PDA

View Full Version : 486 performance: cpu vs bus speed



carangil
September 30th, 2009, 05:07 PM
Ok here's a (probably dumb) question:

a 486 dx4 runs at 100 mhz, and has a bus speed of 25mhz.
a 486 dx2 runs at 66 mhz, and has a bus speed of 33mhz.

Here's the question: Which is faster/smoother for games? Does the faster memory bus of the 66 system make up for the slower processing?

Old Thrashbarg
September 30th, 2009, 05:53 PM
I can't offer much insight as to the performance difference between bus speeds on the 486 platform, other than to say that faster bus speeds feel more responsive overall. As far as actual benchmarks and such, though, I couldn't tell you... I never ran any benchmarks when I was using 486 machines.

However... You are working from an incorrect assumption. The 486 DX4/100 runs at a bus speed of 33mhz, not 25. Despite the name, DX4 chips used a 3X multiplier. So the comparison is really a non-issue; all other things will be equal, so the DX4 will naturally be faster due to the clock speed advantage.

krebizfan
September 30th, 2009, 06:22 PM
The Intel 486-DX4 had an internal multiplier of 3 so it had a 33 Mhz bus. AMD and Cyrix had chips with an internal multiplier of 4 but I don't have much experience with them.

Now, if the comparison was between DX2-66 or DX4-75, the DX2-66 was faster if the code could not fit inside the cache.

Basically, for big games give me the faster bus if the clock speed was close. Small games just tended to run too fast on either CPU.

lutiana
September 30th, 2009, 07:49 PM
Simply put the speed of the chip (internal clock) is how fast the processor can handle data. The bus speed on the other hand is how fast the data can get to the chip from the RAM.

The two things put together determine the overall performance, but in general you get more of a performance boost by upping the bus speed and not the core speed.

My guess is that a 486DX-2 66 on a 33Mhz bus probably performs very much like a 486DX4-100 on a 25Mhz bus. What I would do is over clock the DX4 but upping the bus to 33Mhz and dropping the multiplier to 3x. If this runs stable is should give you a nice boost. Just remember to keep the chip cool while over clocking.

Most of this is not very relevant on modern machines since they are far more complex and in the case of AMD there is no "front end bus" anymore since the memory management is done directly on the CPU die.

Ole Juul
September 30th, 2009, 08:32 PM
@lutiana: I'm having a little trouble understanding your concept of dropping the multiplier from 4X to 3X. The DX4 was designed to run at 3X, or sometimes 2X for the 100MHz version.

Here is review of the DX4 on Wikipedia. (http://en.wikipedia.org/wiki/Intel_80486DX4)

lutiana
September 30th, 2009, 08:51 PM
@lutiana: I'm having a little trouble understanding your concept of dropping the multiplier from 4X to 3X. The DX4 was designed to run at 3X, or sometimes 2X for the 100MHz version.

Here is review of the DX4 on Wikipedia. (http://en.wikipedia.org/wiki/Intel_80486DX4)

I read the OP and he said his DX4 was a 25Mhz bus with a 4x multiplier, I did not know that they were designed for 3x 66Mhz or 2x 50Mhz.

Basically upping the BUS and dropping the multiplier should give you more over all system performance and keep the core speed the same. This means no extra cooling on the CPU is needed (IIRC).

So on a 486DX4-100 running a 33Mhz bus and a 3x multiplier, I would most likely up the bus to 50Mhz and drop the multiplier to 2x. Or if it were a 50Mhz bus and a 2x multiplier model I'd most likely up the bus to 66Mhz and 1.5x multiplier.

Most likely neither of these 2 cases would be very stable, but you never know till you try. But the basic concept I am trying to get across here is that BUS speed tends to be more important than core speed for overall system performance.

Unknown_K
September 30th, 2009, 08:55 PM
News to me of the dx2/100 being 2x50. From what I recall Intel sold the original DX50 on 50Mhz bus and had issues with it so they switched to the DX/2 50 (25mhz bus) soon after.

Itel pretty much stuck to 25/33 Mhz bus speeds while later AMD loved 40Mhz (40,80,120 models) and some DX4/133's could run at 4x40 for 160Mhz.

As far as speed goes I think 40 and 50Mhz FSB models added another wait state to the memory cpu bus and if you had VLB with more then 1 VLB card that also slowed down the cpu-memory bus (VLB cards ran at the cpu FSB).

