My guess is any upgrade that changes the base FSB the most will have the most impact on overall scores (assuming cache issues aren’t present)

Numerically the overdrives that increase clock speed the most as a percentage will have the most affect on numerical calculations , though frames per second not necessarily.

The p24t will have the largest floating point improvement over the 486

A mixed bag if I ever saw one.

Should be interesting

I'm guessing the Socket 4 system will show the most improvement. Posting to follow. Interesting topic!

rmay635703 wrote on 2020-03-29, 15:33:

My guess is any upgrade that changes the base FSB the most will have the most impact on overall scores (assuming cache issues ar […]
Show full quote

My guess is any upgrade that changes the base FSB the most will have the most impact on overall scores (assuming cache issues aren’t present)

Numerically the overdrives that increase clock speed the most as a percentage will have the most affect on numerical calculations , though frames per second not necessarily.

The p24t will have the largest floating point improvement over the 486

A mixed bag if I ever saw one.

Should be interesting

I agree, that changing FSB will have a big impact. So we can't give good estimations 😁
So do you change the FSB or not?

Socket 3 - 486DX2-66 P24D to PODP5V83 (Pentium Overdrive 83 MHz) FSB stays at 33MHz
So no dramatical gain in integer performance (+35%?), but very big gain in floating point performance, maybe 3 times faster.
Assuming that L1 cache works in WB mode, the overdrive will gain even more, because it's cache is 4 times bigger! (16k + 16k vs. 8k) and in write back mode it will be more effective.

Socket 4 - Pentium-60 P5 to PODP5V133 (Pentium Overdrive 133 MHz) FSB changes from 60 to 66MHz

Socket 5 - Pentium-75 P54C to PODP3v150 (Pentium Overdrive 150 MHz) FSB changes from 50 to 60MHz
So maybe here is the most gain, because FSB +20%

Socket 7 - Pentium-120 P54CQS to PODMT66x200 (Pentium Overdrive 200 MHz with MMX) FSB changes from 60 to 66MHz
The MMX has double the L1 cache, so I would suspect 2nd most gain for that upgrade.

CoffeeOne wrote on 2020-03-29, 16:43:

Socket 3 - 486DX2-66 P24D to PODP5V83 (Pentium Overdrive 83 MHz) FSB stays at 33MHz
So no dramatical gain in integer performance (+35%?), but very big gain in floating point performance, maybe 3 times faster.

35%? Even at the same clock speed, the Pentium is at least two times faster in integer maths than a 486, and the POD is based on the later P54C and clocked higher. Integer speed will be about 2.5 to 3 times faster than a DX2-66. And floating point? The Pentium was a monster when it came to floating point maths. I have no idea how it compares to a DX2-66, but I'm guessing it's somewhere between 5 to 10 times faster. (talking about raw performance here - not fps gain in games)

I had an Am5x86 system at 160 MHz which I upgraded with the PODP5V83 and got a speed increase even there - despite halving the clock rate.

CoffeeOne wrote on 2020-03-29, 16:43:

I agree, that changing FSB will have a big impact. So we can't give good estimations 😁
So do you change the FSB or not?

Yes. I will change FSB as needed to satisfy nominal ratings of overdrives chips. Those FSB jumps are there to make things more interesting 😀

derSammler wrote on 2020-03-29, 16:53:

CoffeeOne wrote on 2020-03-29, 16:43:

Socket 3 - 486DX2-66 P24D to PODP5V83 (Pentium Overdrive 83 MHz) FSB stays at 33MHz
So no dramatical gain in integer performance (+35%?), but very big gain in floating point performance, maybe 3 times faster.

35%? Even at the same clock speed, the Pentium is at least two times faster in integer maths than a 486, and the POD is based on the later P54C and clocked higher. Integer speed will be about 2.5 to 3 times faster than a DX2-66. And floating point? The Pentium was a monster when it came to floating point maths. I have no idea how it compares to a DX2-66, but I'm guessing it's somewhere between 5 to 10 times faster. (talking about raw performance here - not fps gain in games)

I had an Am5x86 system at 160 MHz which I upgraded with the PODP5V83 and got a speed increase even there - despite halving the clock rate.

🤣 your are by far more optimistic than Intel's marketing department at that time. No chance that integer performance of the POD 83 is 2.5 to 3 times faster than a DX2/66.

In the 486 socket 3 motherboard I use the Intel 486dx4-100 overdrive CPU which gives me about a 10% CPU performance boost over the 486dx2-66.

The AMD 5x86-133 socket-3 CPU is about 20% faster than the 486dx4-100. And then over clock it to 160mhz is about another %20 faster.

But you should NOT compare different fruits to each-other.

Keep and 486 a 486
And keep a 5x86 a 5x86
And Keep a Pentium a Pentium.

