Horun wrote on 2023-11-07, 02:28:

Standard Def Steve wrote on 2023-11-07, 00:35:

Aww, that's no fun. But a Northwood Celeron would be just as slow (or worse, even) so I guess the bottleneck fetishist with one of those boards could take that route. 😜
--
Speaking of which, the Willamette and Northwood Celerons are excruciatingly slow in 3DMark 2000. There's just something about that benchmark - I guess the data sets just don't fit at all into 128K. I kid you not, a 1GHz PIII will legit outperform a 2.6GHz Celeron in 3DMark 2000.

Hmm ok so what about Celeron and 3DMark 2001 ?

3DMark 2001 (and '99, for that matter) isn't as cruel to the Celerons, although even in these benchmarks they look mighty weak next to the full-fat P4s.

But 3DMark 2000 is just on another level. I don't believe there exists another benchmark or application that paints Netburst-128 in such a curiously bad light. And the thing is, there's next to zero scaling with clock speed. A 1.7GHz Celeron will only score around 3900 (really, they are that slow) and increasing the clock by a full gigahertz will only net you an additional ~450 points.

So in my tiny mind, the near total lack of scaling indicates that this particular benchmark is constantly overstepping the cache.

BitWrangler wrote on 2023-11-07, 00:33:

Horun wrote on 2023-11-07, 00:15:

Hahaa I am going with my only Willi a 1.7G Celeron if the board supports it and a Geforce 3. Am sure will be down in a benchmark list but will give it a try 😀

If you are quick of the mark you might be able to squeak 13-16,000 out of it fully wrung out and look pretty good until the 6800 and ATI x series crowd wakes up.

Does this mean that my score of 11330 is low ?
I don't have any clue of what could be a good score for this type of system.

I used one of the most recent AGP card I have but the "modern" driver should add overhead.

I can give another try with a 6800NU AGP (with unlocked units). But Nvidia series 6 and7 are more or less the same arch iirc.
I don't think i have AGP ATI series X but i might also try FX5900 and ti4800

acl wrote on 2023-11-07, 07:06:

Does this mean that my score of 11330 is low ? I don't have any clue of what could be a good score for this type of system. […]
Show full quote

BitWrangler wrote on 2023-11-07, 00:33:

Horun wrote on 2023-11-07, 00:15:

Hahaa I am going with my only Willi a 1.7G Celeron if the board supports it and a Geforce 3. Am sure will be down in a benchmark list but will give it a try 😀

If you are quick of the mark you might be able to squeak 13-16,000 out of it fully wrung out and look pretty good until the 6800 and ATI x series crowd wakes up.

Does this mean that my score of 11330 is low ?
I don't have any clue of what could be a good score for this type of system.

I used one of the most recent AGP card I have but the "modern" driver should add overhead.

I can give another try with a 6800NU AGP (with unlocked units). But Nvidia series 6 and7 are more or less the same arch iirc.
I don't think i have AGP ATI series X but i might also try FX5900 and ti4800

Possibly is slow. I am not sure where the fault lies. I would guess that the earliest drivers for cards "past" a CPU's era are more likely to score best on older benchmarks. Later in their life the drivers may be optimized for newer CPUs and benchmarks/games. We notice a thing with 300-900Mhz processors where score seems to peak with GF2/3 and drop off with "fast" GPU. In later generations still DX8 seems very badly optimized and 3DMark2001 scores don't look very good. However 2001 also likes bus speed and penalises SDRAM, when I was thinking of that max potential score with the GF3, I was thinking of the FSB being up over 133/533 and the core and RAM tweaked out to max on GF3, something like 220/560... which also needs a "lucky" GF3.

BitWrangler wrote on 2023-11-07, 12:52:

acl wrote on 2023-11-07, 07:06:

Does this mean that my score of 11330 is low ? I don't have any clue of what could be a good score for this type of system. […]
Show full quote

BitWrangler wrote on 2023-11-07, 00:33:

If you are quick of the mark you might be able to squeak 13-16,000 out of it fully wrung out and look pretty good until the 6800 and ATI x series crowd wakes up.

Does this mean that my score of 11330 is low ?
I don't have any clue of what could be a good score for this type of system.

I used one of the most recent AGP card I have but the "modern" driver should add overhead.

I can give another try with a 6800NU AGP (with unlocked units). But Nvidia series 6 and7 are more or less the same arch iirc.
I don't think i have AGP ATI series X but i might also try FX5900 and ti4800

Possibly is slow. I am not sure where the fault lies. I would guess that the earliest drivers for cards "past" a CPU's era are more likely to score best on older benchmarks. Later in their life the drivers may be optimized for newer CPUs and benchmarks/games. We notice a thing with 300-900Mhz processors where score seems to peak with GF2/3 and drop off with "fast" GPU. In later generations still DX8 seems very badly optimized and 3DMark2001 scores don't look very good. However 2001 also likes bus speed and penalises SDRAM, when I was thinking of that max potential score with the GF3, I was thinking of the FSB being up over 133/533 and the core and RAM tweaked out to max on GF3, something like 220/560... which also needs a "lucky" GF3.

