VOGONS


The Ultimate 486 Benchmark Comparison

Topic actions

Reply 60 of 280, by Tetrium

User metadata
Rank l33t++
Rank
l33t++

The MMX chip will have a minimum multiplier of 2x as this multi is remapped to 3.5x. I'm not sure but there may be some early MMX chips that hadn't been remapped yet (The MMX 233 chip was released somewhat later then the 166 and 200 ones).

Also, it may prove a problem to run the Tillamook with the L2 cache enabled. So far finding a suitable board for this chip has proven quite fruitless 🙁

Whats missing in your collections?
My retro rigs (old topic)
Interesting Vogons threads (links to Vogonswiki)
Report spammers here!

Reply 61 of 280, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Are both the low-end and high-end of the Tillamook locked to 3.5x? I haven't experimented any with locked CPUs. I can run at 3.5x75 and 3.5*83 but if this is the case, a 233MMX would be suitable enough for the OC work.

I hope the 233MMX isn't locked on the low end, or else I won't have 166MMX results.

@sebaz_ri
When doing CPU comparisons, it is best that they all be done on the exact same system. So, I will likely carry out this work over the course of several months.

Plan your life wisely, you'll be dead before you know it.

Reply 62 of 280, by Tetrium

User metadata
Rank l33t++
Rank
l33t++

I know many Intel CPU's of the P1 and P2 kind were often unlocked and sometimes partially locked.
Some Pentium and many Pentium MMX chips are locked upward, so that shouldn't be of any concern when it comes to underclocking.

I'd say use a single motherboard if you can, I think a Super 7 board would be best and make sure all the memory is cached at all times.

Whats missing in your collections?
My retro rigs (old topic)
Interesting Vogons threads (links to Vogonswiki)
Report spammers here!

Reply 63 of 280, by elianda

User metadata
Rank l33t
Rank
l33t

About locked Multipliers, how about this information given here?
http://www.netreach.net/~wieland7/Hardware/In … ultipliers.html

Retronn.de - Vintage Hardware Gallery, Drivers, Guides, Videos. Now with file search
Youtube Channel
FTP Server - Driver Archive and more
DVI2PCIe alignment and 2D image quality measurement tool

Reply 64 of 280, by feipoa

User metadata
Rank l33t++
Rank
l33t++

@elianda
That was a fun read, particurlarly the first paragraph. This is what I suspect Cyrix did with the 5x86/4X CPUs -- Test them, then stamp them with what frequency ratings passed specs. Thank you for the information.

@tetrium
I'm going to run the 586 test on a very versitile socket 7 motherboard, which from what I can tell, is basically a super7 without 100 MHz FSB and AGP support. Albiet, some highly overclocked CPUs can't be tested, but that is fine with me.

This AZZA PT-5IT v2.1 has support for:

FSB (MHz): 50, 60, 66, 75, 83

Multiplier (x): 1.5, 2.0, 2.5, 3.0, 3.5 (via 1.5), 4.0, 4.5, 5.0, 5.5, and 6.0 (via 2.0)

Voltages (V): 3.3, 3.5, 3.2, 2.9, 2.8, 2.5, 2.1

Extra support for: Cyrix MII/4X and AMD II/III+

Supports 256 MB of DIMMs (2x128MB), 512KB pipeline burst, i430TX

Agreed, all RAM should be kept cacheable here so as not to distract from the extended cacheable range of the AMD K6-III+. For this i430TX, I need 64 MB. I have 64 MB SIMMs, a single 32 MB DIMM, and some 128/256/512/1024 MB DIMMs, but would rather plop in a fast 64 MB DIMM. I might need to do some shopping. EDIT: $3 spent, 64 MB DIMMs achieved.

EDIT2: Crap, that means for a valid Cyrix 5x86 vs Cyrix 6x86 comparison, I need to redo the 5x86 and POD with all cached RAM. heh, what's another 5 CPUs anyway? It only really affected 3 or 4 of the Windows-based scores.

Plan your life wisely, you'll be dead before you know it.

