VOGONS


=My Cyrix 5x86 systems : 120MHz vs 133MHz=

Topic actions

First post, by rgart

User metadata
Rank Oldbie
Rank
Oldbie

EDIT: 21.04.16 - Some minor adjustments.

Here are the two systems that have been consuming some of my time

[SYSTEM 1]

◙ Cyrix 5x86 120GP x2 x3 (stepping 1 revision 3)
◙ ECS Elite UM8810P-AIO Rev 2.1
◙ Phoenix BIOS 4.04 PLUG AND PLAY (18/09/95) UMC8881F, UMC8886F
◙ 256 KB 15ns TAG - 15ns Cache (UM6125AK-15)
◙ 64MB RAM Fast Page Memory 60ns (2 x 32MB SIMMS)
◙ PCI Matrox Mystique 220 (MGA-1164SG) (1997)
◙ ISA Creative Labs Sound Blaster 32 CT-3670 (PNP)
◙ IDE Maxtor 1.01GB Hard Disk Drive
◙ IDE Maxtor 4.01GB Hard Disk Drive
◙ IDE Generic CDR Drive

CPU Voltage 3.45 (Interposer PGA Adapter)
PCI CLCK / HOST CLCK - 1:1
FSB (BUS) SPEED – 40mhz
CPU SPEED – 120mhz
CLCK MULTIPLIER - 3
CACHE TIMINGS – 2,1,1,1
L1: Write Back(WB)
L2: Write Back(WB)
DRAM Read Wait State – 1WS
DRAM Write Wait State – 1WS
Host-to-PCI Wait State – 0WS

ZgcCUmt.jpg
dKUdZIg.jpg
a0BpufM.jpg
H3T0WMh.jpg
kfqzv1S.jpg
D2iWXFA.jpg
FosZVIS.jpg
wieFzyp.jpg
0fozEAm.jpg

[SYSTEM 2]

◙ Cyrix 5x86 133GP x4 (stepping 0 revision 5)
◙ Biostar MB-8433UUD-A v2.0 (voltage trimmer mod, feipoa’s modded bios) AWARD 4.51PG BIOS (20/05/96) UMC 8881F
◙ 512 KB 10ns TAG (W24257AK-10) 10 ns cache (IS61C1024-10N) – Chinese Reprints
◙ 64 MB RAM “SEC KOREA” Fast Page Memory – 16x36 PARITY SIMM (60ns) (1 x 72 pin)
◙ PCI Matrox Millennium G200, 8 MB 250mhz SDRAM (MGA-G200) GPU Fan (1998)
◙ ISA Creative AWE64 Gold, CT4390 w/28MB RAM (Simmconn board)
◙ SCSI Adaptec AHA-2940U2W Ultra2-LVD (Bios 2.00.0)
◙ SCSI Compaq 18.2GB Wide Ultra3 SCSI 10,000RPM
◙ SCSI Plextor PX-32TS CD-ROM
◙ Roland Sound Canvas External Module (SC-55ST)

CPU Voltage (Trimmer)- 3.6
PCI CLCK / HOST CLCK - 1:1
FSB (BUS) SPEED – 33mhz
CPU SPEED – 133mhz
CLCK MULTIPLIER - 4
CACHE TIMINGS – 2,1,1,1
L1: WRITE BACK (WB)
L2: WRITE THROUGH(WT)
DRAM Read Wait State – 0WS
DRAM Write Wait State – 0WS
Host-to-PCI Post Write – 0WS

rEbRhBx.jpg
g5TbQTp.jpg
BU9MRH5.jpg
OkfwnMl.jpg
CROwxv6.jpg
9iunEGh.jpg
RUwf96q.jpg
NQGKmWv.jpg
lYidTCr.jpg
tDIvtIZ.jpg
ie4haeU.jpg
1tfB6EP.jpg
pa6Dj4V.jpg
71JEeOe.jpg
3e8ntJf.jpg

BENCH RESULTS:

(Cyrix 120MHz)

Operating System: MSDOS 6.22

◙ speedsys (v4.78): 67.81
◙ 3dbench2 (v1.0): 96.0
◙ Pcpbench /vgamode v1.97 (320x200): 20.7
◙ Pcpbench 101 v1.97 (640x480): 7.4
◙ Pcpbench 103 v1.97 (800x600): 5.5
◙ Doom 2 Full Version v1.666 FPS: 39.8
◙ Quake Share Ware v1.06 FPS (320x200): 15.4