The fastest 486 CPU's I have used were the AMD DX/4 133 running at 160 on a PCI motherboard and also the Cyrix Pentium cored 120Mhz 486 chip (also PCI bus).

Ole Juul
September 30th, 2009, 10:17 PM
lutiana: I read the OP and he said his DX4 was a 25Mhz bus with a 4x multiplier, I did not know that they were designed for 3x 66Mhz or 2x 50Mhz.
I think that, as Old Thrashbarg pointed out, the OP had it wrong. :)

lutiana: But the basic concept I am trying to get across here is that BUS speed tends to be more important than core speed for overall system performance.
That makes sense to me.

Unknown_K: News to me of the dx2/100 being 2x50.
Was it? Not that Wikipedia is much of an authority, but here is the list of i486 models (http://en.wikipedia.org/wiki/80486).

IBMMuseum
September 30th, 2009, 10:23 PM
Here is review of the DX4 on Wikipedia. (http://en.wikipedia.org/wiki/Intel_80486DX4)

Well, a little information there is wrong/needs to be updated. Not all Intel 486DX4s have a 16Kb L1 cache (some have the original 8Kb), and not all are 3V CPUs (there are 5V Overdrive models). Iīve never heard of the 50MHz x 2 as a official Intel DX4 release, however not to say it doesnīt exist.

It appears that it has been over three years since my last updates, but I tested a number of Intel 486 CPUs for capabilities, and posted what I learned on a webpage. I really should strip a couple places that identify me, and provide the link. Maybe not all-encompassing, but it is the most complete list I am aware of.

EDIT: Here is the page: http://www.IBMMuseum.com/Intel486.htm

IBMMuseum
September 30th, 2009, 10:45 PM
...Not that Wikipedia is much of an authority, but here is the list of i486 models (http://en.wikipedia.org/wiki/80486).


...an on-chip floating-point unit (FPU), except in the SX and SL models...

False, the Intel 486SL has the FPU, and was a low-power version of the 486DX (later in the article the 486SL is listed correctly as this) in a PQFP design, with the added PC/ISA support chips. Iīve not heard of the Ļi486SL-NMĻ mentioned that didnīt have the FPU, but I suppose it was considered, especially with the heat/power the FPU could use where it detracted from the system design (like laptops). The thing I noted about the 486SL is that it was marked as a later 486 design, but didnīt have the ability to give the CPUID like other Intel 486s of the same timeframe.

Old Thrashbarg
October 1st, 2009, 04:30 AM
So on a 486DX4-100 running a 33Mhz bus and a 3x multiplier, I would most likely up the bus to 50Mhz and drop the multiplier to 2x. Or if it were a 50Mhz bus and a 2x multiplier model I'd most likely up the bus to 66Mhz and 1.5x multiplier.

486 chips had the multipliers set internally... as far as I know they cannot be changed, so you cannot just reduce the multiplier and up the bus speed. And anyhow, 50mhz is well-known to be problematic, 66mhz simply isn't doable. Even 40mhz can be a little iffy with some expansion cards, though it's usually workable.

The best OP is going to do is running the DX4 at 3x40mhz, for 120mhz total. I'd say most 100mhz chips will run at that setting. However, even that might not be as big a gain as you might expect, depending on what the motherboard does with wait-states at the higher bus speed.

Anonymous Coward
October 1st, 2009, 05:54 AM
*All* Intel DX4 CPUs had 16kb internal cache, but only a limited number had writeback cache. Certain steppings of the Intel DX4 were officially rated by Intel to do 2x50MHz. Intel DX4 officially supported *BOTH* 2X and 3X multipliers and could be changed by jumper (DX2 CPUs generally could not do this, unless they were really relabeled DX4s).
AMD was much more ambitious with its DX4 chips, and they were actually available for sale to the general public (unlike Intel's DX4, which was OEM only [excluding overdrive]). I believe AMD had at least *four* versions of DX4. 8kb and 16kb with or without writeback cache! I think some of their DX4 chips also supported the 2x50 mode.
Systems that use the 50MHz bus can be perfectly stable, and it's not hard to build one. The problem arises when you want to use VESA or PCI bus. Those two really shouldn't be clocked over 33MHz. However, on ISA or EISA system 50MHz is perfectly fine. I have found that often times a 40MHz bus will be faster than 50MHz bus because of the additional wait states required to make it stable.

