Here's my system:

I'm awaiting for another "Pentium" sticker to arrive. The 1st edition type.




The Socket 5 with P133 processor:







As you can see, the PSU is fixed upside down.




Part of the chipsets:





3 PCI and 4 ISA Slots:





The header pins showing both the turbo switch and turbo led, among others :

I have special "attachment" to this board. Unlike many other items that I have collected from ebay, this system is actually one of the first I got - 15 years ago. My dad bought it for me when I was in pre-university college.

Bought it brand new, I was the first owner (so to speak) of this board and still am! 😁

All the discussions about the sound recordings in the other thread makes me want to insert my SB 1.5 with the C/MS chips (thanks for the chips, pianoman72!!!) on it, in this system. (Also when I got this system, it came with this original Sound Blaster 1.5 brand new in it's box. Still have it - box, disks and all.)

I have been inspired by all of your Pentium slow down projects. I have just tested a socket 7 mobo; MSI MS5158. It's just a regular socket 7, with no AGP. In the BIOS, you can disable L1 and also L2 cache. Is this normal for most socket 7 boards?

I tested this board with a S3 PCI video card. I used a Pentium 233 MMX CPU, and also a Pentium 90 MHz CPU. I used 3dbench. The full range of cache disabled scores were: 10.4, 12.0, 13.3, 16.3, 17.8, 19.6, 27.0, 28.5, 35.7, 38.4. Full details below. (To obtain higher values, I could increase the fsb from 66 to 75 or 83.)

P233, 66mhz fsb, multi 3.5
L1 disabled = 38.4, L1+L2 disabled = 17.8

66mhz fsb, multi 2
L1 disabled = 35.7, L1+L2 disabled 16.3

50mhz fsb, multi 3.5
L1 disabled = 28.5, L1+L2 disabled = 13.3

50mhz fsb, multi 2
L1 disabled = 27.0, L1+L2 disabled = 12.0

P90, 50mhz fsb, multi 1.5
L1 disabled = 19.6, L1+L2 disabled = 10.4

Cool my SS7 board can't do 50 MHz, it will only go as low as 66...

The mmx scores are very high! My Pentium is non MMX and I'm surprised you get such a high score with both Caches turned off.

I thought that the MMX are the same as a Pentium but with MMX extensions, but looks like Intel has tweaked a bit more...

I did a few more tests. I increased the FSB to 75 and also 83.

P233, 75mhz fsb, multi 3.5
L1 disabled = 41.6, L1+L2 disabled = 20.0

83mhz (slow BIOS timings), multi 3.5
L1 disabled = 47.6, L1+L2 disabled = 17.5

83mhz (faster BIOS timings), multi 3.5
L1 disabled = 47.6, L1+L2 disabled = 21.7

The full range of cache disabled scores is now: 10.4, 12.0, 13.3, 16.3, 17.5, 17.8, 19.6, 20.0, 21.7, 27.0, 28.5, 35.7, 38.4, 41.6, 47.6

Mau1wurf1977 wrote:

The mmx scores are very high! My Pentium is non MMX and I'm surprised you get such a high score with both Caches turned off.

I thought that the MMX are the same as a Pentium but with MMX extensions, but looks like Intel has tweaked a bit more...

I am unable to answer this, because of ignorance! 😉 I just tested the chips, and wrote down the scores. 😀

Very nice!

Comparing these scores with our database you can go as low as a 386DX25 and as high as 486DX2 66

Though the MMX will give you a 386DX-40 as it's lowest setting...

Very interesting. The MMX has a similar results as my Cyrix. Intel must have tweaked the design a little bit...

At 75 MHz FSB my Cyrix scores 41.6 just like yours (with L2 on) and 16.5 with L2 off...

PS: I think we can conclude that an Pentium PC makes a great "slowdown" machine. Lot's of options (L1, L2 Cache and often Turbo), ISA slots, PS/2 mouse and not as "fragile" as 386 hardware because of younger age...

Mau1wurf1977 wrote:

PS: I think we can conclude that an Pentium PC makes a great "slowdown" machine. Lot's of options (L1, L2 Cache and often Turbo), ISA slots, PS/2 mouse and not as "fragile" as 386 hardware because of younger age...

Definitely. 😀

This benchmark was done using the P90 chip, and the FSB @ 50 MHz. L1 and L2 BIOS caches were disabled. I used PCTools SI.exe - the same tool that Malik used earlier on in this thread.

I removed the P90, and dropped in the P233 again. I set the BIOS timings to max, enabled both L1 and L2 caches, and also increased the mobo's FSB from 50 to 83.

I have found another way of tweaking this board. It can accept either 72 pin DIMMs, or SDRAM. Up to now, I was using PC-133 SDRAM. Now, I have removed this type of memory, and installed some 72 pin FPM DIMMs. I also set the BIOS timings to "slow". Using a P90 - 50mhz fsb - multi 1.5, I get these 3DBench comparison scores:

SDRAM
L1+L2 disabled = 10.4

DIMMs
L1+L2 disabled = 8.1

retro games 100 wrote:

I have been inspired by all of your Pentium slow down projects. I have just tested a socket 7 mobo; MSI MS5158. It's just a regular socket 7, with no AGP. In the BIOS, you can disable L1 and also L2 cache.