I picked the older driver i had that was compatible with the 7900GS but there should be older drivers available online. I just used the older i had.
I will try other GPU and drivers our of curiosity.

Standard Def Steve wrote on 2023-11-07, 00:35:

VivienM wrote on 2023-11-06, 23:04:

Standard Def Steve wrote on 2023-11-06, 16:31:

I believe Asus made a 915 based S478 board. It'd be neat to seat a Willamette in one of those and let 'er rip alongside an RTX 2080 Ti.

Somebody already inventoried all these boards - Socket 478 Motherboards with PCI-E Slots . Maybe I'm missing something but I can't see which ones the OP identified as not supporting Willamette.

I did check the supported CPU list for the Asus P4GD1, no Willamettes listed... 🙁

Aww, that's no fun. But a Northwood Celeron would be just as slow (or worse, even) so I guess the bottleneck fetishist with one of those boards could take that route. 😜

The Northwood Celeron, will probably clock way higher, though ... 😉

BTW, do "we" have some smart nicknames for the P4 Celerons like the Celemine, Tualeron etc.?

If not, I'd suggest CeleWood for the Northwood-128 🤣

Celemette?

H3nrik V! wrote on 2023-11-07, 16:22:

BTW, do "we" have some smart nicknames for the P4 Celerons like the Celemine, Tualeron etc.?

Back in the day, I think the Prescott Celeron (which IIRC upped the cache to 256K and was less bad as a result) D was regularly referred to as a Deleron.

BitWrangler wrote on 2023-11-07, 12:52:

acl wrote on 2023-11-07, 07:06:

Does this mean that my score of 11330 is low ? I don't have any clue of what could be a good score for this type of system. […]
Show full quote

BitWrangler wrote on 2023-11-07, 00:33:

If you are quick of the mark you might be able to squeak 13-16,000 out of it fully wrung out and look pretty good until the 6800 and ATI x series crowd wakes up.

Does this mean that my score of 11330 is low ?
I don't have any clue of what could be a good score for this type of system.

I used one of the most recent AGP card I have but the "modern" driver should add overhead.

I can give another try with a 6800NU AGP (with unlocked units). But Nvidia series 6 and7 are more or less the same arch iirc.
I don't think i have AGP ATI series X but i might also try FX5900 and ti4800

Possibly is slow. I am not sure where the fault lies. I would guess that the earliest drivers for cards "past" a CPU's era are more likely to score best on older benchmarks. Later in their life the drivers may be optimized for newer CPUs and benchmarks/games. We notice a thing with 300-900Mhz processors where score seems to peak with GF2/3 and drop off with "fast" GPU. In later generations still DX8 seems very badly optimized and 3DMark2001 scores don't look very good. However 2001 also likes bus speed and penalises SDRAM, when I was thinking of that max potential score with the GF3, I was thinking of the FSB being up over 133/533 and the core and RAM tweaked out to max on GF3, something like 220/560... which also needs a "lucky" GF3.

I tried with an FX5900XT @~Ultra (500/410) and with older drivers (FW53.03)
I got a better score in 3DM2000 but a bit less in 2001SE
I was able to push the Willamette 50Mhz higher with the FX5900, it is probably more tolerent to AGP freq. But the 50Mhz themselves cannot explain the better results in 3DM2000.

The system became quite unstable with my 6800NU and would require a fresh Windows install... and i don't have time to re-install everything. But i would be interested to see results from this card.

I think with GeForce 6/7, we're right on a turning point. When these late AGP card would theoretically give good results, but drivers are not optimized at all for "old" CPUs like the P4 Willamette.
I'm curious to see other results to see if their card is held back by the P4.

Uploaded my FX5900 scores, just as example.

acl wrote on 2023-11-07, 23:57:

I tried with an FX5900XT @~Ultra (500/410) and with older drivers (FW53.03) I got a better score in 3DM2000 but a bit less in 20 […]
Show full quote

BitWrangler wrote on 2023-11-07, 12:52:

acl wrote on 2023-11-07, 07:06:

Does this mean that my score of 11330 is low ? I don't have any clue of what could be a good score for this type of system. […]
Show full quote

Does this mean that my score of 11330 is low ?
I don't have any clue of what could be a good score for this type of system.

I used one of the most recent AGP card I have but the "modern" driver should add overhead.