IBMMuseum
October 1st, 2009, 08:00 AM
*All* Intel DX4 CPUs had 16kb internal cache, but only a limited number had writeback cache...

Certainly I am not adverse to retesting, but I did check for L1 cache size with what should be a reliable program (sizes were reported correctly as 8 or 16Kb across the Intel 486-class CPUs) and would have logged those DX4 ODPs if it had shown me 16Kb...


...Certain steppings of the Intel DX4 were officially rated by Intel to do 2x50MHz...

Do you know which ones? The only "stepping" value I saw from Intel was listed as "A" for the DX4s. By what I have written on the page:

CPUID 0480h: 'A' stepping
CPUID 0483h: 'A' stepping, 16Kb L1 Write-Back capable CPU, in Write-Through mode
CPUID 0490h: 'A' stepping, 16Kb L1 Write-Back capable CPU, in Write-Back mode
CPUID 1480: 'A' stepping, ODP DX4, 5V package

With the same stepping value it is hard to tell the difference in capabilities (such as different clock multipliers). In particular, marking an ODP as version '1.0' or '1.1' with no change in stepping confuses me. What did they actually change between the versions?


...Intel DX4 officially supported *BOTH* 2X and 3X multipliers and could be changed by jumper (DX2 CPUs generally could not do this, unless they were really relabeled DX4s)...

I knew this at least...

lutiana
October 1st, 2009, 08:38 AM
486 chips had the multipliers set internally... as far as I know they cannot be changed, so you cannot just reduce the multiplier and up the bus speed.


Ok, so they were multiplier locked, I really think I had one at some point that was not, or perhaps I am thinking of an early Pentium. I know that I was able to defeat this on some of the intel chips I have worked with but I forget which.



Systems that use the 50MHz bus can be perfectly stable, and it's not hard to build one. The problem arises when you want to use VESA or PCI bus. Those two really shouldn't be clocked over 33MHz. However, on ISA or EISA system 50MHz is perfectly fine. I have found that often times a 40MHz bus will be faster than 50MHz bus because of the additional wait states required to make it stable.

Yes, overclocking is about taking into consideration the entire system and not just the CPU, it is a very difficult balancing act of all the factors involved. But alot of times I was able to squeeze just a little more performance out of my systems by upping the bus speed a bit (usually just a single jumper needed to be moved).

Anonymous Coward
October 1st, 2009, 08:46 AM
I could be wrong, but I believe that there was something about the ODP DX4s that tricked the BIOS into thinking they only had 8kb internal cache to make them more compatible. I have heard that some boards won't boot up if the BIOS detects anything other than 8kb internal cache.

The information concerning operation at 50MHz should be available in the official intel white paper. I remember downloading it from intel's website more than 5 years ago, but it is long since gone. I can only find the datasheets for the embedded version at the moment, but I will keep looking.

Here it is:

http://web.archive.org/web/20000818071658/support.intel.com/support/processors/embedded/intel486/7824.htm

50MHz is officially supported on the following S-Specs:

SX876, SK053, SX877, SK050, SK099

It is likely that it works on most others as well, but it just hadn't been tested by intel.

Old Thrashbarg
October 1st, 2009, 09:11 AM
I was going by vague memories when talking about multipliers. Anonymous Coward says they can be changed, so I'll take his word on that. I remember that I couldn't get it to work, but it could have been due to some other factor. It's been about 10 years since I've touched such a system, so I just can't recall exactly.

In any case, I believe you still only have 2X and 3X available to work with... I've never heard of half multipliers except on the never-released DX3. Somebody mentioned a 1.5X multiplier in an earlier post... I don't think that can be done.

As for the 50mhz bus, I still really don't care much for the idea, regardless of how 'official' it is. Yeah, you can make it stable if you choose the right parts, but it's not likely to work very well on the average middle-of-the-road 486 board, if it even offers the setting at all. Also, I personally consider it a huge problem not to be able to use VLB or PCI. EISA would be a good alternative, sure, but good EISA boards are not exactly easy to find. Add in the factor of wait states, and you're much better off sticking with a 40mhz bus, the fastest cache you can use, and a good VLB video card. 40mhz still isn't ideal for VLB, and it could still get a little dodgy with SCSI cards and the like, but it usually works OK with video cards.

luckybob
October 1st, 2009, 09:24 AM
Ok here's a (probably dumb) question:

a 486 dx4 runs at 100 mhz, and has a bus speed of 25mhz.
a 486 dx2 runs at 66 mhz, and has a bus speed of 33mhz.

