Hi, quake & doom benchmarks what screen size? what version? thanks

What’s the point of installing a DX4/586 processor without L2 cache?

dominusprog wrote on 2024-05-04, 10:56:

What’s the point of installing a DX4/586 processor without L2 cache?

I have a cyrix 5x86 without L2. After many test, i checked the L2 cache did not make almost any positive difference, in fact, oposite

theelf wrote on 2024-05-04, 16:49:

dominusprog wrote on 2024-05-04, 10:56:

What’s the point of installing a DX4/586 processor without L2 cache?

I have a cyrix 5x86 without L2. After many test, i checked the L2 cache did not make almost any positive difference, in fact, oposite

That’s odd..

Btw, I’ve pondered how a pod would do against a 120gp in a all out OC competition

Any chance you have a POD?

DId you have a fan on the cyrix? passive cooling from the stock heatsinks isn't enough really on 100mhz parts at 120.

Sphere478 wrote on 2024-05-04, 17:21:

That’s odd.. […]
Show full quote

theelf wrote on 2024-05-04, 16:49:

dominusprog wrote on 2024-05-04, 10:56:

What’s the point of installing a DX4/586 processor without L2 cache?

I have a cyrix 5x86 without L2. After many test, i checked the L2 cache did not make almost any positive difference, in fact, oposite

That’s odd..

Btw, I’ve pondered how a pod would do against a 120gp in a all out OC competition

Any chance you have a POD?

No sorry, i dont have a POD, and sadly i dont find good data to compare. I was checking phil database, but is not clear to me if quake results from people are fullscreen or not, then impossible to compare data

BitWrangler wrote on 2024-05-04, 17:49:

DId you have a fan on the cyrix? passive cooling from the stock heatsinks isn't enough really on 100mhz parts at 120.

Hi, i used the cx586 for years without any air cooling, just a heatsink. I addapted one from socket 370 and the cyrix is always in good temp. Im running at 4v

theelf wrote on 2024-05-04, 18:13:

No sorry, i dont have a POD, and sadly i dont find good data to compare. I was checking phil database, but is not clear to me if […]
Show full quote

Sphere478 wrote on 2024-05-04, 17:21:

That’s odd.. […]
Show full quote

theelf wrote on 2024-05-04, 16:49:

I have a cyrix 5x86 without L2. After many test, i checked the L2 cache did not make almost any positive difference, in fact, oposite

That’s odd..

Btw, I’ve pondered how a pod would do against a 120gp in a all out OC competition

Any chance you have a POD?

No sorry, i dont have a POD, and sadly i dont find good data to compare. I was checking phil database, but is not clear to me if quake results from people are fullscreen or not, then impossible to compare data

BitWrangler wrote on 2024-05-04, 17:49:

DId you have a fan on the cyrix? passive cooling from the stock heatsinks isn't enough really on 100mhz parts at 120.

Hi, i used the cx586 for years without any air cooling, just a heatsink. I addapted one from socket 370 and the cyrix is always in good temp. Im running at 4v

When comparing to pentium, it’s probably wise to disregard quake results as quake used some special Intel only tricks here

Cyrix got some undeserved criticism about their floating point unit because of quake it wasn’t that cyrix had a bad floating point unit it was that quake figured out a way to make the Intel one run very well using kind of an unintended trick. Which is nice, but is kind of a little bit unfair in the competition.

Sphere478 wrote on 2024-05-05, 01:05:

theelf wrote on 2024-05-04, 18:13:

No sorry, i dont have a POD, and sadly i dont find good data to compare. I was checking phil database, but is not clear to me if […]
Show full quote

Sphere478 wrote on 2024-05-04, 17:21:

That’s odd..

Btw, I’ve pondered how a pod would do against a 120gp in a all out OC competition

Any chance you have a POD?

No sorry, i dont have a POD, and sadly i dont find good data to compare. I was checking phil database, but is not clear to me if quake results from people are fullscreen or not, then impossible to compare data

BitWrangler wrote on 2024-05-04, 17:49:

DId you have a fan on the cyrix? passive cooling from the stock heatsinks isn't enough really on 100mhz parts at 120.

Hi, i used the cx586 for years without any air cooling, just a heatsink. I addapted one from socket 370 and the cyrix is always in good temp. Im running at 4v

When comparing to pentium, it’s probably wise to disregard quake results as quake used some special Intel only tricks here

Cyrix got some undeserved criticism about their floating point unit because of quake it wasn’t that cyrix had a bad floating point unit it was that quake figured out a way to make the Intel one run very well using kind of an unintended trick. Which is nice, but is kind of a little bit unfair in the competition.

Beside software like autocad, etc i cant think what else in dos use fpu and can use for benchmark and compare

Sphere478 wrote on 2024-05-05, 01:05:

When comparing to pentium, it’s probably wise to disregard quake results as quake used some special Intel only tricks here

More specifically special Pentium trick. Quake was not meant to be playable on 486, thus testing on 486 doesnt make much sense.

Sphere478 wrote on 2024-05-05, 01:05:

quake figured out a way to make the Intel one run very well using kind of an unintended trick.