(Cyrix 133MHz)

Operating System: Windows 98 SE with Unofficial Windows98 SE Service Pack 3 Update
MSDOS.SYS : GUI = 0

◙ speedsys (v4.78): 75.26
◙ 3dbench2 (v1.0): 90.1
◙ Pcpbench /vgamode v1.97 (320x200): 23.7
◙ Pcpbench 101 v1.97 (640x480): 7.2
◙ Pcpbench 103 v1.97 (800x600): 5.3
◙ Doom 2 Full Version v1.666 FPS: 42.1
◙ Quake Share Ware v1.06 FPS (320x200): 16.3

AUTOEXEC.BAT and CONFIG.SYS by-passed for testing.

Cyrix Enhancements the same for both systems. (Peter Moss Utility)

sQdPeQA.jpg

WINDOWS: LH 5x86 /loop_en=on /rstk_en=on /lsser=off /fp_fast=on /btb_en=off
DOS: LH 5x86 /loop_en=off /rstk_en=on /lsser=off /fp_fast=on /btb_en=on

14.05.16 NOTE: RTSK_EN=ON should be turned OFF or it can cause some issues.
14.05.16 NOTE: BTB_EN=ON in Windows & DOS for Stepping 1 Revision 3 Cyrix 5x86 CPU
14.05.16 NOTE: BTB_EN=OFF in Windows and ON in DOS for Stepping 0 Revision 5 Cyrix 5x86 CPU

Other Video hardware tested:
Rendition Verite V2200 (4MB)
Matrox Millenium (MGA-2064W Storm) (2MB)
S3 325 VIRGE GUI Accelerator (2MB)

Cyrix 5x86 120 Comments:

This system is stable and fast. I have never had it crash. If i tighten the DRAM wait states any further and take them to 0 the PC will not POST and I have to Jumper JP30 and clear CMOS to boot.
This system has successfully run 5 passes in Memtest86 v4.0 + loads himem.sys without issue and has had hours of testing in Doom, Heretic, Duke3d, Hexen and Quake.
The UM8810P-AIO motherboard along with a Matrox Millenium/Mystique video card appears to be a solid choice for a Cyrix 5x86 based system.

Cyrix 5x86 133 Comments:

This system is extremely fast. Windows 98 is very responsive and It's easy to forget your using a Socket 3 system.
The SCSI interface really makes testing between these two systems unfair as it completely blows away the other systems IDE interface for load time on almost every piece of software tested.
This PC has successfully run 10 passes in Memtest86 4.0 + loads himem.sys without issue and has been tested extensively in Doom, Heretic, Duke3d, Hexen, CPUBURN, Prime95 and Quake.

Since making some minor adjustments I have had the system crash a few times while playing Hexen today 🙁

I'll continue to try and ID this problem

jPwT5nN.png

Last edited by rgart on 2017-04-15, 13:21. Edited 27 times in total.

=My Cyrix 5x86 systems : 120MHz vs 133MHz=. =My 486DX2-66MHz=

Reply 1 of 142, by feipoa

User metadata
Rank l33t++
Rank
l33t++

Does the part number on that memory module chip say KM44V1600AK? If so, then you are using 3.3V memory in a 5V motherboard. Remove it immediately before you damage the chipset. Which eBay seller did you buy this from? If it is the one I am thinking of, I already warned him of this, but he continues to sell these as 5 V modules.

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

Reply 2 of 142, by rgart

User metadata
Rank Oldbie
Rank
Oldbie

Its a KM44V1600AK part.

I cant recall where I got them it was so long ago. I think it was 1800-MEMORY

RAM removed - they were the only single 72 pin 64MB Simms I had 😀

=My Cyrix 5x86 systems : 120MHz vs 133MHz=. =My 486DX2-66MHz=

Reply 3 of 142, by feipoa

User metadata
Rank l33t++
Rank
l33t++

That seller has a history of selling 3.3V memory advertised as 5V.

For the Cyrix 5x86-120 system, are you sure you need both of these settings at 1:

DRAM Read Wait State – 1WS
DRAM Write Wait State – 1WS

With a 40 MHz bus, the UMC board should be able to get away with 0ws / 0ws. At the worst case, I would expect 1ws / 0ws. Try different memory - a single stick if possible.