Reply 65 of 280, by Tetrium

User metadata
Rank l33t++
Rank
l33t++
elianda wrote:

About locked Multipliers, how about this information given here?
http://www.netreach.net/~wieland7/Hardware/In … ultipliers.html

Hehe, it was exactly -that- website that sparked a testing marathon, me testing all of my 166 and 200 MMX chips to see what multipliers worked and which ones wouldn't! 😁

And as clever as I was, I never bothered to write down the results on the CPU's themselves, so no idea which one is which! 😜

That's irrelevant to me now though, I got a whole bunch of the 233MMX chips now, but still was a funny afternoon 😀

@feipoa: Just for the record, Super 7 is by definition a board with official 100Mhz FSB 😉
A board can support even 133Mhz FSB and not have AGP, but still be a Super 7 board 😉

I've found that quite a lot of the TX boards (which Intel of course intended to be maxed with their 233MMX chip) support CPU voltages downwards to 2.2v often (maybe some even lower) and could in theory work with virtually any Socket 7 chip made 😉

I really appreciate all of your work in this area, I know it sucks up a LOT of hours, even if everything goes right!

I have 1 suggestion though, 430TX doesn't support a feature that made the Cyrix chips faster, Linear Burst.
If possible, you might retest the Cyrix chips only and only to test the difference in Linear Burst mode and non-Linear Burst mode 😉

Edit:Did a bit of googling, and found this thread AMD K6-266 vs. Cyrix MII-400GP @ 263 Mhz
I uploaded a small text file with possible undocumented jumper settings for some extra voltages.
Have you by any chance tried a couple of them out?
HINT:What I do when testing undocumented voltage settings is underclock the CPU by as much as possible (so you have the greatest chances of the system booting successfully and the system not killing the chip 😉. I realize it's not 100% safe, but risks are not that great actually) and then write down what CPU voltage I see in the BIOS (that is...IF it's in the BIOS 😕 ).

Whats missing in your collections?
My retro rigs (old topic)
Interesting Vogons threads (links to Vogonswiki)
Report spammers here!

Reply 66 of 280, by feipoa

User metadata
Rank l33t++
Rank
l33t++

I always thought the definition of a super 7 was it being a late-era socket 7 which was able to run faster MII's and K6II/III+ chips, and possibly had DIMMs, and sometimes AGP. I was unaware that "100 MHz" defined "super", but I'm glad to know now.

You have raised a very good point about LINBRST mode. Some time ago I did a Cyrix 6x86 program search and grabbed about 5 apps to alter Cyrix 6x86 register bits. Unfortunately I got side tracked and never tested them out with this TX board. It would be a pitty to loose 5-10% performance on a Cyrix without LINBRST.

I do not have a VIA or ALi or other chipset based socket 7 to test these chips on. I suppose, I could test a Cyrix 5x86-133 with and without LINBRST enabled on my Biostar or M919 socket 3 to see what enhancement it has. If you recall from the Ultimate 486 Benchmark Comparison, with ALL Cyrix enhancements off, the ALU performance dropped 13%. But there were a lot of other features in enabled/disabled here. I suspect LINBRST not being that significant on the 5x86 chips because I recall the other features that I enabled making a significant difference, however I am not exactly sure how this scales to the 6x86. Maybe somebody with a non-TX chipset can give this a quick test with Landmark, Speedsys, and Quake?

I will need to test all stable register bits with the Cyrix 6x86's before tabulating any results.

I saw the text file you uploaded in the other thread about undocumented voltage, and was intent on testing the settings with a multi-meter, but I somehow got distracted. I will test the undocumented voltage settings with a multi-meter; no CPU required, that could be too scary for me.

It may be that those voltage predictions are way off (or right on the ball). It really depends on what the motherboard manufactorer selected for the resistance of the reference resistor on the regulation circuit. I've seen some strange things with different makers. For variable voltage regulators, sometimes they vary the reference resistance, and sometimes the standard resistance. And sometimes, you don't know what the heck is going on, particularly on PC Chips boards. I've traced out one PC Chips regulation circuit and thought it was the siliest way to do things -- either the engineer had no idea what he was doing, had only select spare parts on hand, or they were trying to save 5 cents per board with a Mickey Mouse approach.