Because these CPU’s are different architecture designed for different motherboards.
Different eras and different operating systems.

Personally, I think people are missing the message.
It’s NOT about being fastest or best quality components But best having the best best software and hardware compatibility.
What goo is the fastest computer if it will NOT run the software or games you want to run.

So for a 486 I think the 486dx2-66 has the best game compatibility for DOS and Win3x games.
And for Windows 95 the Pentium is bests.
In combination with the AWE64 sound card.

Intel486dx33 wrote on 2020-03-29, 17:58:

In the 486 socket 3 motherboard I use the Intel 486dx4-100 overdrive CPU which gives me about a 10% CPU performance boost over t […]
Show full quote

In the 486 socket 3 motherboard I use the Intel 486dx4-100 overdrive CPU which gives me about a 10% CPU performance boost over the 486dx2-66.

The AMD 5x86-133 socket-3 CPU is about 20% faster than the 486dx4-100. And then over clock it to 160mhz is about another %20 faster.

But you should NOT compare different fruits to each-other.

Keep and 486 a 486
And keep a 5x86 a 5x86
And Keep a Pentium a Pentium.

Because these CPU’s are different architecture designed for different motherboards.
Different eras and different operating systems.

Personally, I think people are missing the message.
It’s NOT about being fastest or best quality components But best having the best best software and hardware compatibility.
What goo is the fastest computer if it will NOT run the software or games you want to run.

So for a 486 I think the 486dx2-66 has the best game compatibility for DOS and Win3x games.
And for Windows 95 the Pentium is bests.
In combination with the AWE64 sound card.

What do you mean?
The pentium overdrive 83 has pentium architecture inside.
But it was made to work in a 486 mainboard AND it shall replace an existing 486 cpu. So blame Intel, if you find that bad.

Not really, pentium OD 83mhz, intel custom designed the die and left off many features that makes up real pentium so you cannot get full benefit of pentium.

https://www.os2museum.com/wp/intel-overdrive- … tium-overdrive/

Cheers,

pentiumspeed wrote on 2020-03-29, 19:09:

Not really, pentium OD 83mhz, intel custom designed the die and left off many features that makes up real pentium so you cannot get full benefit of pentium.

https://www.os2museum.com/wp/intel-overdrive- … tium-overdrive/

Yes, I know about this. In the article Michal is making a claim that PODP83 doesn't have a V-Pipe because it cannot be disabled by a test register.

However, I have reasons to believe this is not accurate::

1. Arguably if the PODP83 was really "crippled" like that, the performance difference compared to "real" Pentium would be bigger than it actually is. In fact in pure integer performance benchmark (Dhrystone) the PODP83 easily beats Pentium 75. Surely that wouldn't be the case if the PODP83 was unable to perform two instructions per clock in many cases.
2. When I run software on PODP83 using Pentium hardware counters profiler I can see that the counter showing instruction executed on the v-pipe is clearly increasing. Also the proportion of overall instructions and V-pipe instructions is in line with expectations (considering the V-Pipe cannot execute all instructions).
3. This would for sure must have been mentioned in manuals, erratas and data-sheets. Intel wouldn't be able to advertise the PODP83 like they did. I mean numerous references to superscalar technology) in manuals, etc.. Otherwise the world would sue the #%$^ out of them.

mpe wrote on 2020-03-29, 20:31:

However, I have reasons to believe this is not accurate:

Whether or not it's accurate, he's also completely ignoring the fact the the mainboard used has a huge impact on the speed gain you get. With a proper late 486 board, the POD can run to its full potential. But if you use it on some old, crappy board, which can hardly push 20mb/sec. through memory, the POD can't do wonders and may not be much faster than a DX4.

The explained reason why he thinks the second v-pipe is disabled is probably just off. He's rather not getting a slow-down when forcing single-pipe execution because with the 32-bit bus instead of a 64-bit bus, the CPU simply can not fetch instructions fast enough anyway for the second v-pipe to matter - unless you have all code and data in the L1 cache already.

Also, some of his negative statements are kind of self-fulfilling, like:

"although a PODP5V83 still doesn’t perform anywhere near a 90 MHz Pentium."

Why would one even think it does? It's not running at 90 MHz and sits on a 486 bus. It simply can't perform anywhere near a 90 MHz Pentium.