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

Reply 4 of 142, by rgart

User metadata
Rank Oldbie
Rank
Oldbie

Like I say if I change even one of those wait states to 0 the PC will not post until I jumper JP30 to clear CMOS

The latest revision of the UM8810 manual claims the PCI CLCK: HOST CLCK needs to be on 1: 1/2 specifically for the Cyrix 5x86 120GP but I have found stability at 1:1 with those wait states.

I have not yet tried all the RAM modules available to me on either of these systems.

What are the chances the RAM running at 5 volts was responsible for my earlier crash?
Have you had first hand experience of 3.3 volt RAM on a 5 volt motherboard damaging the chip-set?

The ebay seller also sells KM44C1000BJ-6 and HYB514100BJ-60 and claims they are 64MB 5V 60ns 72pin simms. I cant find a spec sheet. Can anyone confirm these are 5V? It was 1800-memory I picked up the RAM from.. I just sent him a message and the RAM was listed as 5 Volts.

=My Cyrix 5x86 systems : 120MHz vs 133MHz=. =My 486DX2-66MHz=

Reply 5 of 142, by feipoa

User metadata
Rank l33t++
Rank
l33t++
rgart wrote:

Have you had first hand experience of 3.3 volt RAM on a 5 volt motherboard damaging the chip-set?

Yes, but the damage to the chipset was not so black and white. It caused unpredictable BSOD's on a system that had been running stable for years. The BSOD's started immediately after putting in the wrong voltage RAM and did not go away when 5 V memory was put back in. I am not 100% sure if the chipset got damaged, or if it was some other component on the PCB.

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

Reply 6 of 142, by rgart

User metadata
Rank Oldbie
Rank
Oldbie
feipoa wrote:
rgart wrote:

Have you had first hand experience of 3.3 volt RAM on a 5 volt motherboard damaging the chip-set?

Yes, but the damage to the chipset was not so black and white. It caused unpredictable BSOD's on a system that had been running stable for years. The BSOD's started immediately after putting in the wrong voltage RAM and did not go away when 5 V memory was put back in. I am not 100% sure if the chipset got damaged, or if it was some other component on the PCB.

oh dear 😀

=My Cyrix 5x86 systems : 120MHz vs 133MHz=. =My 486DX2-66MHz=

Reply 7 of 142, by rgart

User metadata
Rank Oldbie
Rank
Oldbie

I'm now using what I can confirm is 16MB 60NS 5V FPM RAM

Part number HYM536410BM-Series
Category Memory => DRAM => EDO/FPM DRAM => Modules => 16 MB => FPM
Title FPM
Description FP DRAM Module Based on 4mx4 DRAM, With Parity, 5v, 2k-refresh: 4mx36
Company Hyundai Electronics America

fvbMS9i.jpg

Ill continue testing with these. Hopefully the chipset is fine.

Here are the 64MB Modules I used in the MB8433 which I thought were 5V and were listed as on ebay as "64MB FPM PARITY 60NS SIMM 72-PIN 5V"

Filename
Memory-Samsung-KM44V16100A(1).pdf
File size
249.91 KiB
Downloads
208 downloads
File license
Fair use/fair dealing exception

3.3V
tDIvtIZ.jpg

I haven't been able to find a spec sheet on my 32 MB modules.

Last edited by rgart on 2016-04-24, 09:39. Edited 4 times in total.

=My Cyrix 5x86 systems : 120MHz vs 133MHz=. =My 486DX2-66MHz=

Reply 8 of 142, by rgart

User metadata
Rank Oldbie
Rank
Oldbie

I believe the 32MB BUFFALO EMW-64M (TC5117400CSJ-60) modules I have are also 5 volt

Filename
TOSHS094-1.pdf
File size
939.27 KiB
Downloads
180 downloads
File license
Fair use/fair dealing exception

slLsZa2.jpg

Last edited by rgart on 2016-04-24, 09:43. Edited 6 times in total.

=My Cyrix 5x86 systems : 120MHz vs 133MHz=. =My 486DX2-66MHz=

Reply 9 of 142, by rgart

User metadata
Rank Oldbie
Rank
Oldbie

I'm now using the above BUFFALO EMW-64M simms in both PC's.

1 x 32MB on the Cyrix 120 system
1 x 32MB on the Cyrix 133 system

I have also removed rtsk=on from my dos batch file that turns on dos related cyrix enhancements. I did try timings of 3,1,1,1 with 0,0,0 WS on the Cyrix 120 system but it refuses to post without wait states 1. I installed 3 x 120mm deepcool case fans just for the hell of it - to move a bit of air and keep those parts that like to heat up cooler. So far things are stable on both systems.

