For the time being, it's more economical to just get a wide SCSI drive.

I have an IBM 5x86C coming in the mail. I'm really interested in getting a Cyrix based 5x86 chip going on this board. I was not able to get anywhere with the 3 other Cyrix branded 5x86 chips I had already tried.

I can't remember if I read about the IBM 5x86C on VOGONS or on the CPU-world forums, but my understanding about it is:

-IBM rated it's own IBM 5x86C chips much more conservatively than Cyrix, potentially making them better for overclocking
-IBM 5x86C chips tend to be much later revisions than Cyrix branded chips, and often work well with branch prediction enabled
-Has nifty blue heatsink

Hopefully IBM also made a few tweaks to improve hardware compatibility too...

I have three VRMs in my possession. Hopefully one of them will do the job.

I got my Cyrix 5x86 chips going on this board, however the results are not what I had in mind. For whatever reason I can only use the 2x multiplier, not the 3x setting. What's even stranger is that in order to get the chip going in 2X mode I need to set the jumpers to 2.5x. When the jumpers are set to 2X or 3X, the system will not boot.

I did however get the Cyrix chip going at 100MHz by placing it in 2x50MHz mode. I have not yet turned on any of the goodies (stepping 1, rev3), however as it currently stands the CPU benches only 2/3 the speed of am5x86 at 160 but with similar memory I/O scores.

Perhaps my IBM 5x86C chip will fare better.

Wow. You guys weren't kidding about the Intel 486DX/4 chips being a lot faster than AMD/Cyrix parts. I tested mine at 120MHz (WT cache). It gave my am5x86 at 160 a serious run for its money.

Unfortunately my intel DX4 doesn't work on 50MHz bus at 100MHz, so I guess the chances of hitting 150MHz aren't good either. An intel dx4 at 150MHz would be pretty sweet though...

I noticed that my motherboard supports a 66MHz bus setting. I have been experimenting to see if any of my CPUs will work with it. Neither of my Cyrix 5x86 CPUs work. I don't believe I have any other CPUs except for maybe a few DX-50s that could run at that setting without being seriously overclocked. I am hoping that I will have a chance with the IBM 5x86C that is in the mail.

In the mean time, I have been lining up candidates for possible VLB cards that might work at 66MHz. It sounds pretty ridiculous, but I have a few that work well at 50MHz with 0WS, and I wonder if perhaps any of them will operate at 66MHz with 1WS. My 4 best cards are a Hercules Dynamite Power, Cardex Challenger (both based on ET4000W32P) and two cards based on ARK1000 from Octek and 2theMax.

While I was screwing around, I realised that I hadn't really done much real world testing with the ARK cards. I think in the benchmarks they didn't do quite as well as the Tseng chips. But I tried one out at 50MHz 1WS with Grand Theft Auto 1. Holy crap, this chip really shines at 50MHz. It was so fast I almost couldn't control it. I will have to run a few other tests on it to verify the results.

I was successfully able to get my system working at 66MHz clock for a few moments yesterday. I tested both Cyrix 5x86 CPUs and two intel DX-50 CPUs. This time I was actually able to get the Cyrix 5x86 chip to boot up at 133MHz for a few moments, but it wasn't stable.

The first DX-50 chip worked, but only with L1 cache turned off.
The second DX-50 chip worked with both L1 and L2 caches turned on, at least long enough to run a few benchmarks. Nothing was terribly spectacular. The memory scores were about the same as operation at 33MHz. Strangely the graphics score was identical to operation at 50MHz, and naturally CPU score was pretty low.

However, all was not lost. Although I started with an ISA graphics adapter (running in spec), I later switched to the ARK1000 VLB card. It seems that this card is perfectly comfortable at 66MHz, as I did not see any graphics corruption, and the IC was not hot at all. I also tested it with a VGA Sierra game. However, Grand Theft Auto 1 crashed the system. I highly suspect this was due to the CPU as the same thing happened with the ISA card installed.

Assuming I could actually get a chip that is stable at 66MHz, I think this ARK1000 card could prove to be quite useful.

