gerwin wrote:

meljor wrote:

But I agree, i think every p3 above 1200mhz is a p3-s but i could be wrong. I think normal p3 stopped at 1133mhz and from there p3-s took over.

I found a picture of a 1400MHz one (SL64W):
http://chook.x86-guide.com/en/collection/Inte … pu-no10561.html

Yes, nice find. And as Tetrium said there are others as well. According to wiki there is a tualatin 1ghz 512kb up to 1400mhz 512kb (ofcourse) but also a 1ghz tualatin 256kb up to 1400mhz 256kb (the one you found).

So I stand corrected. I didn't know. Most of the time you see 1200mhz+ tualatins and since a lot of them come from servers they have 512kb cache.

I like the tualatin celerons for 440BX boards as they are fast without overclocking anything at 100mhz fsb.

I never gave it much thought before. Practically all that is different between these 256kB PIIIs and Tualerons is the stock multiplier and the FSB identification. FSB identification is easily changed with a motherboard setting or a slotket jumper. That leaves the multiplier it comes with, and the Tualerons have the highest ones of course.
There is one unlocked Pentium III-S 1266 here, but it does not support anything beyond multiplier 9.5x, just many lower ones like 4.0x. That is fine for me, as anything above like 1GHz seems unnecessary for my use cases, and also does not translate into much benefit in my setup. In the past I actually benchmarked it:

1Motherboard:      Soyo SY-6BA+III i440BX, FSB set at 105MHz. CL3 memory.
2Graphics Card:    AGP Abit Siluro Geforce MX440 128-bit 64MB
3Operating System: Windows 98SE
4Tualatin-S, SL5QL, @1000MHz, 3Dmarks2001SE:  3420 
5Tualeron,   SL5ZJ, @1366MHz, 3Dmarks2001SE:  3440 

appiah4 wrote:

Is the PIII-S 1400 anything rare? Because I can grab one for around $20. Should I? And is it compatible on all Tualatin boards?

I'd guess it's about as rare as K6-III+/400 ATZ 😁

(so not really rare at all, which is actually a good thing for us retro enthusiasts 😀)

Yes go grab a couple! They are basically the best you can get for s370 (if you can make it work). But do keep in mind the things that are already mentioned here previously about Tualatin requiring some special attention due to them not being as compatible compared to Coppermines).

Not rare on Ebay. Same thing with k6-3+ as here in NL you can't find them anywhere.

And i know because i keep looking for a spare k6-3+.

gerwin wrote:

I never gave it much thought before. Practically all that is different between these 256kB PIIIs and Tualerons is the stock mult […]
Show full quote

I never gave it much thought before. Practically all that is different between these 256kB PIIIs and Tualerons is the stock multiplier and the FSB identification. FSB identification is easily changed with a motherboard setting or a slotket jumper. That leaves the multiplier it comes with, and the Tualerons have the highest ones of course.
There is one unlocked Pentium III-S 1266 here, but it does not support anything beyond multiplier 9.5x, just many lower ones like 4.0x. That is fine for me, as anything above like 1GHz seems unnecessary for my use cases, and also does not translate into much benefit in my setup. In the past I actually benchmarked it:

Copy code to clipboard

1Motherboard:      Soyo SY-6BA+III i440BX, FSB set at 105MHz. CL3 memory.
2Graphics Card:    AGP Abit Siluro Geforce MX440 128-bit 64MB
3Operating System: Windows 98SE
4Tualatin-S, SL5QL, @1000MHz, 3Dmarks2001SE:  3420 
5Tualeron,   SL5ZJ, @1366MHz, 3Dmarks2001SE:  3440 

Thank you . This is precisely the comparison I was looking for . The main advantage I see in running a downclocked Tualatin-S (other than lower TDP) is the fact that you can slow it down further than a Tualeron by disabling caches and lowering FSB ( 693 MHz vs 858 MHz when comparing a Tualatin-S @1400 vs Tualeron@1300) . Whether this usefully significant is something I have yet to try (my BX-based is still a work in progress and my Tualatin-S is in the mail).