I don't think I'll source 5V 60ns 64MB 72 pin simms because for the programs I run I don't really need them.

Last edited by rgart on 2016-04-21, 07:37. Edited 5 times in total.

=My Cyrix 5x86 systems : 120MHz vs 133MHz=. =My 486DX2-66MHz=

Reply 10 of 142, by sliderider

User metadata
Rank l33t++
Rank
l33t++

Beige PCI Powermacs numbered 7200-9600 use 5v 72 pin memory, so any memory specc'ed for those should work. They can use either FPM or EDO, though, so if your m'board doesn't support one of those then you'll have to buy RAM appropriate to the particular m'board.

Reply 11 of 142, by NJRoadfan

User metadata
Rank Oldbie
Rank
Oldbie
sliderider wrote:

Beige PCI Powermacs numbered 7200-9600 use 5v 72 pin memory, so any memory specc'ed for those should work. They can use either FPM or EDO, though, so if your m'board doesn't support one of those then you'll have to buy RAM appropriate to the particular m'board.

PCI Powermacs use 5V 168pin 64-bit DIMMs, not SIMMs. The Nubus Powermacs use 5v 72-pin SIMMs in pairs and should work on 486 boards.

That infamous 3.3v memory is rearing its ugly head again. I'm lucky none of it made it onto the 16MB 30pin SIMMs I got from them.

Reply 13 of 142, by rgart

User metadata
Rank Oldbie
Rank
Oldbie

Regarding the RAM.
I dont know if 3.3V RAM will damage a motherboard designed for 5V.
Has anyone besides Feipoa had first hand experience?
I received a response from 1800-memory this morning...

Dear xxxxxx,

My technician became available this afternoon.

I did show him the picture you sent. He immediately said it is 5V memory and that the chips on the module are 3.3V but are 5V-tolerant. I then asked him if using this memory in a 5V motherboard would damage the motherboard. He did say that there is no way this memory could damage a motherboard.

We will be happy to accept this memory back for a full refund of the purchase price of the memory.

Last edited by rgart on 2013-08-27, 10:58. Edited 1 time in total.

=My Cyrix 5x86 systems : 120MHz vs 133MHz=. =My 486DX2-66MHz=

Reply 14 of 142, by rgart

User metadata
Rank Oldbie
Rank
Oldbie

the RAM issue aside....

I have no idea why the Cyrix 120 systems is scoring better than the Cyrix 133 in pcpbench and 3dbench2

Both systems are bypassing autoexec.bat and config.sys and using the same Cyrix enhancements.
The only difference software wise is the Cyrix 120 is running DOS 6.22 while the Cyrix 133 runs Windows 98 but not the GUI mode. Only a command prompt at startup.

The Cyrix 120 system has a higher bus speed of 40mhz compared to 33 but the CPU Is obviously slower than the Cyrix 133mhz

The Cyrix 120 system has wait states and only 256k 15ns cache.
The Cyrix 133 system has 0 wait states and 512k 10ns cache.

Feipoa has mentioned my Chinese cache reprints look fake but even with 256k 15ns cache from a 1996 era machine it is still scoring lower than the Cyrix 120.

Can the bus speed really be that influential? I somehow doubt it when we are talking 40mhz compared to 33mhz.

=My Cyrix 5x86 systems : 120MHz vs 133MHz=. =My 486DX2-66MHz=

Reply 15 of 142, by sliderider

User metadata
Rank l33t++
Rank
l33t++

Bus speed is more important. If you have the 120 running at 60mhz x2 it would be even faster than the 133 running at 33 x 4. A 50mhz 486DX would be faster than a 486 DX2-50 because the bus speed is double even though the CPU core runs at the same speed. The CPU may be running 13mhz slower but the rest of the system is running 7mhz faster and the CPU is just one component among many that determines how fast a system is. That's why when upgrade manifacturers were cranking out 450mhz K6-2 upgrades and such for socket 7 boards, the tech writers warned people to test other components first to see if the CPU was the biggest bottleneck or whether it was something else that would be more effective at speeding up the system for less money or to just save for a new motherboard that supports higher fsb clocks instead.

Reply 16 of 142, by dirkmirk

User metadata
Rank Oldbie
Rank
Oldbie

I had a quick look through feiopas ultimate 486 benchmark programme again and he only tested the cyrix 5x86-120/40fsb without the cyrix enhancements? Could the 120 actually be a better cpu when we're talking stock fsb clocks? Obviously transfer speeds would be quicker but I though the higher cpu speed would help for gaming in the 133? perhaps im wasting my time trying to track down a 133 looks like overclocking a 120 to 133 is the way to go.

