VOGONS

Fast Pentium MMX compared to a Pentium II

Topic actions

First post, by SSTV2

Posted on 
User metadata
Rank Oldbie
Rank
Oldbie

I have long wanted to know how a heavily tweaked Pentium MMX system, from the year of 1997, would compare to Pentium II systems of the time. I recently had the opportunity to build one such Pentium MMX PC, tweak and test it 😀

Specs of the Pentium MMX system:

MB - ASUS VX97 w/buck converter, i430VX, 512KB L2;
CPU - Pentium MMX 233@292MHz / AMD K6-2 500MHz @83.3MHz FSB;
RAM - 64MB EDO;
*RAM TIMINGS: DRAM Read/Write Burst Timing x222, RAS to CAS 3T, RAS Precharge 3T;
Video - EvilKing Voodoo 3 PCI @167MHz;
Sound - SB AWE64 Gold;
BIOS - Modernized Award BIOS.

Specs of the PII system:

MB - ACER V66XA, i440BX;
CPU - Pentium II Klamath 266MHz / Pentium III Katmai 600MHz;
RAM - 256MB SDRAM;
*RAM TIMINGS @ 100MHz FSB: CAS 3T, RAS to CAS 2T, RAS Precharge 2T;
*RAM TIMINGS @ 66MHz FSB: CAS 2T, RAS to CAS 2T, RAS Precharge 2T;
Video - Voodoo 3 2000 PCI @167MHz;
Sound - Integrated Crystal CS4235;
BIOS - ACER V3.2 /w Katmai CPU uCode.

I decided to do the comparison by benchmarking only Unreal, because the performance of this game heavily depends on the CPU and the bandwidth of the system bus. The given frame rate average and min/max will allow us to directly compare CPUs of different generations and manufacturers.

Testing conditions and environment on both systems:

Win98SE, sound drivers disabled, video driver V1.07.00, USB - Enabled, Unreal ver. 226, GameRenderDevice - 3dfx Glide, default Glide Rendering settings, 640x480 resolution, texture detail - High, monsters disabled at intro Castle Flyby for frame rate consistency, results are collected after 2nd cycle and rounded to its nearest tenth.

*** PII system results:


PII KLAMATH @133MHz

avg 23.6
min 8.6
max 51.4
---
PII KLAMATH @166 MHz

avg 29.2
min 10.9
max 64.9
---
PII KLAMATH @200MHz

avg 34.7
min 12.9
max 76.5
---
PII KLAMATH @233MHz

avg 39.5
min 13.3
max 85.3
---
PII KLAMATH @266MHz

avg 44.2
min 16.2
max 95.2
---
PII KLAMATH @300 MHz, @66MHZ FSB, CL2-3-3

avg 47.6
min 18.1
max 102.4
---
PII KLAMATH @300 MHz, @66MHZ FSB

avg 48.4
min 18.3
max 104.9
---
PII KLAMATH @300MHz, @75MHZ FSB

avg 49.8
min 18.2
max 105.5
---
PII KLAMATH @333MHz, @83.3MHZ FSB

avg 55.5
min 20.8
max 116.6
---
PII DESCHUTES @350 MHz

avg 60.4
…Show last 151 lines
min 22.9
max 124.8
---
PII KLAMATH @263MHz, @75MHZ FSB