The conclusion one can make from these two benchmarks is that memory access is the bottleneck. In this use case the 105 MHz bus with CL3 SDRAM is insufficient for CPU speeds over 1 GHz. I had one other benchmark where I put in CL2 memory: it added 180 3Dmark points.
The benchmark that Meljar quoted though: there is actually some benefit measured there when going from a 1000 MHz Tualeron to a 1300 MHz one. Is it the graphics card used?
I would like to see more benchmarks, but I only have DOS retro systems ready at the moment. Maybe in the future, as I expect to receive an unlocked coppermine-128kB, I can then benchmark at least four Pentium IIIs at the exact same clock speeds, with the only difference being their L2-cache size: 128kB Celeron Coppermine, 256kB Coppermine, 256kB Tualeron and 512kB Tualatin.

When you disable L1 cache it is actually the same thing: main memory access time / FSB speed matters more then CPU speed.

Iirc the Tualeron also had extra waitstates or something in addition to having less cache and a lower FSB?
But did the vanilla Tualatin (the 133MHz FSB 256kb cache one) also have these extra waitstates?

gerwin wrote:

The conclusion one can make from these two benchmarks is that memory access is the bottleneck. In this use case the 105 MHz bus […]
Show full quote

The conclusion one can make from these two benchmarks is that memory access is the bottleneck. In this use case the 105 MHz bus with CL3 SDRAM is insufficient for CPU speeds over 1 GHz. I had one other benchmark where I put in CL2 memory: it added 180 3Dmark points.
The benchmark that Meljar quoted though: there is actually some benefit measured there when going from a 1000 MHz Tualeron to a 1300 MHz one. Is it the graphics card used?
I would like to see more benchmarks, but I only have DOS retro systems ready at the moment. Maybe in the future, as I expect to receive an unlocked coppermine-128kB, I can then benchmark at least four Pentium IIIs at the exact same clock speeds, with the only difference being their L2-cache size: 128kB Celeron Coppermine, 256kB Coppermine, 256kB Tualeron and 512kB Tualatin.

When you disable L1 cache it is actually the same thing: main memory access time / FSB speed matters more then CPU speed.

Here's my P6 comparison thread. Celeron-1400 and overclocked PIII-S are in it.
Intel P6 Comparison with charts + additional CPUs

The conclusion one can make from these two benchmarks is that memory access is the bottleneck. In this use case the 105 MHz bus with CL3 SDRAM is insufficient for CPU speeds over 1 GHz

While memory bandwidth is most definitely the issue, the bigger cache on the Tualatin-S seems to essentially offset a 366 Mhz CPU clock difference, all other things being equal, at least in terms of 3DMark 2001 performance (unless the MX440 is the bottleneck) .

When you disable L1 cache it is actually the same thing: main memory access time / FSB speed matters more then CPU speed.

I agree, but I would still expect a lower CPU clock to still have some effect when trying to achieve the greatest slowdown possible

darry wrote:

the bigger cache on the Tualatin-S seems to essentially offset a 366 Mhz CPU clock difference

Find it hard to believe, need a better benchmark. Problem: I have loads of lower-end graphics cards, but no 'fast' Windows 98 compatible ones.

gerwin wrote:

darry wrote:

the bigger cache on the Tualatin-S seems to essentially offset a 366 Mhz CPU clock difference

Find it hard to believe, need a better benchmark. Problem: I have loads of lower-end graphics cards, but no 'fast' Windows 98 compatible ones.

As soon as I get my 1400 MHz Tualatin-S , I will be able able to compare it running at 1050 MHz to my 1300 Mhz Tualeron on a BX board (100 MHz FSB, 512MB CL2 SDRAM)
I am currently using a Quadro FX1100 (FX5700 derivative), but also have a Radeon 9700 (somewhere).