I can give another try with a 6800NU AGP (with unlocked units). But Nvidia series 6 and7 are more or less the same arch iirc.
I don't think i have AGP ATI series X but i might also try FX5900 and ti4800

Possibly is slow. I am not sure where the fault lies. I would guess that the earliest drivers for cards "past" a CPU's era are more likely to score best on older benchmarks. Later in their life the drivers may be optimized for newer CPUs and benchmarks/games. We notice a thing with 300-900Mhz processors where score seems to peak with GF2/3 and drop off with "fast" GPU. In later generations still DX8 seems very badly optimized and 3DMark2001 scores don't look very good. However 2001 also likes bus speed and penalises SDRAM, when I was thinking of that max potential score with the GF3, I was thinking of the FSB being up over 133/533 and the core and RAM tweaked out to max on GF3, something like 220/560... which also needs a "lucky" GF3.

I tried with an FX5900XT @~Ultra (500/410) and with older drivers (FW53.03)
I got a better score in 3DM2000 but a bit less in 2001SE
I was able to push the Willamette 50Mhz higher with the FX5900, it is probably more tolerent to AGP freq. But the 50Mhz themselves cannot explain the better results in 3DM2000.

The system became quite unstable with my 6800NU and would require a fresh Windows install... and i don't have time to re-install everything. But i would be interested to see results from this card.

I think with GeForce 6/7, we're right on a turning point. When these late AGP card would theoretically give good results, but drivers are not optimized at all for "old" CPUs like the P4 Willamette.
I'm curious to see other results to see if their card is held back by the P4.

Uploaded my FX5900 scores, just as example.
3DMark2000-FX5900.png3DMark2001SE-FX5900.png

Updated your your score for 3dmark2k. You dont have to use the same gpu for both benchmarks. You dont have to submit scores for both benchmarks.

Oh, this is gonna be fun. I only have an AGP motherboard for P4. The Ti4200 I have, apparently is in my storage, so as of now, the options are TNT2 M64 or Intel i740 🤣 hoping to get to building tonight ...

Found another Celery in my stuff but is a Northwood 2Ghz not a Willy so the 1.7 is all I got 🙁...It will take a few days to get mine together due to work and family, hoping by Sat or Sun.....
added: found the 1.8Ghz willy (knew I had it somewhere if my cpu list was correct). so maybe can do two (1.7 Celery and 1.8 P4) tests...

I've wanted to build a 423 Willamette system for a while now. Just for the pointless fun of it.

My S478 1.9 Willy and Asrock P4VM890 arrived.

Time to pair a needlessly modern PCIe card with a Willamette.

paradigital wrote on 2023-11-10, 17:08:

My S478 1.9 Willy and Asrock P4VM890 arrived.

Time to pair a needlessly modern PCIe card with a Willamette.

Can't wait to see how it performs.
At least DDR should be better than RDRAM...

acl wrote on 2023-11-10, 18:19:

paradigital wrote on 2023-11-10, 17:08:

My S478 1.9 Willy and Asrock P4VM890 arrived.

Time to pair a needlessly modern PCIe card with a Willamette.

Can't wait to see how it performs.
At least DDR should be better than RDRAM...

Wasn't the performance of RDRAM quite excellent? And the main problem was just the... price?

Just failed XP install on my Asus P4P800 SE ... I'm too lazy to slipstream SATA drivers into the ISO, and I don't have any PATA harddisks at hand ... Getting ready to RUFUS a Win7 boot - hoping the hardware can keep up ...

Didn’t go well. Two caps literally exploded on first power-on. Was supposed to be a NOS board, and certainly looked it on first glance.

Disappointed 🙁

VivienM wrote on 2023-11-10, 18:31:

acl wrote on 2023-11-10, 18:19:

paradigital wrote on 2023-11-10, 17:08:

My S478 1.9 Willy and Asrock P4VM890 arrived.

Time to pair a needlessly modern PCIe card with a Willamette.

Can't wait to see how it performs.
At least DDR should be better than RDRAM...

Wasn't the performance of RDRAM quite excellent? And the main problem was just the... price?

Not sure, so please, correct me if I'm wrong.
I read that while RDRAM was generally better than SDRAM (especially with speed above 800) it was more or less on par or a bit below DDR in most use cases due to higher latency.

I never experienced this back in the days because I still had a P3 during early P4 era and bought an Athlon64 later.
And I don't run any other retro P4 today 😅 Just my RDRAM Willamette.

paradigital wrote on 2023-11-10, 18:41:

Didn’t go well. Two caps literally exploded on first power-on. Was supposed to be a NOS board, and certainly looked it on first glance.

Disappointed 🙁

Same as my first s423 board.
I was refund and had to find another one.
2000-2003 hardware don't age very well...

paradigital wrote on 2023-11-10, 18:41:

Didn’t go well. Two caps literally exploded on first power-on. Was supposed to be a NOS board, and certainly looked it on first glance.

Disappointed 🙁

Happens a lot. Capps can still go bad even if the board wasn't used. Sucks, hopefully you can replace them.