallatin has L3 cache, doesn't it help?

Pentium D EE 965 and disable one core then you have what amounts to a 65nm 2m 3.73 EE Presshot, kinda defeats the purpose of buying a 965 honestly. But it should run a bit cooler than the 90nm 3.73 single core version, again if you have money to burn then have at it.

Well, it would serve XP better, but yeah I get it, I'm here for W98 firstly.

they were the pinnacle of Intel's absurd push for clock speeds over IPC at the cost of TDP and Power

Wait, this is 2020s thinking!

CM 661 are the best you can get in the Pentium 4 line up.

As in the "smartest", I see Prescott 1m and 2M that go to 3.8, so cooling over +200mhz (that can also be gained back with OC)?

Xeons which can be modded to run on socket 775

Saw similar about Mobile Pentiums 479 with that 478 adapter, but nah, I'm good with "consumer" grade stuff.

Gallatin L3 didn't help at all for gaming or cache dependent tasks, it was slow cache and unable to be used in the same manner that L2 cache is. Gallatin was roughly 10-20% faster than Northwood due to the L3 cache and had it been fast L2 that performance increase would have been larger.

CM stopped at 3.6 because Core2 had been released so Intel dropped all Pentium4 lines, its capable of hitting 4.3 Ghz on Air/AIO with little issue ..Presshot would have extreme stability issues past 4Ghz due to heat without exotic cooling of some form. There are records of the 661 hitting 4.5Ghz stable on HWbot, there are also crazy people who have a Presshot EE at 4.4Ghz they also needed a nuclear reactor to power it and HVAC for the room it was running in. (Likey used some exotic cooling like LN2 or chiller for it)

Pressler and Smithfield are two dies glued together, Intel never produced multicore Pentium4 CPUs. Multicore didn't happen till Core2 CPUs.

TrashPanda wrote on 2022-12-05, 07:47:

Gallatin L3 didn't help at all for gaming or cache dependent tasks, it was slow cache and unable to be used in the same manner that L2 cache is. Gallatin was roughly 10-20% faster than Northwood due to the L3 cache and had it been fast L2 that performance increase would have been larger.

CM stopped at 3.6 because Core2 had been released so Intel dropped all Pentium4 lines, its capable of hitting 4.3 Ghz on Air/AIO with little issue ..Presshot would have extreme stability issues past 4Ghz due to heat without exotic cooling of some form. There are records of the 661 hitting 4.5Ghz stable on HWbot, there are also crazy people who have a Presshot EE at 4.4Ghz they also needed a nuclear reactor to power it and HVAC for the room it was running in. (Likey used some exotic cooling like LN2 or chiller for it)

Pressler and Smithfield are two dies glued together, Intel never produced multicore Pentium4 CPUs. Multicore didn't happen till Core2 CPUs.

All Smithfield processors were made of two 90 nm Prescott cores, next to each other on a single die with 1 MB of Level 2 (L2) cache per core.
Presler introduced the 'multi-chip module, or MCM, which consisted of two single-core dies placed next to each other on the same substrate package.
There was this slight difference, just mentioning it since I saw it on Wiki, doesn't matter in the grand scheme of things anyway.
Extreme Editions had more tech things going for them (XE versions of Smith/Press).

So in some order I got from this:
1. Pentium D 965 in "single core mode" = 65mm Presscot-2m (a cooler Press-2m at 3.73Ghz), ability to run "dual core mode" for WinXP (but hot)
2. CedarMill 661 (smartest choice due to lower thermals/TDP + higher OC potential)
3. 2m 3.73 EE Presshot (burn baby burn, but faster than CM661 at stock)

Then it all boils down to availability, price, and mobo CPU support (965 works with like 1 or maybe 2 Intel only chipsets).

Correct?

Gallatin L3 didn't help at all for gaming or cache dependent tasks

Gallatin has 15-20% performance boost over Northwood in games with heavy CPU load. Pretty impressive for "slow" cache.

The Serpent Rider wrote on 2022-12-05, 15:36:

Gallatin L3 didn't help at all for gaming or cache dependent tasks

Gallatin has 15-20% performance boost over Northwood in games with heavy CPU load. Pretty impressive for "slow" cache.

Yup slow as shit L3 cache, it should have been on the order of 30% or more uplift over Northwood but Intel cut corners and stuck it with 512kb of L2 instead of 2Mb. 15% uplift over Northwood really isn't that impressive when looking at actual real world numbers when Northwood performed terribly out of the gate. However you could clock the snot out of Gallatin due to the unlocked multi to make up for the deficit if you have the cooler that can handle the ~110watt TDP.

I actually don't get why they went with the slower L3 but I suspect its because Gallatin was a Xeon variant of the P4 ported to the consumer space to give them some breathing room to get Presshot out, in any case Gallatin never saw use outside of Extreme edition CPUs.

One other point to note about Gallatin .. it didn't have SSE3 or x64.