1300 Mhz Tualeron on a BX board (100MHz FSB) with Quadro FX1100 : 7002 3D marks in 3Dmark 2001

@ Standard Def Steve
Thanks for the link, I looked over it at first though. At first glance: Graphics card is plenty fast here. but FSB difference between the tualatins makes it an unequal comparison to answer the question: How much of an FPS increase does the extra cache provide.

darry wrote:

As soon as I get my 1400 MHz Tualatin-S , I will be able able to compare it

Looking forward to it. 😀

PS. I have an exotic benchmark here, but it is just for Quake 1 / PC Player Bench: Multiplier 4.0x Challenge

Thanks for the very resourceful discussion.

I have a tangential question. I wanted to drop in a Radeon 8500 to my Tualatin build for period correctness but I cant seem to find any for a reasonable price. I instead grabbed a 9600XT and a 9800PRO for five bucks each. Is the 9800PRO way overkill for a Tualatin or eould I see improvement over the 9600XT despite the CPU bottleneck?

melbar wrote:

The difference between the Celeron-Tualatin and the "-S" Tualatin is: […]
Show full quote

The difference between the Celeron-Tualatin and the "-S" Tualatin is:

• - higher FSB clock
  - Data prefech
  - cache latency

Be aware that the "-S" Tualatin was a server CPU. Not for the normal home market.

Regarding the market back in the day, i would compare the Celeron-Tualatin with a high-clocked P3-Coppermine or the P3-Tualatin (with 256kb cache).

Back in the day, the Celeron Mendocino compete with a PII, not with the PIII.

Tualatin Celeron does have Data prefech, but the slower 100 mhz FSB doesn't help with that feature, about the cache, i don't know really if there a difference or it's higher latency comapred to P3, the both the P3 Tualatin (both 256 and 512 KB versions) and the Tualatin Celeron has the same cache asociativity

It would be interesting a Tualatin comparasion, Celeron 1400 vs Pentium 3 1400 vs Pentium 3-S 1400

i remember trying 7-Zip's benchmark tool in some CPUs and my Celeron 1300 outperformed a P4 2.0A GHz with PC133 and came close to the same CPU on DDR on that bench, some games like Quake 3 benefies more form P4's high bandwith of the FSB and RAM. So this is why the Willamette Celeron beats the Tualatin there

Carlos S. M. wrote:

Tualatin Celeron does have Data prefech,

Do you have a link to some tests? I have found a lot of discussions from 2003 to 2004. In one forum, they say finally it's data-prefech is disabled...

melbar wrote:

Carlos S. M. wrote:

Tualatin Celeron does have Data prefech,

Do you have a link to some tests? I have found a lot of discussions from 2003 to 2004. In one forum, they say finally it's data-prefech is disabled...

on this thread, someone contacted Intel Support and they replied that the Celeron has all the same basic features as the P3 tualatin and then DPL is a basic tualatin feature, the only difference seems to be the lower FSB and the higher cache latency. Also they mention about xbitlabs review of the tualatin Celeron confirming they ineeded have DPL as well, sadly, xbitlabs site is down
http://www.overclockers.com/forums/showthread … ion-Accelerator

http://www.overclockers.com/forums/showthread … and-p3-tualatin

http://www.overclockers.com/forums/showthread … III-a-Tualatin/

many forums say is enabled, and some other forums says they are not aparently

Got a new graphics card so I could redo my benchmarks properly.

3DMark2001SE build 330 - Full run at default settings 1024x768x32 NoAA:

Celeron 'Coppermine-128' QBC6ES, 128kB L2 (05-2017 benchmarks)
_7.5 x 100 = _750 MHz: 4987 3DMarks
_8.5 x 100 = _850 MHz: 5254 3DMarks (default setting)
_9.5 x 100 = _950 MHz: 5478 3DMarks
_7.5 x 133 = 1000 MHz: 6294 3DMarks

Pentium III 'Coppermine' SL5FQ, 256kB L2 (05-2017 benchmarks)
_7.5 x 100 = _750 MHz: 5300 3DMarks
_7.5 x 133 = 1000 MHz: 6593 3DMarks (default setting)