avg 44.3
min 16.7
max 93.9
---
PII KLAMATH @292MHz, @83.3MHZ FSB

avg 48.9
min 18.2
max 105.1
---
Celeron COVINGTON @233 MHz, CL2-3-3

avg 32.2
min 11.3
max 65.1
---
Celeron COVINGTON @233 MHz

avg 33.3
min 11.7
max 71.9
---
Celeron COVINGTON @266 MHz, CL2-3-3

avg 34.6
min 12.8
max 73.6
---
Celeron COVINGTON @266 MHz

avg 36.2
min 12.8
max 75.8
---
Celeron COVINGTON @300 MHz, @66MHZ FSB, CL2-3-3

avg 36.8
min 13.2
max 75.8
---
Celeron COVINGTON @300 MHz, @66MHZ FSB

avg 38.4
min 13.3
max 79.9
---
Celeron COVINGTON @333 MHz

avg 40.8
min 14.4
max 84.3
---
Celeron COVINGTON @350 MHz, @100MHZ FSB

avg 49.0
min 18.4
max 100.6
---
Celeron COVINGTON @400 MHz, @100MHZ FSB

avg 52.8
min 19.4
max 108.8
---
Celeron COVINGTON @448 MHz, @112MHZ FSB, CL3-2-2

avg 59.4
min 21.8
max 122.0
---
Celeron MENDOCINO @333 MHz

avg 54.1
min 20.1
max 109.9
---
Celeron MENDOCINO @375MHz, @75MHZ FSB

avg 60.0
min 23.0
max 122.7
---
Celeron MENDOCINO @417MHz, @83.3MHZ FSB

avg 66.4
min 25.5
max 138.2
---
Celeron MENDOCINO @400MHz, @66MHz FSB

avg 60.9
min 24.5
max 125.7
---
Celeron MENDOCINO 450MHz, @75MHz FSB

avg 68.5
min 26.7
max 139.1
---
Celeron MENDOCINO 500MHz, @83.3MHz FSB

avg 76.0
min 28.8
max 158.0
---
Celeron MENDOCINO 600MHz, @100MHz FSB

avg 87.1
min 34.6
max 176.4
---
PIII KATMAI @600MHz
 
avg 85.6
min 32.3
max 184.4
---

*** Pentium MMX system results: 


AMD K5 PR166 @116MHz, RAS to CAS - 2T

avg 18.6
min 6.4
max 42.2
---
Pentium MMX @233MHz, RAS to CAS - 2T

avg 29.3
min 10.5
max 63.0
---
Pentium MMX @292MHz

avg 36.1
min 12.9
max 75.9
---
AMD K6-2 @500MHz, WRITE ALLOCATION ENABLED

avg 46.9
min 18.3
max 99.6
---
The attachment Socket_7.png is no longer available
The attachment Celeron_M_&_PIII_K.png is no longer available
The attachment Celeron_C.png is no longer available
The attachment Pentium_II_K&D.png is no longer available

Conclusions:

Now we can see that a nearly maxed out Pentium MMX PC back in 1997 would deliver us just a slightly higher average performance than that of a PII Klamath clocked at 200MHz and a future upgrade to a 500MHz K6-2 CPU would guarantee a slightly higher performance than that of a PII Klamath clocked at 266MHz.

Last edited by SSTV2 on 2023-06-05, 18:53. Edited 2 times in total.

Reply 1 of 13, by auron

Posted on 
User metadata
Rank Oldbie
Rank
Oldbie

why skip stock results for pentium mmx? most people didn't overclock anyway. also if just comparing the CPUs i wouldn't have used 256mb on the BX board because with 64 there is potentially some swapping in unreal. 440BX is a year later anyway, a proper 1997 comparison would be 440LX with a celeron 533 as the fastest upgrade option.

i think no one should be surprised that p6 outperforms p5 by a lot in unreal/UT, even the readme for those games mentions this. however if you go pre-1998 there also exist games that see no benefit or even run slower clock-for-clock on p6, making those much more expensive and power hungry klamaths look not so great.

Reply 2 of 13, by SSTV2

Posted on 
User metadata
Rank Oldbie
Rank
Oldbie

The main goal of this comparison was to determine the equivalent performance of a P6-class CPU compared to a P5-class CPU, that is pushed to its limits on a period correct socket 7 motherboard. I could redo the benchmark at a stock 233MHz or much slower Pentiums/K5s/K6s/Cyrix 6x86s, just point out what interests you?

Regarding the RAM, I made sure that there would be no HDD activity during benchmarks, if it so happened during 2nd cycle, I'd just restart the timedemo (technically, there were no swapping even on a 64meg system 😉).

About the chipset - 440BX is just an updated 440LX with some features added and others removed, it performs the same as a 440LX at 66MHz FSB, unless RAM timings are set different. Celeron A @ 533MHz beats any Klamath, I just don't see any point of comparing it to a P1, it has no chance of getting anywhere close to a such Celly's results.

Reply 4 of 13, by A001

Posted on 
User metadata
Rank Newbie
Rank
Newbie

233Mhz vs 233Mhz would be interesting too. I've got MMX233 430HX and PII233 440FX both sporting EDO SIMMs but no idea how much faster PII is.

Reply 6 of 13, by SSTV2

Posted on 
User metadata
Rank Oldbie
Rank
Oldbie

I have updated original post with more results from Celeron Covington/Mendocino, PII Deschutes CPUs at different frequencies, also included Pentium MMX @233MHz.

Now we can see that a 300MHz Klamath @66MHz FSB and with lowered RAM timings now performs essentially the same as a 500MHz AMD K6-2 on a i430VX based motherboard.

swaaye wrote on 2023-06-03, 03:03:

One of my favorite old sites comes to mind:
https://web.archive.org/web/19980220065941/ht … .com/bench.html

Intel's Tillamook Pentium also blurred the lines. Which is also like what Pentium M was doing to Pentium 4 a few years later.

Those Doom v1.9 benchmark results can't get any more authentic.

Garrett W wrote on 2023-06-03, 08:51:

You might like this thread:

Pentium Pro 233 vs Pentium MMX 233 vs Pentium II 233

Yes, I have seen it 😉

Reply 7 of 13, by AlexZ

Posted on 
User metadata
Rank Oldbie
Rank
Oldbie

It would be great if you could test Mendocino Celeron at 83Mhz FSB. I used that back in the day instead of K6-2 for gaming.