Here's the question: Which is faster/smoother for games? Does the faster memory bus of the 66 system make up for the slower processing?

My first thought was to ask why you cant run the DX4 on the 33fsb?

Adjust its voltage slightly higher and keep it cool, but why not have both worlds?

FishFinger
October 1st, 2009, 09:32 AM
Despite its name the DX4 (generally) has a 3x multipler, so its already running on a 33MHz bus - The 100mhz DX4 is 3x 33MHz, not 4x 25MHz.

IBMMuseum
October 1st, 2009, 09:45 AM
I could be wrong, but I believe that there was something about the ODP DX4s that tricked the BIOS into thinking they only had 8kb internal cache to make them more compatible. I have heard that some boards won't boot up if the BIOS detects anything other than 8kb internal cache.

My testing platform checked and saw the correct 16Kb sizes when it was there. I've documented a system not booting (that is, going up to where the POST checked for it) based on CPUID values. The BIOS has to be able (or a short utility program could do it too) to enable the L1 Write-Back mode on those CPUs that have that functionality, otherwise the CPU remains in the default Write-Through mode.

But what I will attempt to do is go through a whole series of retesting, just to satisfy myself...


...The information concerning operation at 50MHz should be available in the official intel white paper. I remember downloading it from intel's website more than 5 years ago, but it is long since gone. I can only find the datasheets for the embedded version at the moment, but I will keep looking.

Here it is:

http://web.archive.org/web/20000818071658/support.intel.com/support/processors/embedded/intel486/7824.htm

50MHz is officially supported on the following S-Specs:

SX876, SK053, SX877, SK050, SK099

It is likely that it works on most others as well, but it just hadn't been tested by intel.

I'll note and retest these as well...

Anonymous Coward
October 1st, 2009, 09:53 AM
As for the 50mhz bus, I still really don't care much for the idea, regardless of how 'official' it is.

I think the only reason intel even offered a 50MHz bus option was to make DX-50 users happy. Buyers of very expensive (in their day) 50MHz EISA systems were a bit upset because they did not have an upgrade path. Apparently this was intel's answer, even though it was not very convenient since you would still need to buy a separate VRM to make it work.

I certainly don't recommend a 50MHz ISA/EISA system over a good 33/40MHz VLB/PCI solution. But it's still interesting if you have limited components and want to try something different.

IBMMuseum
October 1st, 2009, 09:57 AM
...As for the 50mhz bus, I still really don't care much for the idea, regardless of how 'official' it is. Yeah, you can make it stable if you choose the right parts, but it's not likely to work very well on the average middle-of-the-road 486 board, if it even offers the setting at all. Also, I personally consider it a huge problem not to be able to use VLB or PCI. EISA would be a good alternative, sure, but good EISA boards are not exactly easy to find. Add in the factor of wait states, and you're much better off sticking with a 40mhz bus, the fastest cache you can use, and a good VLB video card. 40mhz still isn't ideal for VLB, and it could still get a little dodgy with SCSI cards and the like, but it usually works OK with video cards.

One bus (Microchannel) has been forgotten here, although there was only one real 486DX-50 microchannel platform: a Type 3 'M' complex on the PS/2 Models 90 and 95. The T3 (http://www.ibmmuseum.com/OhlandL/complexes/Type3.html) is a double-decker, which doesn't leave too much room for voltage interposers or heatsink/fans, but the systems should tolerate the bus speed just fine. I've heard of (probably AMD or Cyrix) CPUs being run on it at a 3x multiplier (150MHz).

With a 16Kb L1 cache, 256Kb L2 cache, and 150MHz internal speed (plus SCSI drive(s) instead of IDE, and a microchannel bus) you would think such system would really rock for a 486...

carangil
October 1st, 2009, 10:42 AM
Thanks everyone for your good responses. I stand corrected on the bus speed. I knew the intel dx4 was a 3x multiplier but for some reason i thought the am486 was 4x. I guess dx4 was a silly name.

I can tell all of you take old computers seriously... You jumped right on this thread. Im glad i've found a nice friendly group of computer users that isn't mostly focused on only the latest and greatest hardware. :)

IBMMuseum
October 1st, 2009, 10:58 AM
...50MHz is officially supported on the following S-Specs:

SX876, SK053, SX877, SK050, SK099

It is likely that it works on most others as well, but it just hadn't been tested by Intel.

