VOGONS


GeForce 4 vs. GeForce FX?

Topic actions

Reply 100 of 217, by Joseph_Joestar

User metadata
Rank l33t
Rank
l33t
bloodem wrote on 2022-01-15, 11:19:

Oh, OK! That would explain it, then. The thing is... I also tested with what should be a pretty early Thoroughbred 1700+ (year 2002 / week 36), but on my KT133/KT266 boards I get no post with lower multipliers. The same CPU works great on the KT400/KT600 boards that I have.

I've posted a picture of my CPU here in case it helps.

The date code seems to be 0302 so early 2002 maybe? Or is that second week of 2003. Not sure how to read the code.

PC#1: Pentium MMX 166 / Soyo SY-5BT / S3 Trio64V+ / Voodoo1 / YMF719 / AWE64 Gold / SC-155
PC#2: AthlonXP 2100+ / ECS K7VTA3 / Voodoo3 / Audigy2 / Vortex2
PC#3: Athlon64 3400+ / Asus K8V-MX / 5900XT / Audigy2
PC#4: i5-3570K / MSI Z77A-G43 / GTX 970 / X-Fi

Reply 101 of 217, by bloodem

User metadata
Rank Oldbie
Rank
Oldbie
Joseph_Joestar wrote on 2022-01-15, 11:38:

I've posted a picture of my CPU here in case it helps.
The date code seems to be 0302 so early 2002 maybe? Or is that second week of 2003. Not sure how to read the code.

Interesting, your CPU is actually newer than mine: 03 is the year, 02 is the week (Thoroughbred was launched in June 2002, so week 3 of 2002 is not possible, anyway).
So it's still a mystery how much the motherboard matters in this equation. Looking at this old Tomshardware forum post, it does seem like motherboards have a lot to do with the "unlocking" process, but it's hard to say if that information is accurate or not.

Looking at this Anandtech article, it does seem like even the earliest Thoroughbred CPUs were not exactly unlocked by default, which further reinforces the idea that the motherboard could play a big role in unlocking these CPUs without the need for conductive ink in order to join the L1 bridges.

One thing is certain, you have an Abit motherboard, and these were known to always be extremely flexible and feature-packed in terms of overclocking (or, in this case, downclocking) capabilities.

1 x PLCC-68 / 2 x PGA132 / 5 x Skt 3 / 9 x Skt 7 / 12 x SS7 / 1 x Skt 8 / 14 x Slot 1 / 5 x Slot A
5 x Skt 370 / 8 x Skt A / 2 x Skt 478 / 2 x Skt 754 / 3 x Skt 939 / 7 x LGA775 / 1 x LGA1155
Current PC: Ryzen 7 5800X3D
Backup PC: Core i7 7700k

Reply 102 of 217, by Joseph_Joestar

User metadata
Rank l33t
Rank
l33t
bloodem wrote on 2022-01-15, 12:20:

So it's still a mystery how much the motherboard matters in this equation. Looking at this old Tomshardware forum post, it does seem like motherboards have a lot to do with the "unlocking" process, but it's hard to say if that information is accurate or not.

Interesting read. It could well be that motherboards play a certain role in this as well. That said, I was unable to downclock a Duron 700 in this exact motherboard, so there must be something on the CPU side as well.

One thing is certain, you have an Abit motherboard, and these were known to always be extremely flexible and feature-packed in terms of overclocking (or, in this case, downclocking) capabilities.

Very true. The KT7A has a "SoftMenu" section in the BIOS which pretty much lets you tweak CPU settings however you want. I've added a picture of that screen here.

PC#1: Pentium MMX 166 / Soyo SY-5BT / S3 Trio64V+ / Voodoo1 / YMF719 / AWE64 Gold / SC-155
PC#2: AthlonXP 2100+ / ECS K7VTA3 / Voodoo3 / Audigy2 / Vortex2
PC#3: Athlon64 3400+ / Asus K8V-MX / 5900XT / Audigy2
PC#4: i5-3570K / MSI Z77A-G43 / GTX 970 / X-Fi

Reply 103 of 217, by bloodem

User metadata
Rank Oldbie
Rank
Oldbie
Joseph_Joestar wrote on 2022-01-15, 13:15:

Interesting read. It could well be that motherboards play a certain role in this as well. That said, I was unable to downclock a Duron 700 in this exact motherboard, so there must be something on the CPU side as well.

Oh, the CPU also plays a big part, for sure! I know this for a fact, because later Bartons are definitely fully locked, even when using the same KT400/KT600/KT880 boards.
So there's definitely something specific to the Thoroughbreds that makes them very flexible (at least when used on certain boards?).