Pentium III 900E,ECS P6BXT-A+,384MB,GeForce FX 5600, Voodoo 2,Yamaha SM718
Athlon 64 3400+,Gigabyte GA-K8NE,2GB,GeForce GTX 275,Audigy 2 ZS
Phenom II X4 955,Gigabyte GA-MA770-UD3,8GB,GeForce GTX 780
Vishera FX-8370,Asus 990FX,32GB,GeForce GTX 980 Ti

Reply 8 of 13, by SSTV2

Posted on 
User metadata
Rank Oldbie
Rank
Oldbie

Further update with a few more Mendocino CPUs and an AMD K5. For some reason K5 refuses to boot @ PR200 clock on this motherboard, works fine on older motherboards with a linear voltage regulator /shrugs.

Reply 9 of 13, by AlexZ

Posted on 
User metadata
Rank Oldbie
Rank
Oldbie

Thanks for adding more Mendocinos. Your results clearly confirm my findings from long ago that Mendocino 500 at 83Mhz FSB was an excellent CPU for gaming, almost a match for PIII at the same clock. Getting it work at 100Mhz FSB was very rare though. K6-2 was not worth keeping. Mendocinos are an excellent choice for old 440LX boards that do not support PIII.

Pentium III 900E,ECS P6BXT-A+,384MB,GeForce FX 5600, Voodoo 2,Yamaha SM718
Athlon 64 3400+,Gigabyte GA-K8NE,2GB,GeForce GTX 275,Audigy 2 ZS
Phenom II X4 955,Gigabyte GA-MA770-UD3,8GB,GeForce GTX 780
Vishera FX-8370,Asus 990FX,32GB,GeForce GTX 980 Ti

Reply 10 of 13, by rasz_pl

Posted on 
User metadata
Rank l33t
Rank
l33t
auron wrote on 2023-06-02, 21:27:

i think no one should be surprised that p6 outperforms p5 by a lot in unreal/UT, even the readme for those games mentions this. however if you go pre-1998 there also exist games that see no benefit or even run slower clock-for-clock on p6, making those much more expensive and power hungry klamaths look not so great.

P 233MMX was $594 in June of 97
In April 1998 $112 300 Covington and $149 300A Mendocino killed it completely. Less than a year later that fastest possible CPU is slower than the lowest end.
I dont think this happened before or ever since.

https://github.com/raszpl/FIC-486-GAC-2-Cache-Module for AT&T Globalyst
https://github.com/raszpl/386RC-16 memory board
https://github.com/raszpl/440BX Reference Design adapted to Kicad
https://github.com/raszpl/Zenith_ZBIOS MFM-300 Monitor

Reply 11 of 13, by gerry

Posted on 
User metadata
Rank l33t
Rank
l33t
rasz_pl wrote on 2023-06-06, 13:39:

P 233MMX was $594 in June of 97
In April 1998 $112 300 Covington and $149 300A Mendocino killed it completely. Less than a year later that fastest possible CPU is slower than the lowest end.
I dont think this happened before or ever since.

a good reminder of the rapid changes at that time, and how computers really are very cheap now!

Reply 12 of 13, by BitWrangler

Posted on 
User metadata
Rank l33t++
Rank
l33t++
rasz_pl wrote on 2023-06-06, 13:39:
P 233MMX was $594 in June of 97 In April 1998 $112 300 Covington and $149 300A Mendocino killed it completely. Less than a year […]
Show full quote
auron wrote on 2023-06-02, 21:27:

i think no one should be surprised that p6 outperforms p5 by a lot in unreal/UT, even the readme for those games mentions this. however if you go pre-1998 there also exist games that see no benefit or even run slower clock-for-clock on p6, making those much more expensive and power hungry klamaths look not so great.

P 233MMX was $594 in June of 97
In April 1998 $112 300 Covington and $149 300A Mendocino killed it completely. Less than a year later that fastest possible CPU is slower than the lowest end.
I dont think this happened before or ever since.

IDK if it was somewhat similar that Pentium P5 60/66 launched march 93, dx4 wasn't until march 94 and they were a bit faster, depending on code. then by 95 the low end was Pentium 75 which was faster than Dx4s by intel.

But then until socket H1/G1 for core i series, there wasn't really so many clean platform breaks, it was all overlap.

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 13 of 13, by Gmlb256

Posted on 
User metadata
Rank l33t
Rank
l33t
rasz_pl wrote on 2023-06-06, 13:39:
P 233MMX was $594 in June of 97 In April 1998 $112 300 Covington and $149 300A Mendocino killed it completely. Less than a year […]
Show full quote
auron wrote on 2023-06-02, 21:27:

i think no one should be surprised that p6 outperforms p5 by a lot in unreal/UT, even the readme for those games mentions this. however if you go pre-1998 there also exist games that see no benefit or even run slower clock-for-clock on p6, making those much more expensive and power hungry klamaths look not so great.

P 233MMX was $594 in June of 97
In April 1998 $112 300 Covington and $149 300A Mendocino killed it completely. Less than a year later that fastest possible CPU is slower than the lowest end.
I dont think this happened before or ever since.

Yep, Celeron was Intel's way to address the gap after leaving the Socket 7 platform. That Mendocino was legendary due to being very overclockable and having L2 cache running at full CPU speed unlike any PII or Katmai PIII.

Go to top of page Go to top of page
Back to General Old Hardware
Close window
Close window