Its funny to think about people calling Michael Abrash mister Quake ;]

rasz_pl wrote on 2024-05-05, 12:52:

Sphere478 wrote on 2024-05-05, 01:05:

When comparing to pentium, it’s probably wise to disregard quake results as quake used some special Intel only tricks here

More specifically special Pentium trick. Quake was not meant to be playable on 486, thus testing on 486 doesnt make much sense.

It make all sense if you want to compare high end 486 to pentium, because one of the idea of build a high end 486 is to test software from latest dos era, and first windows 9x era too

theelf wrote on 2024-05-04, 16:49:

dominusprog wrote on 2024-05-04, 10:56:

What’s the point of installing a DX4/586 processor without L2 cache?

I have a cyrix 5x86 without L2. After many test, i checked the L2 cache did not make almost any positive difference, in fact, oposite

That's interesting, because when I upgraded the cache on my DX2-66 from 128 to 256KiB it improved the performance. I can imagine that removing the entire cache can affect the performance.

dominusprog wrote on 2024-05-05, 18:51:

theelf wrote on 2024-05-04, 16:49:

dominusprog wrote on 2024-05-04, 10:56:

What’s the point of installing a DX4/586 processor without L2 cache?

I have a cyrix 5x86 without L2. After many test, i checked the L2 cache did not make almost any positive difference, in fact, oposite

That's interesting, because when I upgraded the cache on my DX2-66 from 128 to 256KiB it improved the performance. I can imagine that removing the entire cache can affect the performance.

Same here in various 486 i have, including a IBM SX25 upgraded to DX4100, that get a nice improvement from L2 cache both at 25 and 100mhz, 5 to 10% aprox, and a TX DLC40mhz that get a HUGH improvement with L2, more or less 10 to 20%

But with this board and cyrix CPU, i did not see (almost) any improvement, like 1% to none, and to keep stability i need to use 3-1-1-1 in fact using L2 make system slower

dominusprog wrote on 2024-05-04, 10:56:

What’s the point of installing a DX4/586 processor without L2 cache?

We're not talking about FSB speeds that allow a -1-1-1 burst.
Please read this here: 486 board with UMC 8881E/8886B: The winner is: EDO without L2 (if your only other option is L2 at 3-2-2-2)

Disruptor wrote on 2024-05-07, 19:19:

dominusprog wrote on 2024-05-04, 10:56:

What’s the point of installing a DX4/586 processor without L2 cache?

We're not talking about FSB speeds that allow a -1-1-1 burst.
Please read this here: 486 board with UMC 8881E/8886B: The winner is: EDO without L2 (if your only other option is L2 at 3-2-2-2)

Sure, thanks for the link.

I think that's a really quite contrived example. mkarcher clearly recognises that Quake, in particular, is not very cache friendly (the memory access being somewhat 'random'); one would suspect that a larger cache would alleviate this to a certain degree since the hit rate would surely increase.

The above may be true for certain single threaded workloads, for certain applications, but I suspect that as soon as you encounter software that has relatively standard memory access patterns (i.e large data arrays in contiguous memory locations, instructions without wild branching patterns) or you move to a mixed software load like in Windows (or Linux, or OS/2) then the assumption will go out of the window.

There has been a lot of research into the design and sizing of cache memory over many decades, and unless the implementation is completely and utterly broken (which is always possible) or an exploit results in the performance increase being negated by security measures (e.g. cache flushing on context change), in the majority of workloads it is better to have it in place than not.

megatron-uk wrote on 2024-05-08, 08:03:

There has been a lot of research into the design and sizing of cache memory over many decades, and unless the implementation is completely and utterly broken (which is always possible) or an exploit results in the performance increase being negated by security measures (e.g. cache flushing on context change), in the majority of workloads it is better to have it in place than not.

While this is clearly true, special circumstances (and FSB60 on a 486 mainboard clearly counts as "special circumstance") may invalidate otherwise perfectly well working cache strategies. Note that in those speed tests, I had to insert a memory read wait state at FSB60 as soon as I added the cache. This makes the cache miss performance plummet from 76MB/sequential read to 54MB/s sequential read. At the same time, writes also suffer, because the possibly required cache update (3 clocks) takes longer than the actual write to RAM on a page hit (just 2 clocks). The cache subsystem is clearly not designed to deliver adequate performance at FSB60, at least with the cache chips I used for that test. On the other hand, the EDO RAM that "won" the competition is able to handle FSB60 quite well.

A cache only makes sense if it is significantly faster than RAM, and the cache management overhead is negligible. Both is not the case with the 15ns cache chips and the required 3-clock leadoff, if I am able to run EDO RAM without extra waitstates at 4-2-2-2. Basically the thread shows that an unfit cache does not help if the system has high-performance RAM. That 15ns cache (including a 15ns tag) is unfit for FSB60.

megatron-uk wrote on 2024-05-08, 08:03:

Quake, in particular, is not very cache friendly

Its more of Quake exuberating exacerbating and exposing poorly implemented caches. Same situation on socket 5/7 in non PB configurations.

theelf wrote on 2024-05-04, 10:17:

Hi, quake & doom benchmarks what screen size? what version? thanks

Hi, I will check the version at this weekend

BitWrangler wrote on 2024-05-04, 17:49:

DId you have a fan on the cyrix? passive cooling from the stock heatsinks isn't enough really on 100mhz parts at 120.

A big heatsink with active USB powered fan.