At any rate, direct measurement with the AZZA using all possible jumper combinations is probably the quickest, surest approach.

I won't get to this "586" project for some time. I am going on holiday for a month starting now and will be without internet access. All the best!

Plan your life wisely, you'll be dead before you know it.

Reply 67 of 280, by Tetrium

User metadata
Rank l33t++
Rank
l33t++

If you have a Pentium non-MMX chip, theres no way you can overvolt those, as those are supposed to be run at 3.3v. Underclocking them to, say, 75Mhz (1.5 x 50Mhz) will make a later non-MMX chip probably boot at much lower voltages (the non-MMX chip came in different manufacturing processes, the later ones scale better).
And even with the MMX chip (which runs at 2.8v or 2.9v) theres really no danger if you run them for 20 seconds at 3.3v, especially if you underclock them.
Remember that some of these chips are even able to run @ a whopping 300Mhz!

They can hardly be killed and a replacement is like €1 each anyway 😉

Edit:I "could" do the LINBRST vs non-LINBRST benchmarking, but only after I finished with my attic which is probably a couple weeks later (sorted all the motherboards, atm busy with the optical/floppy drives and disks and I got a LAN party this weekend).
I do have the required Cyrix chips laying around so that shouldn't be a problem.

Whats missing in your collections?
My retro rigs (old topic)
Interesting Vogons threads (links to Vogonswiki)
Report spammers here!

Reply 68 of 280, by swaaye

User metadata
Rank l33t++
Rank
l33t++

Socket 5/7 boards support at least up to 3.52v because that's what many older CPUs need. This includes chips like K5, some Pentiums, and the Winchips.

So you can overvolt a 3.3v chip a little.

Reply 70 of 280, by Tetrium

User metadata
Rank l33t++
Rank
l33t++

Not by me, still working on the attic.

Btw, what (preferably bootable) program would you suggest I use?

Whats missing in your collections?
My retro rigs (old topic)
Interesting Vogons threads (links to Vogonswiki)
Report spammers here!

Reply 71 of 280, by feipoa

User metadata
Rank l33t++
Rank
l33t++

All the DOS programs from the Ultimate 486 Benchmark Comparison that are listed in Appendix 1 should be runable from a diskette (with the exception of Doom and Quake). I'd probably start with these programs as they are pretty quick to run, then maybe squeeze in Quake if conditions permit.

Without digging into the mechanism behind Cyrix's LINBRST, this (experimental) shotgun approach is by far the simplest method to determine how effective LINBRST is. Unfortunately, I have had to pack my testbed for the time being and will not have the opportunity to run this test myself for some time.

Plan your life wisely, you'll be dead before you know it.

Reply 72 of 280, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Corrected some typos and grammatical errors to The Ultimate 486 Benchmark Comparison. Added a few lines here and there where appropriate. This seems to be a pretty good way to circumvent occasional insomnia.

Plan your life wisely, you'll be dead before you know it.

Reply 73 of 280, by iulianv

User metadata
Rank Member
Rank
Member

Is there any newer version of the .pdf other than the one on the first thread page? I'm about to get my hands on an IBM Blue Lightning DX2/66 CPU, but until I do I was curious about what to expect from it... and I cannot seem to find it anywhere through the benchmark results, although it is present in the "family photo".

Thanks.

Reply 74 of 280, by luckybob

User metadata
Rank l33t
Rank
l33t
iulianv wrote:

Is there any newer version of the .pdf other than the one on the first thread page? I'm about to get my hands on an IBM Blue Lightning DX2/66 CPU, but until I do I was curious about what to expect from it... and I cannot seem to find it anywhere through the benchmark results, although it is present in the "family photo".

Thanks.

I maybe wrong, but the IBM chips were in fact re-branded Cyrix chips.

It is a mistake to think you can solve any major problems just with potatoes.