That is some nice motherboard you have -- this document states that it can go up to 400 MHz with a K6-2 (does it really have 6.0x multiplier?) -- which is the highest I've seen for a 430TX based board. If that's true, then it has quite a good range all the way down to 50MHz x 1.5 as you demonstrated above.

archsan wrote:

retro games 100 wrote:

I have been inspired by all of your Pentium slow down projects. I have just tested a socket 7 mobo; MSI MS5158. It's just a regular socket 7, with no AGP. In the BIOS, you can disable L1 and also L2 cache.

That is some nice motherboard you have -- this document states that it can go up to 400 MHz with a K6-2 (does it really have 6.0x multiplier?) -- which is the highest I've seen for a 430TX based board. If that's true, then it has quite a good range all the way down to 50MHz x 1.5 as you demonstrated above.

(I know this topic is like 2 years old now)
There are several TX boards that will work with the K6-2.
The K6-2 interprets the 2x jumper as a 6x jumper, enabling 66Mhz x 6.
Many TX boards will actually work with the K6-2 (though with limitations), there even was a website about it, no idea if it still exists.

Ah, I see, thanks, I forgot that bit. Found that here. Also, looks like the K6-2/400 won't be able to use 1.5x multiplier. And the K6/233 will use 1.5x as 3.5x, so the one that seems fit for underclocking to 1.5x50MHz would be the K6/200 and K6/166. Now I'm not even sure they can do that since I haven't actually seen it myself. Have someone actually done this? (Ugh, I need to get a decent socket7 mobo ASAP!)

I see the K6-2 is performing relatively poorly compared to other slower Pentium cpu's when disabled L1 or L2.

Btw i see that most pc crippled down by disabling cache are just a tiny bit faster than 386 SX's and DX.
Just imagine what power you would have back then when you would have a 386 with 256K L2 cache. 😀

I think this comparison is nonsense since the cache architecture of a K6 strongly differs from a 386. A 386 system usually has just SRAM based Read Cache that is on the mainboard. The 386 has no L1 cache internally, so basically the Cache on the board is the first level. The speed gain going from the usual 64 kB to 256 kB is rather small. Adding a second Cache level wouldn't speed it up, it would probably even slow it down due to additional latencies.
It should also be considered that the ratio between execution cycles vs. cycles required for memory reads/writes is much higher on a CISC 386 than on a pipelined K6. At the same time the 386 clock speed is closer to memory clock, so overall latency (in cycles) should be lower on a 386 already.

So does the Turbo button actually alter the clock speed on a pentium? On my 486, all it does is disable the caches. Still runs at 66MHz.

Elianda,

Of course i was joking, i thought everyone knew this.

Nonetheless, it is pretty amazing that any pentium running stock clock and disabled L1 L2 cache is bringing it so low to 386 regions.

Most 386s i have seen even havent got any cache mem, and they run a 16 or 32bit memory bus, the pentium with sd or fp / edo runs 64bit so there is quite some advantage compared to the 386 let alone the core clock doesnt even do much on the pentium side.

In simpler terms, I think, the huge bottleneck is one reason why these more modern but cacheless CPUs would crawl down to (cached) 386 speed. By disabling all caches, practically the system RAM becomes its 'cache' and therefore, all cycles where it has nothing to do but wait are wasted/stalled. That's why the bottlenecking factor, i.e. the system RAM/bus speed (latency, clock) plays much more important role in determining overall system speed.

That at least would partially explain it -- the other reason must lay in how the CPU is designed, which I guess would depend a lot on the existence of cache memory for it to perform properly. Because in the example above by RG100, the cacheless Pentium is shown to be performing near a 286 system, which AFAIK is also cacheless and has even lower CPU/bus clock speed (and lower RAM latency as well, I'd guess). This makes me think there's more to it than just latency differences. A more educated explanation is welcome, of course.

Ah, of course, that comment above shows my ignorance of microprocessor architecture in general. Of course, pipelined CPUs such as K6, Pentium and all modern x86 CPUs would "depend a lot" on the cache, duh... 😀

I just stumbled onto this article which gives a quick introduction to modern microprocessor architectures:
http://www.lighterra.com/papers/modernmicroprocessors/

Random 'study notes':

* 286 is not a pipelined design (which means that instructions are executed sequentially), but 386 is (depth: 6 stages -- as a comparison, Pentium is 5 stages deep). My hunch: could it be that by disabling caches on the Pentium, not only we're screwing it up with memory latency, but also with its branch predictor inefficiency? (From what I read somewhere else: Pentium has quirks in its branch prediction, but P-MMX/P-Pro/II/III has a much improved branch predictor)

* In other discussions here, we have also seen how a cacheless P2 would crawl to a 286-speed. The P2, however, has a deeper pipeline than Pentium at 12+ stages, so disabling its caches would give a much greater hit. This leads me to...

* Random idea: Core / Bulldozer with ALL caches disabled ... ??? hehe (if that's even possible, I don't remember seeing such options on my mobo's BIOS menu).

who needs pentiums.
the 286 386s are just awesome. when you want to use real slow speeds there is nothing but the real deal.
any emulator also has no charm to me.
luckily i have a real 286-25 😀