Perhaps because the 486 has more cache on the motherboard and the P3 doesn't?
edit:Talking about L2 cache, the P3 has it on the cpu while Socket 7 and before have it on the motherboard 😉

Yes and you can disalbe L1 as well as L2 in the BIOS of most PIII boards. L2 cache has basically zero impact. And L1 cripples the cpu...

Perhaps L1 cache is more important for newer processors as they have a higher CPU/FSB ratio

Yea it seems so!

Well I didn't have much luck getting my Slot 1 > 370 adapter to work in my new BX440 board. That board seems a bit flaky anyway, got a few weird messages and required a reboot (that was without an adapter, just a normal slot 1 cpu).

With the adapter it would lock up at post...

I have played more with my S370 board and tightening the RAM timings helped a little bit. I get a Norton SI of 20 and I tried all my favourite games (WC1, WC2, Monkey Island 1 and 2, Fate of Atlantis, Hearth of China and so on) and they all worked perfectly! FreeDos seems very compatible and it has plenty of memory plus CD drive and mouse support out of the box.

I keep reading that Wing Commander runs better on something faster, but maybe it's because these S370 boards have really fast video cards for DOS? I use the onboard video card that comes with the intel 815 chipset.

I have ordered a 1.4 GHz S370 cpu and hoping that this will give me 386DX 25 MHz performance which I would be happy with. Then I can close that part of the project and settle for recommending S370 boards with a ISA slot and a 1+ GHz cpu...

Next issue will be dealing with a single ISA slot...

I might post a new thread but maybe you know. Is there a Soundblaster clone card that has a real MPU401 (with intelligent mode) port.

So I can have digital speech and MPU401 for MT-32 and General midi (I will use external modules) with a single card...

CMIIW, but if you disable internal caches on later CPUs, meaning no cache whatsoever, all the instructions must be passed back and forth the RAM (which is why you will find the RAM's timing/frequency will quite affect the performance) ... and that is simply "not the way it's meant to be".

When i disabled the L1 cache, i expected the L2 cache to take the L1 cache's role though maybe half as fast, but it doesn't behave like that. It really depends on the design, i guess.

Yea I guess if P3 boards had DDR memory, we wouldn't be having this discussion 😁

My netbook with cached enabled is the perfect 386 machine. But no sound at all under DOS...

Playing with that AWE64 now. Lot's of noise I am trying to find the mixer to increase the volume...

Installation was easy, some plug and play tool which did everything for me.

I have read so much about AWE cards not having a genuine OPL chip and call me ignorant but Wing Commander and Monkey Island sound just like I remember them... 😁

Mau1wurf1977 wrote:

I came across this excellent post: The 486, The Turbo Button And The Internal Cache - Rudimentary Speed Analysis […]
Show full quote

I came across this excellent post: The 486, The Turbo Button And The Internal Cache - Rudimentary Speed Analysis

Now I wonder why the 486DX2-66 only gets a slight performance drop from disabling L2 cache, yet when I do the same on any Socket S370 cpu, the performance nose dives and it doesn't even reach the speed of a 486 cpu with cache disabled?

How is this possible? The clock speed of the PIII is much higher, the memory runs faster at tighter timings. The PIII "should" be faster, but it's not...

Do the later cpus simply have so much L1 cache that they are basically useless without it?

Look at what happened to the performance of the PII 266 when they removed the cache and created the Celeron. The performance hit was so severe that Intel had to add back some of the cache in later Celerons because nobody was buying them without it. Cacheless Celerons=major fail.

But that was the L2 cache wasn't it?

Under DOS it makes no difference. I actually have a Celeron 300 and a Celeron 300A. The 300A has Cache running at full speed. They benchmark the same under DOS...

So maybe the difference shows up in Windows (being a lot more memory intense...)

But still I don't understand how cacheless 486 can beat a cacheless Pentium III...

Doesn't quite make sense to me...

Mau1wurf1977 wrote:

But that was the L2 cache wasn't it? […]
Show full quote

But that was the L2 cache wasn't it?