Reply 17 of 142, by rgart

User metadata
Rank Oldbie
Rank
Oldbie

I was messing around with the Cyrix 133 system last night.

L2 on Write Back seems to work great from a command prompt.

But windows 98 GUI doesn't like it at all.

I am considering retiring the 133 for now and going back to running the Cyrix 120 at 60 x 2.

I was reading my other thread where I was running the 120 at 60x2 and 66x2 and I cant help but notice the pcpbench scores were a full two points higher.

dirk: I made some vid's for myself testing the 120 and 133 system side by side in a few games and they are ever so close. In some ways the 120 IDE system appears faster than the 133 SCSI system. The 133 system killed the 120 on loading time but that was only because of the SCSI interface. Like some here have noted previously the UM8810 board bus mastering leaves much to be desired. It really dislikes all the SCSI cards I have tried so far. The other thing about SCSI is you have to search for quiet drives - Some are terribly loud and NOT for gaming because of this.

=My Cyrix 5x86 systems : 120MHz vs 133MHz=. =My 486DX2-66MHz=

Reply 18 of 142, by feipoa

User metadata
Rank l33t++
Rank
l33t++
rgart wrote:
Regarding the RAM. I dont know if 3.3V RAM will damage a motherboard designed for 5V. Has anyone besides Feipoa had first hand […]
Show full quote

Regarding the RAM.
I dont know if 3.3V RAM will damage a motherboard designed for 5V.
Has anyone besides Feipoa had first hand experience?
I received a response from 1800-memory this morning...

Dear xxxxxx,

My technician became available this afternoon.

I did show him the picture you sent. He immediately said it is 5V memory and that the chips on the module are 3.3V but are 5V-tolerant. I then asked him if using this memory in a 5V motherboard would damage the motherboard. He did say that there is no way this memory could damage a motherboard.

We will be happy to accept this memory back for a full refund of the purchase price of the memory.

According to the specification sheet, that memory is only 3.0 - 3.6 V tolerant under sustained use. The absolute maximum ratings this memory can handle is 4.6 V. I will quote from the specification sheet, "Permanent device damage may occur if 'ABSOLUTE MAXIMUM RATINGS' are exceeded. Functional operation should be restricted to the conditions as detailed in the operational sections of this data sheet [3.3 V +- 0.3V]. Exposure to absolute maximum rating conditions for extended periods may affect device reliability."

This is the company who also sold me the same 64 MB sticks. The 4 pieces I have look identical to yours. I ran them in a system for 2 hours to run MemTest. I do not recall which motherboard it was, but it wasn't one of great importance, so I didn't care. Although I did not receive any MemTest errors, the memory chips get a lot hotter than standard 5V SIMMs per the same 2 hr test. It was the heat from the SIMMs which tipped me off to check the datasheet for their voltage. I would not run a system long-term with these SIMMs installed. Actually, I wouldn't run them at all in a 5 V system (refer to dire warnings from manufacturer spec. sheet noted above in bold).

These particular SIMMs seem a lot less damaging than the other IBM 3.3V SIMMS I tested 10 years ago, which damaged my UMC-based PC Chips M919 motherboard.

I strongly recommend you not use this guy's 3.3V memory in your 5V motherboard. The only way to properly get away with 3.3V memory in a 5V system is if the memory module has a drop regulator of some sort on it (e.g. double doide or voltage regulator IC), and these sticks do not.

Onto the bus topic.

1) Highly graphic applications are very sensitive to PCI bus speed. The benchmarks you have chosen to compare the two systems are highly dependent on the PCI bus speed.

2) Double-banked cache should be faster than single-banked cache because it can interleave between the banks, however I have not found this to matter much for benchmarks (but I admit, I have not tested it extensively either). Your 120 MHz system uses 256KB double-banked, whereas your 133 MHz system uses 512KB single-banked. With 512KB over 256KB, you probably get a few percent increase in cache hits (if you are using 32MB memory), however I'm not sure what the performance loss is due to the non-interleaved operation.