Joseph_Joestar wrote on 2022-01-15, 13:15:

Very true. The KT7A has a "SoftMenu" section in the BIOS which pretty much lets you tweak CPU settings however you want. I've added a picture of that screen here.

Yeah, I know it all too well. The menu is very similar to that of other Abit motherboards that I have, like the ST6 or even the older BH6. Great board you have there, I'm jealous! 😁

1 x PLCC-68 / 2 x PGA132 / 5 x Skt 3 / 9 x Skt 7 / 12 x SS7 / 1 x Skt 8 / 14 x Slot 1 / 5 x Slot A
5 x Skt 370 / 8 x Skt A / 2 x Skt 478 / 2 x Skt 754 / 3 x Skt 939 / 7 x LGA775 / 1 x LGA1155
Current PC: Ryzen 7 5800X3D
Backup PC: Core i7 7700k

Reply 104 of 217, by pixel_workbench

User metadata
Rank Member
Rank
Member
bloodem wrote on 2022-01-15, 10:31:

P4 is terrible when it comes to speed flexibility. There's no way of reaching 386 / 486 speeds, because even when disabling the L1 cache, the Netburst architecture is still too fast.

This is not completely accurate. The reason it appears so is because using "SETMUL L1D" on a P4 actually disables the L2 cache, and not the L1. I tested this on a board that can disable caches in BIOS, and then the P4 slows down to a level of a slow 486. Looking at the memory test graph in Speedsys supports this theory. Below are some DOS benches from my testing.

P4 2.8 Northwood 400MHz FSB, Soyo P4I845PE, 512MB DDR-266, Radeon 9550

Normal:
3dbench1.0c = 460.6
PCPbench = 406.1
Doom fps = 121.45

Using SETMUL L1D:
3dbench1.0c = 333.5
PCPbench = 132.1
Doom fps = 84.78

L1 & L2 caches disabled in BIOS:
3dbench1.0c = 21.5
PCPbench = 6.6
Doom fps = 10.36

My Videos | Website
P2 400 unlocked / Asus P3B-F / Voodoo3 3k / MX300 + YMF718

Reply 105 of 217, by bloodem

User metadata
Rank Oldbie
Rank
Oldbie
pixel_workbench wrote on 2022-01-20, 04:31:

This is not completely accurate. The reason it appears so is because using "SETMUL L1D" on a P4 actually disables the L2 cache, and not the L1. I tested this on a board that can disable caches in BIOS, and then the P4 slows down to a level of a slow 486. Looking at the memory test graph in Speedsys supports this theory. Below are some DOS benches from my testing.

Awesome find, pixel_workbench!
True, I was disabling the L1 cache with setmul, because none of the socket 478 boards that I tested had the BIOS option to do so.
Anyway, although that slow 486 speed is actually very nice for me (I've found that most/all of my favorite early DOS games work just fine at this speed), this doesn't really change the fact that, overall, P4 platforms are still far from being flexible.

1 x PLCC-68 / 2 x PGA132 / 5 x Skt 3 / 9 x Skt 7 / 12 x SS7 / 1 x Skt 8 / 14 x Slot 1 / 5 x Slot A
5 x Skt 370 / 8 x Skt A / 2 x Skt 478 / 2 x Skt 754 / 3 x Skt 939 / 7 x LGA775 / 1 x LGA1155
Current PC: Ryzen 7 5800X3D
Backup PC: Core i7 7700k

Reply 106 of 217, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

Hmmm wonder what it's like with an 800 Mhz FSB P4 when you do L1 off, Pentium class? then maybe L1 off at 533 is 486 and at 400 is nearer 386 (bearing in mind multi lower on 800 chip, so at 400 it's at early willamette clock speeds, but IPC might be higher)

Though I guess I could scare up the P4M 1.8 I thought I had kicking round somewhere, and see how speedstep behaves.

Edit: reviewing my options of hardware to play around with this on, and I don't seem to have a "goldilocks" CPU or motherboard combo, stuff that's either too good or too bad. Too good mainly meaning I've got different plans for it already. In investigating whether a 331 Celerunt would take 200/800Mhz I found that there's an additional FSB step that might be available on some socket 775 boards. Intel specced a 166/667 FSB which they never released a desktop CPU for, but only used on mobile. However, it might have been implemented with correct PCI/AGP dividers in some boards/chipsets, because MFers never knew Intel wouldn't release 166/667 desktop CPUs. Taping over a pad is meant to enable it. Useful to make settings more granular, modular, interactive-odular. Or because it's a huge leap between 533 and 800 if you've got 533 CPUs. 533 to 800 is possible but you need a low multi (slow) 533 CPU that's practically perfect.