Reply 75 of 280, by sliderider

User metadata
Rank l33t++
Rank
l33t++
luckybob wrote:
iulianv wrote:

Is there any newer version of the .pdf other than the one on the first thread page? I'm about to get my hands on an IBM Blue Lightning DX2/66 CPU, but until I do I was curious about what to expect from it... and I cannot seem to find it anywhere through the benchmark results, although it is present in the "family photo".

Thanks.

I maybe wrong, but the IBM chips were in fact re-branded Cyrix chips.

Seeing as Cyrix didn't have their own fab, wouldn't that make the Cyrix chips the rebranded ones?

Reply 76 of 280, by feipoa

User metadata
Rank l33t++
Rank
l33t++

@iulianv

The pdf file on the first thread page is the newest one. It has been updated several times since the original post. If you look at the footer of the PDF, it has a date stamp of 23 July 2011; so its pretty new.

I beleive the Cyrix DX2-66 and IBM Blue Lightening DX2-66 are the same chips, so there would have been little purpose in testing the IBM faced version.

Plan your life wisely, you'll be dead before you know it.

Reply 77 of 280, by luckybob

User metadata
Rank l33t
Rank
l33t
sliderider wrote:
luckybob wrote:
iulianv wrote:

Is there any newer version of the .pdf other than the one on the first thread page? I'm about to get my hands on an IBM Blue Lightning DX2/66 CPU, but until I do I was curious about what to expect from it... and I cannot seem to find it anywhere through the benchmark results, although it is present in the "family photo".

Thanks.

I maybe wrong, but the IBM chips were in fact re-branded Cyrix chips.

Seeing as Cyrix didn't have their own fab, wouldn't that make the Cyrix chips the rebranded ones?

Same difference.

It is a mistake to think you can solve any major problems just with potatoes.

Reply 78 of 280, by Tetrium

User metadata
Rank l33t++
Rank
l33t++
luckybob wrote:

I maybe wrong, but the IBM chips were in fact re-branded Cyrix chips.

Cyrix designed their own chips but had other fabs produce them. In effect, NVidia does the same, they don't produce their own chips
And yes, those Cyrix's and IBM's were in effect the same chip

Whats missing in your collections?
My retro rigs (old topic)
Interesting Vogons threads (links to Vogonswiki)
Report spammers here!

Reply 79 of 280, by feipoa

User metadata
Rank l33t++
Rank
l33t++

As I eluded to in the monologue of the Ultimate 486 Benchmark Comparison, I'm now beginning to setup a test environment to determine what effect the noted Cyrix 5x86 enhancements have over the resultant benchmarks of this study.

I'll be employing a Cyrix 5x86-133 with all enhancements ON (*except for BTB). These results will be considered optimal. I'll run the test on a Biostar MB8433-UUD v3 with 64 MB RAM and 512 KB cache in Write-back mode (all RAM cached). L2 cache wait states at 2-1-1 and RAM at 0/0.

The user-adjustable features I'll be testing are:

1) I/O delay, 0 to 128 clock delay (0 = optimal)
2) DTE_E (1 = optimal)
3) MEM_BYP (1 = optimal)
4) FP_FAST (1 = optimal)
5) LINBRST (1 = optimal)
6) BWRT (1 = optimal)
7) WBAK (1 = optimal)
8) LSSER (0 = optimal)
9) WT1 (0 = optimal, but I don't expect 1 to yield better performance)
10) LOOP (1 = optimal)
11) RSTK (1 = optimal)
12*) BTB (1 = optimal, if supported)

*BTB may work if LOOP is off (0), but is flakey in Windows. Better luck with Stepping 1 CPUs.

The strategy is to leave all features on, then turn one off and note the decrease in bechmark scores (run all tests per change), turn it back in, then turn off a different feature until all features have been tested. I'd then create a bar chart of the results, again with Overall, ALU, and FPU performance per feature.

So the gong show of tests will need to be run 13 times. Does anybody have any suggestions for improvement before I get started (this will take a long time, fyi)?

Plan your life wisely, you'll be dead before you know it.