3) When using a 40 MHz bus, if you are able to leave the L1 cache wait states at the very fastest, this setup is naturally faster than a 33 MHz bus, also with the fastest wait states. Most motherboards will tolerate 40 MHz and 256KB double-banked with the fastest wait states. Most motherboards will NOT tolerate 40 MHz, fastest wait states, when the cache is greater than 256KB and the CPU is a Cyrix 5x86. For other motherboards, they will tolerate 40 MHz, 1024KB cache double-banked, fastest wait states, but only with an AMD X5-160, and not the Cyrix 5x86-120. If you want to limit your cache to 256KB, double-banked, then a Cyrix 5x86-120 at 3x40 is a nice system. Your cacheable range is limited to 32 MB (WB) or 64 MB (WT). I like to run systems with 128 MB, so you need at least 512 KB cache to have all your memory cacheable.

4) As for the Ultimate 486 Benchmark Comparison, it was noted that for the 40 MHz CPUs tested (except POD100), the employed motherboard adds an auto 2/3 PCI multiplier, so the PCI bus in those cases was running only at 27 MHz. I wish I had caught this before running the tests. In the write-up, I note which benchmarks are affected by this PCI bus speed decrease. Sometime before I die, I'll fix the 486 benchmark comparison such that I use a 40 MHz PCI bus for 40 MHz FSB tests. I'll also correct the cacheable limit issue (though this is not important for a CPU-to-CPU comparison from this table only, it only becomes an issue for outsiders comparing their systems to this chart, and for certain Windows-based benchmarks only. Refer to documentation.).

5) To run a Cyrix at 2x60 has one drawback. Your PCI bus will be run at 30 MHz, however if you have a 2/3-capable motherboard, you can run it at 40 MHz if you can find PCI cards comfortable with this speed. Refer to the Ultimate 686 Benchmark Comparison for results of running various 486 CPUs at above stock FSBs (Cyrix 2x50, 2x60, 2x66, and Intel 2x66 - AMD X5-2x66 results tabulated, but not included in the table yet).

For your CPUs of interest, here are the important numbers:

Cyrix 5x86 - 2x60 (assume 40 MHz PCI), 3-2-2-2 L2 and 1ws/0ws memory
Cachechk
L1: 247 MB/s
L2: 92 MB/s
Memory: 63 MB/s

Cyrix 5x86 - 3x40 (assume 40 Mhz PCI), - 2-1-1-1 L2 and 0ws/0ws memory
L1: 247 MB/s
L2: 96 MB/s
Memory: 55 MB/s

If you run the PCI bus at 40 Mhz for both systems, the 2x60 MHz system has the memory speed advantage, whereas the 3x40 system has the L2 cache speed advantage. Personally, I'd take the 2x60 Mhz (3-2-2-2) option because I know from testing that this configuration will work with at least 512KB cache, whereas 512KB with 3x40 Mhz (2-1-1-1) seems a lot more fussy with finding a stable motherboard. Again, this assumption is for a Cyrix 5x86 only. The AMD X5 is naturally slower all around and the same assumptions do not always apply. I only ever had one motherboard which was really stable with the Cyrix 5x86 - 3x40, single-banked 512KB cache, and 2-1-1-1 timing. Keep in mind that your requirements for stability may differ from mine.

I was unable to run at 60/66 MHz FSB with L2 timings faster than 3-2-2-2 and 512KB cache. I did not test the fastest timing with 256KB double-banked cache, however I still suspect 3-2-2-2 is still as good as you'll get. The only way to stably beat the 96 MB/s speed of the 5x86 - 3x40 is to run a 5x86 at 2x66, which yields 274/102/70 MB/s (L1/L2/Memory). Keep in mind that your PCI bus will be at 33 Mhz instead of 40 MHz, so a Cyrix 5x86 - 2x60 (40 MHz PCI) is a good match for a Cyrix 5x86 - 2x66 (33 MHz PCI) inasmuch as heavily graphic applications are concerned. Due to the need to add additional wait states, a 66 Mhz FSB system is not even close to twice as fast as a 33 MHz FSB system.

My own rule of thumb for these two Cyrix 5x86-120 configurations is as follows,
If you have 256KB, double-banked cache, run a Cyrix 5x86 at 3x40, 40 Mhz PCI, and 2-1-1-1, 0ws/0ws.
If you desire 512KB, single-banked cache or greater, run a Cyrix 5x86 at 2x60, 40 MHz PCI, 3-2-2-2, 1ws/0ws.

I forgot to mention, your maximum L2 speeds are also sensitive to the number of RAM sticks you have installed, the less the better!