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 107 of 217, by BitWrangler

User metadata
Rank l33t++
Rank
l33t++

Set up a new thread for discussion of P4 speed fudging etc.. Methods to enhance retro-flexibility on Pentium 4 class hardware...

Sorry/not sorry for the notifications from the quoting 😜

Unicorn herding operations are proceeding, but all the totes of hens teeth and barrels of rocking horse poop give them plenty of hiding spots.

Reply 108 of 217, by Standard Def Steve

User metadata
Rank Oldbie
Rank
Oldbie
pixel_workbench wrote on 2022-01-20, 04:31:
This is not completely accurate. The reason it appears so is because using "SETMUL L1D" on a P4 actually disables the L2 cache, […]
Show full quote
bloodem wrote on 2022-01-15, 10:31:

P4 is terrible when it comes to speed flexibility. There's no way of reaching 386 / 486 speeds, because even when disabling the L1 cache, the Netburst architecture is still too fast.

This is not completely accurate. The reason it appears so is because using "SETMUL L1D" on a P4 actually disables the L2 cache, and not the L1. I tested this on a board that can disable caches in BIOS, and then the P4 slows down to a level of a slow 486. Looking at the memory test graph in Speedsys supports this theory. Below are some DOS benches from my testing.

P4 2.8 Northwood 400MHz FSB, Soyo P4I845PE, 512MB DDR-266, Radeon 9550

Normal:
3dbench1.0c = 460.6
PCPbench = 406.1
Doom fps = 121.45

Using SETMUL L1D:
3dbench1.0c = 333.5
PCPbench = 132.1
Doom fps = 84.78

L1 & L2 caches disabled in BIOS:
3dbench1.0c = 21.5
PCPbench = 6.6
Doom fps = 10.36

I'm surprised that disabling the L2 doesn't have a much larger impact on performance, given how tiny the P4's L1 is.
Plus, the P4's L2 cache is wicked fast. The difference in bandwidth between L1 and L2 is relatively small on a P4; much smaller than it is on a PIII, K7/K8, or even Pentium M.
Edit: Then again, this is DOS. Perhaps the difference between L2 on and L2 off would be much greater under a multitasking OS.

Last edited by Standard Def Steve on 2022-01-20, 23:00. Edited 1 time in total.

P6 chip. Triple the speed of the Pentium.
Tualatin: PIII-S @ 1628MHz | QDI Advance 12T | 2GB DDR-310 | 6800GT | X-Fi | 500GB HDD | 3DMark01: 14,059
Dothan: PM @ 2.9GHz | MSI Speedster FA4 | 2GB DDR2-580 | GTX 750Ti | X-Fi | 500GB SSD | 3DMark01: 43,190

Reply 109 of 217, by Jasin Natael

User metadata
Rank Oldbie
Rank
Oldbie

I find it amusing that nearly any thread discussing the Ge(fail)force FX inevitably devolves into a P4 discussion thread.
They are a match made in heaven, both can be great but are stuck with a stigma of being terrible.
And I'll confess I have zero use for either product range.

Reply 110 of 217, by The Serpent Rider

User metadata
Rank l33t++
Rank
l33t++

It's probably less about P4 itself and more about Intel S478 platform as a whole. Which is not riddled with bugs, poor I/O performance and other problems, usually associated with VIA, Nvidia and to lesser extend SIS chipsets.

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

Reply 111 of 217, by kolderman

User metadata
Rank l33t
Rank
l33t

The stigma makes it cool now IMO. The fact that something people would have vomited over in 2003 is now arguably the best build for retro gaming for that era is kind of awesome. And in case you forget, the noise roar will constantly remind you 😦

Reply 112 of 217, by The Serpent Rider

User metadata
Rank l33t++
Rank
l33t++

have vomited over in 2003

Except it didn't at that time. P4 became a mess in late 2004, when it got obvious that Intel couldn't deliver scalability they've promised. Prescott was hot garbage.

Last edited by The Serpent Rider on 2022-01-22, 11:58. Edited 1 time in total.

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

Reply 115 of 217, by bloodem

User metadata
Rank Oldbie
Rank
Oldbie

IIRC, late Northwood CPUs were usually leading the performance charts (up until when the Athlon 64 was launched in the Fall of 2003).

That being said, the difference in games between Barton and Northwood was still negligible, while the price difference was considerable. Which is why, I only knew one person who had actually bought a Pentium 4 platform during that time.