How do you get a cyrix 5x86 to run in a VLB board? Is'nt yours like a 1992 vintage? I never thought you could run a cyrix 5x86 in a vlb system, your board does'nt have jumpers for the 5x86??

All you need is a voltage regulator to drop 5V down to 3V (see photos above). The jumpers for controlling the multiplier are part of the VRM. However, you can pretty much forget about using writeback cache mode (L1), and linear burst. I was however able to get a number of the other enhancements working.

I have successfully been able to resolve the problem I was previously having getting the Cyrix 5x86 chips into 3X mode. It involves a little trickery, but I think with a jumper cable and some software I am good to go. Basically I booted into 2X mode, switched the jumper to 3X, and then used CTCHIP to set the multiplier to 1X and back to 3X.

The results I am getting with speedsys are pretty impressive for a board of this vintage. I will post a screenshot soon.

Also, my IBM 5x86C arrived. It is stepping 0 rev 5. Seems to work fine at 120MHz so long as a fan is installed. I have not yet tested 2x66MHz, but it's in the pipeline.

I have made some further progress with the Cyrix 5x86. I ended up going back to my Cyrix Stepping 1 Revision 3 chip, as it seems to be stable at 120MHz with BTB enabled. This is preferable, as I would like BTB to work in Windows.

I also made a strange discovery with my memory settings. I was able to get some rather impressive memory speed results by disabling the L2 cache. At 120MHz (40x3) with all memory settings at their maximum speed (not possible with L2 enabled) I was able to get a transfer rate on the main RAM at 61mb/sec. If you look at my last benchmark, you will see that with the am5x86 L2 cache was only operating at 63mb/sec.

Unfortunately, at 50MHz bus I cannot use the fastest DRAM timings, and memory throughput drops to 50mb/sec. It's unfortunate that my board does not have a removable oscillator to adjust the bus speed. I would have liked to see what would happen with a 45MHz bus.

D'oh,

My previous setup with overclocked DRAM and disabled L2 cache has proven to be almost, but not quite stable. I have reverted back to my previous setup with L2 cache enabled.

Just for the hell of it, I decided to try enabling writeback L1 cache. Perhaps if I were running IDE I could get away with it, but writeback L1 and DMA SCSI controllers are a very bad combo. I don't believe a workaround was ever discovered.

I was comparing performance of Cyrix 5x86 and AMD 5x86 using Grand Theft Auto 1 for real world results. Despite the fact that the Cyrix 5x86 at 120MHz whoops am5x86 at 160MHz on speedsys results, in GTA1 the am5x86 at 120MHz was noticeably faster than the Cyrix at 120MHz with enhancements enabled.

Therefore, I will stick with the am5x86 at 160.

Final results with Cyrix 5x86 120. BTB_EN and FP_FAST

I'm actually already running with wide SCSI and 100mbps ethernet on the EISA bus. Unfortunately I need to get a faster drive. I just have some shitty old 2gb seagate barracuda at the moment.

I also already have 128mb installed. But for some reason only 64mb is reported by speedsys and dos. Perhaps I need an updated himem.sys?

Can you use the newer (and very quiet) 68-pin LW Seagate drives on your SCSI adapter? There are some cheap 36 GB 10K Seagate LW drives on eBay that can be relatively noiseless. It can be hit or miss depending on the on-time though.

If you use DOS 6.22's himem, does speedsys report 128 MB? What does Windows show for RAM (if/when you install Windows)?

I was comparing performance of Cyrix 5x86 and AMD 5x86 using Grand Theft Auto 1 for real world results. Despite the fact that the Cyrix 5x86 at 120MHz whoops am5x86 at 160MHz on speedsys results, in GTA1 the am5x86 at 120MHz was noticeably faster than the Cyrix at 120MHz with enhancements enabled.

Do you have a fps vs. fps comparison? Is GTA1 highly ALU intensive? If there's a download shareware for GTA1, I could test it on my CYRIX/PCI System to see what results, that is, if there is a numerical result to output from the test, i.e. in fps.

I'm actually not sure about LVD drives. I assume they might work though, since SCSI is usually pretty good about being backward compatible.