This topic dives quite deep into the frustrating and fuzzy side of 486's.

rgart wrote:

I am considering retiring the 133 for now and going back to running the Cyrix 120 at 60 x 2.

6) If you are unhappy with the Cyrix 5x86-133/4x, I'd be happy to buy it from you at fair market value 😀

rgart wrote:

Was messing around with the Cyrix 133 system last night.

L2 on Write Back seems to work great from a command prompt.

But windows 98 GUI doesn't like it at all.

Did you remember to set the TAG to 7+1 when you set L2 into WB mode? TAG should be 8+0 for L2 in WT mode. If you do not adhere to this, you'll run into all kinds of strange things.

rgart wrote:

Feipoa has mentioned my Chinese cache reprints look fake but even with ...

I think I said only your 32kx8 Winbond TAG SRAM looks fake. The font is very non-standard for Winbond chips.

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

Reply 19 of 142, by rgart

User metadata
Rank Oldbie
Rank
Oldbie
feipoa wrote:

I strongly recommend you not use this guy's 3.3V memory in your 5V motherboard. The only way to properly get away with 3.3V memory in a 5V system is if the memory module has a drop regulator of some sort on it (e.g. double doide or voltage regulator IC), and these sticks do not.

Yup I removed the RAM right after I read your post.
Any crashes I was having have now disappeared and the system has been stable since.
I never use more than 1 SIMM. I also find the system more stable with less.

After some disagreement....1800memory is sending me the correct chips at no cost. Ill be sure to check the data sheet when it arrives.
It will be nice to be using a full 64MB again.

feipoa wrote:
Onto the bus topic. […]
Show full quote

Onto the bus topic.

1) Highly graphic applications are very sensitive to PCI bus speed. The benchmarks you have chosen to compare the two systems are highly dependent on the PCI bus speed.

2) Double-banked cache should be faster than single-banked cache because it can interleave between the banks, however I have not found this to matter much for benchmarks (but I admit, I have not tested it extensively either). Your 120 MHz system uses 256KB double-banked, whereas your 133 MHz system uses 512KB single-banked. With 512KB over 256KB, you probably get a few percent increase in cache hits (if you are using 32MB memory), however I'm not sure what the performance loss is due to the non-interleaved operation.

3) When using a 40 MHz bus, if you are able to leave the L1 cache wait states at the very fastest, this setup is naturally faster than a 33 MHz bus, also with the fastest wait states. Most motherboards will tolerate 40 MHz and 256KB double-banked with the fastest wait states. Most motherboards will NOT tolerate 40 MHz, fastest wait states, when the cache is greater than 256KB and the CPU is a Cyrix 5x86. For other motherboards, they will tolerate 40 MHz, 1024KB cache double-banked, fastest wait states, but only with an AMD X5-160, and not the Cyrix 5x86-120. If you want to limit your cache to 256KB, double-banked, then a Cyrix 5x86-120 at 3x40 is a nice system. Your cacheable range is limited to 32 MB (WB) or 64 MB (WT). I like to run systems with 128 MB, so you need at least 512 KB cache to have all your memory cacheable.

4) As for the Ultimate 486 Benchmark Comparison, it was noted that for the 40 MHz CPUs tested (except POD100), the employed motherboard adds an auto 2/3 PCI multiplier, so the PCI bus in those cases was running only at 27 MHz. I wish I had caught this before running the tests. In the write-up, I note which benchmarks are affected by this PCI bus speed decrease. Sometime before I die, I'll fix the 486 benchmark comparison such that I use a 40 MHz PCI bus for 40 MHz FSB tests. I'll also correct the cacheable limit issue (though this is not important for a CPU-to-CPU comparison from this table only, it only becomes an issue for outsiders comparing their systems to this chart, and for certain Windows-based benchmarks only. Refer to documentation.).

5) To run a Cyrix at 2x60 has one drawback. Your PCI bus will be run at 30 MHz, however if you have a 2/3-capable motherboard, you can run it at 40 MHz if you can find PCI cards comfortable with this speed. Refer to the Ultimate 686 Benchmark Comparison for results of running various 486 CPUs at above stock FSBs (Cyrix 2x50, 2x60, 2x66, and Intel 2x66 - AMD X5-2x66 results tabulated, but not included in the table yet).