Celeron 'Tualeron' QID2QS, 256kB L2
_7.5 x 100 = _750 MHz: 5425 3DMarks (05-2017: 5389 3DMarks)
_9.5 x 100 = _950 MHz: 6187 3DMarks
_9.5 x 133 = 1266 MHz: 7506 3DMarks
12.0 x 100 = 1200 MHz: 6536 3DMarks
14.0 x 100 = 1400 MHz: 6938 3DMarks (default setting)

Pentium III-S 'Tualatin' QEL8ES, 512kB L2
_7.5 x 100 = _750 MHz: 5893 3DMarks
_9.5 x 100 = _950 MHz: 6589 3DMarks
_9.5 x 133 = 1266 MHz: 7845 3DMarks (default setting)

VIA C3 Nehemiah+ ( 6.9.8 )
_7.5 x 100 = _750 MHz: 3272 3DMarks
_9.5 x 100 = _950 MHz: 3828 3DMarks
_9.5 x 133 = 1266 MHz: 4857 3DMarks (had to raise the voltage to 1.6V)
_9.0 x 133 = 1200 MHz: 4814 3DMarks (default setting)

Test System
Motherboard: GA-6BXC rev 2; i440BX chipset; AGP divider 2/3; Upgradeware BIOS. Either Slot-T or MS6905 Slotket.
Memory: 128MB SDRAM; a single DIMM at CL3.
Disk Drive: SD-Card 8GB on IDE adapter; southbridge IDE controller.
Operating System: Windows 98 SE, clean install; DirectX 9.0c.
Graphics Card: Asus V8420/TD Geforce4 Ti4200 AGP; 128MB DDR; Clocks 250/550; Driver v30.82 at default settings. Obtained today from MVZ ICT.

As soon as I get my 1400 MHz Tualatin-S , I will be able able to compare it running at 1050 MHz to my 1300 Mhz Tualeron on a BX board (100 MHz FSB, 512MB CL2 SDRAM)
I am currently using a Quadro FX1100 (FX5700 derivative), but also have a Radeon 9700 (somewhere).

1300 Mhz Tualeron on a BX board (100MHz FSB) with Quadro FX1100 : 7002 3D marks in 3Dmark 2001

This under Windows 98SE .

1050 Mhz Tualatin-S on a BX board (100MHz FSB) with Quadro FX1100 : 7124 3D marks in 3Dmark 2001

I started a separate thread about some issues with the CPU --> Broken pin on socket 370 CPU. Still works but is it reliable ?

That extra/faster cache really does seem to offset a 250 MHz clock difference .
Too bad this board (Biostar M6TBD) does not go over 100 MHz FSB .

darry wrote:

This under Windows 98SE . […]
Show full quote

This under Windows 98SE .

1050 Mhz Tualatin-S on a BX board (100MHz FSB) with Quadro FX1100 : 7124 3D marks in 3Dmark 2001

I started a separate thread about some issues with the CPU --> Broken pin on socket 370 CPU. Still works but is it reliable ?

That extra/faster cache really does seem to offset a 250 MHz clock difference .
Too bad this board (Biostar M6TBD) does not go over 100 MHz FSB .

You can always try out H-Oda!'s SoftFSB for 133MHz FSB, it looks like the clock generator would be the ICS 9148xx-47. Be aware that both AGP and PCI overclocks 33% too, there's no /4 divider for PCI on this generator.

The executable included is a self extracting archive.

gerwin wrote:

Got a new graphics card so I could redo my benchmarks properly. ... Motherboard: GA-6BXC rev 2; i440BX chipset; AGP divider 2/ […]
Show full quote

Got a new graphics card so I could redo my benchmarks properly.
...
Motherboard: GA-6BXC rev 2; i440BX chipset; AGP divider 2/3; Upgradeware BIOS. Either Slot-T or MS6905 Slotket.
...

Interesting values! With you slotket you are able to set the multi down, right?

According to your 3Dmark values at 133FSB, the tualeron at 10x140=1.4Ghz should be >= your PIII-S 1.26GHz.