Under DOS it makes no difference. I actually have a Celeron 300 and a Celeron 300A. The 300A has Cache running at full speed. They benchmark the same under DOS...

So maybe the difference shows up in Windows (being a lot more memory intense...)

But still I don't understand how cacheless 486 can beat a cacheless Pentium III...

Doesn't quite make sense to me...

Maybe you forgot to disable the L2 cache on the 486 motherboard?

sliderider wrote:

Mau1wurf1977 wrote:

Look at what happened to the performance of the PII 266 when they removed the cache and created the Celeron. The performance hit was so severe that Intel had to add back some of the cache in later Celerons because nobody was buying them without it. Cacheless Celerons=major fail.

i got one -- the Sloweron 266 with no L2 cache 😢 -- back in early 1998. 3d games got a lot better when the Banshee came though, but what can i say, the previous card is a SiS 6236 (yea, poor combination i know)

Mau1wurf1977 wrote:

Next issue will be dealing with a single ISA slot...

I might post a new thread but maybe you know. Is there a Soundblaster clone card that has a real MPU401 (with intelligent mode) port.

AFAIK there is none. Full Roland MPU-401 compatible port is only available with Roland soundcards (no SB), and the rest would only have UART mode MIDI port.

You could use a PCI Aureal Vortex2 card for MIDI (UART of course) in DOS though. There are also other PCI soundcards with pretty good DOS support.

Maybe you forgot to disable the L2 cache on the 486 motherboard?

L2 cache isn't mentioned in this test: The 486, The Turbo Button And The Internal Cache - Rudimentary Speed Analysis

But how come the L2 cache on the P III does nothing? e.g. you disable L1 cache and the machine is slow as s dog. So why does the L2 on or off have no impact?

and the rest would only have UART mode MIDI port.

What is the experience with using a SB16s midi port and a MT-32? Would most games work just fine?

You could use a PCI Aureal Vortex2 card for MIDI (UART of course) in DOS though. There are also other PCI soundcards with pretty good DOS support.

I will look into that! But if I have an ISA slot I might as well use a SB16 and that midi port...

Mau1wurf1977 wrote:

Maybe you forgot to disable the L2 cache on the 486 motherboard?

L2 cache isn't mentioned in this test: The 486, The Turbo Button And The Internal Cache - Rudimentary Speed Analysis

It is mentioned as 'external Cache'.

But how come the L2 cache on the P III does nothing? e.g. you disable L1 cache and the machine is slow as s dog. So why does the L2 on or off have no impact?

Maybe use a benchmark that doesn't fit into L1?

And btw disabling L2 on a 386 already drops the speed to ~40% in NU SI 7.0. (64 kB, cacheable size 16 MB)

Yea I guess if P3 boards had DDR memory, we wouldn't be having this discussion

A little off the intended topic, but that brings up an interesting point: there were PIII platforms with faster memory, like the i840 with dual-ported RDRAM, and the Via Apollo 266, with DDR. They weren't all that much better than PC133 in normal operation, but I wonder what sort of results they would give in cacheless operation...

My guess is it would perform relatively better. A guess, but I think the RDRAM would perform better then DDR as DDR had hardly any impact atall. Correct me if I'm wrong though, I'm not exactly an RDRAM scientist 😁

Well, RDRAM didn't have much impact either, since the PIII wasn't particularly starved for memory bandwidth. IIRC, most benchmarks generally showed the i840 to be within 5% or so of the i815... in single-processor configuration, anyway.

But that was with cache. Without cache, and simply going by bandwidth, I'd figure RDRAM>DDR>PC133, with probably a bigger gap between RDRAM and DDR than between DDR and PC133. But OTOH, there also may be the factor of latency to consider, and I don't know how, or if, it would noticeably affect things.

It'd be an interesting thing to test, at least, see if maybe such things would open up better slowdown possibilities. It seems like there's the potential there to fill in that 486-era speed gap. I'd like to test it out myself, but there's the minor little problem that I have no hardware of that sort.

Interesting idea!

Well I guess once you disable L1 cache, your cpu will become memory bandwidth starved 😁

Memory latency might become much more important as the cache has the property that it also compensates for higher latency.