It is interesting to see one S-spec (SX876) here stamped as a 75MHz part, although it probably wasn't a speed failure downgrade of an SX877 (most likely for marketing towards 25MHz base clock motherboard as an upgrade, but being a 3V part). All of the S-specs here I don't have identified as 16Kb L1 CPUs, which could also be a failure on my part (and will be better defined on a retest). Two of the S-specs (SK053 & SK099) I don't even have stamped for a speed (which could be deliberate by Intel, especially if the CPU was designed for 50MHz x 2).

IBMMuseum
October 1st, 2009, 11:04 AM
...although there was only one real 486DX-50 microchannel platform...

Excuse me, two: http://www.ibmmuseum.com/OhlandL/complexes/Type1.html#Upgrade_486DX-50

Old Thrashbarg
October 1st, 2009, 11:37 AM
I certainly don't recommend a 50MHz ISA/EISA system over a good 33/40MHz VLB/PCI solution. But it's still interesting if you have limited components and want to try something different.

Oh, sure, if you just want to experiment with it, it's definitely something interesting to play with. Nevertheless, it is kind of a hack, so I wouldn't want to mess with such things in a 486 I intended to actually use for something.

TNC
October 1st, 2009, 12:21 PM
I knew the intel dx4 was a 3x multiplier but for some reason i thought the am486 was 4x.

These ones had! :)

http://www.cpu-world.com/CPUs/80486/AMD-AMD-X5-133ADZ.html

carangil
October 1st, 2009, 01:07 PM
133... not that's cheating. I know 486 is not really vintage, but sorta considered vintage-esque, but once you get over 100 mhz, its just 'too fast.'

With that said... I now kinda want one :-)

Old Thrashbarg
October 1st, 2009, 03:37 PM
Yeah, the 5x86 was an interesting chip. It was a 486-class CPU, for 486 motherboards, but it was intended to compete with the lower-end Pentium chips, like the P75 and P90. And it did perform about on par with the Pentiums, especially on boards that could handle write-back cache.

And actually, now that I think about it, that might've been where I got the idea that 486 chips were multiplier locked... I think the 5x86, or at least some versions of them, were hard set to a 4X multiplier.

Unknown_K
October 1st, 2009, 08:12 PM
Some were locked but FSB could be changed easily enough 33 to 40 like I did to get 160.

Anonymous Coward
October 1st, 2009, 08:20 PM
Internal writeback cache was such a nightmare with the 486. Even on the boards that support it, there are always problems popping up. I know it wreaks havoc with SCSI VLB controllers.

IBMMuseum
October 1st, 2009, 08:47 PM
Internal writeback cache was such a nightmare with the 486. Even on the boards that support it, there are always problems popping up. I know it wreaks havoc with SCSI VLB controllers.

For the IBM units that ran the DX2 version (CPUID 0436h/0470h) I hadnīt heard of any issues. But there was problems sometimes with the WB cache mode of the POD CPUs on random IBM PS/2s. There it was often solved by bending a pin on the CPU.

Raven
October 11th, 2009, 08:13 AM
Sorry to not have read the entire thread, but does anyone know what sort of multiplier/bus is going on with the IBM DX4/100? That's the only DX4 in my collection. Strangely they don't seem to own one at the cpu-collection site.

Anyone have a lead on where I can get a 150mhz Am5x86 btw?

Anonymous Coward
October 11th, 2009, 12:23 PM
IBM DX4 is the same as Cyrix DX4. Cyrix was a fabless CPU designer, so all their manufacturing was outsourced through other companies. I think around 1994 Cyrix started using IBM to produce their chips, and they also allowed them to sell Cyrix chips under the IBM name. One benefit of using the IBM chips is they are ususally much more conservatively rated. I believe the Cyrix chip has 8kb writeback L1 cache (a source of headaches on older motherboards), support for 2x and 3x multiplier, and support for 50MHz bus.

You should probably give up on 5x86-150. They were never officially released by AMD. A few of them have leaked out, but they are considered somewhat rare (and collectible). As it turns out, because the 5x86 was made on a pretty advanced fabrication process for the time, it is quite easy to overclock the 133MHz part to 150 and 160MHz. I have a 133@160 and it is very stable, many people who owned this chip had similar good experience.