For your CPUs of interest, here are the important numbers:

Cyrix 5x86 - 2x60 (assume 40 MHz PCI), 3-2-2-2 L2 and 1ws/0ws memory
Cachechk
L1: 247 MB/s
L2: 92 MB/s
Memory: 63 MB/s

Cyrix 5x86 - 3x40 (assume 40 Mhz PCI), - 2-1-1-1 L2 and 0ws/0ws memory
L1: 247 MB/s
L2: 96 MB/s
Memory: 55 MB/s

If you run the PCI bus at 40 Mhz for both systems, the 2x60 MHz system has the memory speed advantage, whereas the 3x40 system has the L2 cache speed advantage. Personally, I'd take the 2x60 Mhz (3-2-2-2) option because I know from testing that this configuration will work with at least 512KB cache, whereas 512KB with 3x40 Mhz (2-1-1-1) seems a lot more fussy with finding a stable motherboard. Again, this assumption is for a Cyrix 5x86 only. The AMD X5 is naturally slower all around and the same assumptions do not always apply. I only ever had one motherboard which was really stable with the Cyrix 5x86 - 3x40, single-banked 512KB cache, and 2-1-1-1 timing. Keep in mind that your requirements for stability may differ from mine.

I was unable to run at 60/66 MHz FSB with L2 timings faster than 3-2-2-2 and 512KB cache. I did not test the fastest timing with 256KB double-banked cache, however I still suspect 3-2-2-2 is still as good as you'll get. The only way to stably beat the 96 MB/s speed of the 5x86 - 3x40 is to run a 5x86 at 2x66, which yields 274/102/70 MB/s (L1/L2/Memory). Keep in mind that your PCI bus will be at 33 Mhz instead of 40 MHz, so a Cyrix 5x86 - 2x60 (40 MHz PCI) is a good match for a Cyrix 5x86 - 2x66 (33 MHz PCI) inasmuch as heavily graphic applications are concerned. Due to the need to add additional wait states, a 66 Mhz FSB system is not even close to twice as fast as a 33 MHz FSB system.

My own rule of thumb for these two Cyrix 5x86-120 configurations is as follows,
If you have 256KB, double-banked cache, run a Cyrix 5x86 at 3x40, 40 Mhz PCI, and 2-1-1-1, 0ws/0ws.
If you desire 512KB, single-banked cache or greater, run a Cyrix 5x86 at 2x60, 40 MHz PCI, 3-2-2-2, 1ws/0ws.

I forgot to mention, your maximum L2 speeds are also sensitive to the number of RAM sticks you have installed, the less the better!

This topic dives quite deep into the frustrating and fuzzy side of 486's.

Did you remember to set the TAG to 7+1 when you set L2 into WB mode? TAG should be 8+0 for L2 in WT mode. If you do not adhere to this, you'll run into all kinds of strange things.

Thanks for all that awesome info feipoa. Ill continue to refer back to this. Yeah mate I always use 7+1 with L2 in WB mode and 8+0 for L2 in WT mode. I don't believe my Winbond tag is fake because there appear to be no symptoms. I have had no issues and I am happy with the speed.

I recently changed video cards in this system. I was very reluctant to remove the Matrox....and the Matrox did score better but I'm now using an ATI Rage 128 PRO 32MB. Some of my benches show my FPS has now dropped ever so slightly but I think I'm better off with the ATI RAGE. It probably doesnt matter in reality because the system is seriously bottle-necked by the 33mhz FSB. What a blessing that the ATI Rage works in the Biostar MB-8433 because the ECS UM8810 wouldn't even turn on. I admit I have given up on getting SCSI working in the UM8810 board. No matter what I do with SCSI.....the system is as unstable as hell but as soon as I switch the IDE the system is rock solid stable.

I'm currently customizing the case fan and CPU heat-sink and fan and Ill be sure to post pics when I'm done. I have noted how thermally sensitive the Cyrix 5x86 CPU's are.

I have been putting serious hours into using this system and I think its impressive. Quake runs like a dream with the default resolution. Nothing like what I see on the am5x86 133ADZ YOUTUBE videos

Branch prediction running perfectly stable in DOS.
RSTK, Loop, FpFast = on and Lsser=off running stable in Windows.

Does anyone know the difference between the ATI RAGE 128 PRO 32MB and the ATI RAGE 128 VR 32MB?
Visually all I can see are less chips on the PRO and the BIOS is marked "VGA BIOS Z" instead of just "VGA BIOS"

Last edited by rgart on 2016-04-21, 07:54. Edited 3 times in total.

=My Cyrix 5x86 systems : 120MHz vs 133MHz=. =My 486DX2-66MHz=