Wow, according to the box, the PODP5V83 could also replace an SX-33 (well of course, I just never thought of it). What a great boost that must be .. 😁

[Edit]: Apparently I did think about it before .. 🤣 http://www.cpu-world.com/forum/viewtopic.php?t=32265&start=0

H3nrik V! wrote on 2020-03-30, 09:31:

What a great boost that must be .. 😁

Yes, from jerky DOOM to (almost) smooth Quake. 😁

Yes. Despite only transferring 32bits at once and having lower FSB clock, mature 486 chipsets compare favourably to early Pentium chipsets (Mercury, Neptune, SiS 50x). Even considering memory/L2 bandwidth. That's why Am 5x86 runs so well against entry Pentiums. Especially when overclocked to 150/50 or 160/40 MHz. So when deploying the PODP83 in such a system it isn't necessarily disadvantaged that much.

Obviously things change when you compare against boards with Triton or later chipset with pipelined-burst L2, optimised memory buses with EDO ram.

I think Intel hit a very sweet spot when they compensated the "only"486 bus by doubling L1 caches in the Pentium Overdrive.

it’s NOT about which CPU is faster. You already know which CPU is faster by the MHz rating.
It’s about which CPU will allow for the play of the most DOS games in good quality.
Sure you want the fastest CPU but you also want to be able to down clock it for CPU speed critical games.
Down clocking by using the Turbo button, Disabling cache and using the “Setmul” ( Set Multiplier) utility.
Using these methods you should be able to down clock the CPU to a 286@10 MHz at least.

This will allow for the most DOS games to be played on that CPU.

Intel486dx33 wrote on 2020-03-30, 16:38:

it’s NOT about which CPU is faster. You already know which CPU is faster by the MHz rating. .. the play of the most DOS games in […]
Show full quote

it’s NOT about which CPU is faster. You already know which CPU is faster by the MHz rating.
.. the play of the most DOS games in good quality.
... you also want to be able to down clock it for CPU speed critical games.
This will allow for the most DOS games..

I see your point, but I personally have no desire for playing any sort of games on my retro systems.

Intel486dx33 wrote on 2020-03-30, 16:38:

it’s NOT about which CPU is faster. You already know which CPU is faster by the MHz rating. It’s about which CPU will allow for […]
Show full quote

it’s NOT about which CPU is faster. You already know which CPU is faster by the MHz rating.
It’s about which CPU will allow for the play of the most DOS games in good quality.
Sure you want the fastest CPU but you also want to be able to down clock it for CPU speed critical games.
Down clocking by using the Turbo button, Disabling cache and using the “Setmul” ( Set Multiplier) utility.
Using these methods you should be able to down clock the CPU to a 286@10 MHz at least.

This will allow for the most DOS games to be played on that CPU.

For you. Each person has their own interests and goals related to this hobby.

I look forward to the benchmark results, mpe!

And the winner is….

Pentium Overdrive 83

Despite only offering only a modest clock increase the all-new architecture makes all the difference when compared to the 486DX2-66. This unloved CPU really brings the most significant improvement. I expected it to rock the Quake test. Quake loves Pentium FPU and where Quake is compared no 486 can compete. 2.5x improvement is just massive. However, I was surprised it also won general-purpose Winstone 95 test - 1.6x improvement which is ahead of other overdrives.

The other overdrives were remarkably close to each other.
I'd put Socket 4 / Socket 5 on shared 2-3 spot by having in average 1.5x improvement
The Socket 7 is lagging slightly with 1.4x improvement

It is not without surprise than in absolute terms P120 on 430HX chipsets beats both P133 and P150 on 430LX/NX. It just stress out how important choice of chipset was in the Pentium era.

mpe wrote on 2020-03-30, 23:31:

It is not without surprise than in absolute terms P120 on 430HX chipsets beats both P133 and P150 on 430LX/NX. It just stress out how important choice of chipset was in the Pentium era.

this is the common opinion nowadays, but i'd say you are showing the best case scenario here by comparing the intel LX/NX boards which are well known for not having any performance tuning settings with a late aftermarket HX board with extensive tuning options (which i am assuming is what you did here, couldn't find a mention of it in this thread). so while the HX has the extra PCI buffers that work transparently and obviously gets a solid boost from pipeline burst cache, EDO is a tad overrated and tends to not give a huge benefit unless enabling all the advanced memory tuning features on the HX, which are off by default on the GA-586HX. i also have to wonder what the TC430HX would do here, as that is another intel board with very limited BIOS settings; i wouldn't be surprised if it kept everything at rather slow settings for compatibility sake.

so my point is, while this is a good comparison to show the maximum potential of a later chipset, default settings would show less of a difference and would be more representative of the average experience back then as i'm certain most users didn't tweak the BIOS settings. i also doubt the 430LX/NX were regarded as all that slow in 93/94 as they were simply standard for pentiums then, and there were plenty of cacheless 430FX boards with slow FPM sold early on that have been all but forgotten nowadays in favor of EDO/PB setups that i don't think really became truly dominating in the market until late 1995. the early Advanced/EV boards could even come with async COAST modules which further proves that the transition to PB was not all that instantaneous.