Most gamers I knew (myself included) were buying Athlon XP PCs like crazy. And, truthfully, in hindsight, I don't regret it one bit. I've had a great time with my Athlon XP 2200+ on a Biostar M7VIT PRO/VIA KT400 - it was rock solid (don't remember ever experiencing a crash/BSOD), and performance for the time was more than enough (I was usually GPU limited with my Palit Daytona GeForce 3 Ti 200).

1 x PLCC-68 / 2 x PGA132 / 5 x Skt 3 / 9 x Skt 7 / 12 x SS7 / 1 x Skt 8 / 14 x Slot 1 / 5 x Slot A
5 x Skt 370 / 8 x Skt A / 2 x Skt 478 / 2 x Skt 754 / 3 x Skt 939 / 7 x LGA775 / 1 x LGA1155
Current PC: Ryzen 7 5800X3D
Backup PC: Core i7 7700k

Reply 116 of 217, by Kahenraz

User metadata
Rank l33t
Rank
l33t

I wonder if there isn't a significant bias from this time where people had either too little system memory or too much crapware (or both) that coincided with owning a Pentium 4 that skewed people's perceptions. I have even said myself that, despite how anemic my P4 1.4 GHz felt, it was only during the XP era and I was already aware of my memory size problems.

I can remember asking to use other people's computers at their house only to find that it could take 20 minutes or more for it to boot because of all of the garbage that would run at startup and cause the system to swap forever.

Building retro computers with as much memory as I want, I find the experience to be completely different. I think that the Pentium 4 platform is perfectly fine for the era.

Reply 117 of 217, by mockingbird

User metadata
Rank Oldbie
Rank
Oldbie
bloodem wrote on 2022-01-15, 10:31:

I'm not, I just love to benchmark! 😁
Also, I really enjoy jumping from very slow to very fast platforms. A week ago I tested two 486 systems, so you can imagine the feeling I had when I switched to this beast. 😁

You know, I got to thinking recently: If you're going to get a later Radeon card just for the fog support with later Catalyst drivers (10.2 is stated here, but I think it was fixed as early as Catalyst 7.9), what's stopping you from using a an HD59xx series... You're using XP because the fog only works with Catalyst in XP... And thinking about it further, theoretically, an R9 370 should be the fastest card to support table fog and the last card to work in XP (if iCafe driver 9.000.300.3010 also supports fog).

So I don't see the benefit of a 98-compatible PCIe system...

Now if fog worked with Catalyst under Windows 98, that would be a different story. There is a hack for earlier Radeons, but probably not for R4xx cards. This would need further testing.

mslrlv.png
(Decommissioned:)
7ivtic.png

Reply 118 of 217, by Joseph_Joestar

User metadata
Rank l33t
Rank
l33t
mockingbird wrote on 2022-01-28, 20:01:

You know, I got to thinking recently: If you're going to get a later Radeon card just for the fog support with later Catalyst drivers (10.2 is stated here, but I think it was fixed as early as Catalyst 7.9), what's stopping you from using a an HD59xx series... You're using XP because the fog only works with Catalyst in XP...

Catalyst 7.11 was the driver version which introduced proper table fog rendering on ATi cards under WinXP. From that driver version onward, table fog on ATi cards renders exactly the same as it does on Nvidia cards.

Now if fog worked with Catalyst under Windows 98, that would be a different story. There is a hack for earlier Radeons, but probably not for R4xx cards. This would need further testing.

I got table fog working under Win98 on my Radeon 9000 Pro, but it took a lot of effort and a very specific driver version (4.3 WDM). The end result isn't too bad, but it looks visibly different from the Nvidia/3DFX/Matrox table fog rendering.

img

No idea if this hack will work on later Radeon cards though.

PC#1: Pentium MMX 166 / Soyo SY-5BT / S3 Trio64V+ / Voodoo1 / YMF719 / AWE64 Gold / SC-155
PC#2: AthlonXP 2100+ / ECS K7VTA3 / Voodoo3 / Audigy2 / Vortex2
PC#3: Athlon64 3400+ / Asus K8V-MX / 5900XT / Audigy2
PC#4: i5-3570K / MSI Z77A-G43 / GTX 970 / X-Fi

Reply 119 of 217, by mockingbird

User metadata
Rank Oldbie
Rank
Oldbie
Joseph_Joestar wrote on 2022-01-28, 20:16:

I got table fog working under Win98 on my Radeon 9000 Pro, but it took a lot of effort and a very specific driver version (4.3 WDM). The end result isn't too bad, but it looks visibly different from the Nvidia/3DFX/Matrox table fog rendering.

Thanks. Definitely not with R4xx, which probably need a minimum of 6.x drivers... How did you accomplish that, with ATI Tool or registry hacks?

mslrlv.png
(Decommissioned:)
7ivtic.png