I actually don't have a copy of DOS 6.22. I'm a PC-DOS guy. I'll have to find a copy and give it a shot though. I have a feeling windows NT would probably detect all of the memory properly.

As far as I know, GTA1 and GTA2 were both released as free downloads to the public at some point. I'm pretty sure you can easily find a copy on the net if you look around. Unfortunately GTA1 does not have a built in FPS meter. Is there something else we could use to measure the results?

I figure GTA1 must be ALU intensive, because if it were heavily FPU dependant, it would likely run much better on the Cyrix 5x86 than on the AMD chip.

It seems like I have accidently plugged in my LVD drive into the non-LVD 68-pin connector on my PCI card and it worked. For your case, if it works, you'll love the results.

From Appendix 1 of the U4BC, there are two real-world tests which yield decimal FPS. I'd recommend Doom v1.9s timedemo3 and Quake v1.06 timedemo1. The Cyrix 5x86-120 and AMD X5-160 were a pretty close match, with the AMD leading by about 1 FPS in both tests. Unfortunately, a 1:2/3 FSB-to-PCI divider was in place for these tests, so expect the scores to increase in systems with 1:1.

In general (again from U4BC), the Cyrix 5x86-120 has a 12% FPU advantage over the X5-160, while the X5-160 has a 21% ALU advantage over the Cyrix 5x86-120. The Cyrix 5x86-133 drops the ALU gap to only 9%, while widening the FPU gap further to 20%.

Now that I think about it, removing the L2 cache, and running a Cyrix 5x86-120 at 133MHz (66 x 2) might be the way to go -- if it is stable.

Perhaps if I can boost the voltage on my VRM, I could have a shot at 2x66. My ARK1000 card doesn't seem to have any problems running at this speed.

Now if you can get a Cyrix 5x86-100 w/S1R3 running (stably) at 2x66Mhz w/ branch prediction, that would be fantastic.

From my previous posting,
Re: So you want a Cyrix 5x86-133?

Running the cpu at 4 V was all that was required to run a Cyrix 5x86-120 at 66x2. With more experimentation on my part, it may have been more Windows stable w/out the L2 cache. EDIT: Stable settings for this 5x86-120/133 have been obtained in Windows98SE/WinNT4.0/Win2000. Refer to the above link, page 2.

Anonymous Coward wrote:

-Hercules Dynamite Power VL w/2MB DRAM
I've tried every damn VLB card known to man, and I always come back to the ET4000W32P. It's damn fast in DOS, but it kind of stinks for Windows....which is why I have:

I've been thinking about video cards for my VLB 486 based system.

When you say the Et4000w32 is "damn fast in dos", is that compared to other vlb cards? In my experience with an ISA et4000 vs CL 5434 on a 386DX40 their was no difference is dos performance, not that I tried many games/benchmarks.

Now im looking at 2mb VLB cards I could choose a ET4000w32p or the cirrus logic 5434, the CL having bit 64 memory I would assume would make it a much better card? on my 386 system the CL-5434 is heaps quicker than the et4000 but thats no suprise it doesnt have windows acceleration like the w32p would.

feipoa wrote:

Now if you can get a Cyrix 5x86-100 w/S1R3 running (stably) at 2x66Mhz w/ branch prediction, that would be fantastic. […]
Show full quote

Now if you can get a Cyrix 5x86-100 w/S1R3 running (stably) at 2x66Mhz w/ branch prediction, that would be fantastic.

From my previous posting,
Re: So you want a Cyrix 5x86-133?

Running the cpu at 4 V was all that was required to run a Cyrix 5x86-120 at 66x2. With more experimentation on my part, it may have been more Windows stable w/out the L2 cache. EDIT: Stable settings for this 5x86-120/133 have been obtained in Windows98SE/WinNT4.0/Win2000. Refer to the above link, page 2.

Would a 5x86-120 do 66x2?

I just had a re-read of this post. I love the board and the case. I've always wanted an EISA 486 with VLB video but have enough retro computers at the moment. This hobby takes a lot of patience (7 years in your case!) That is a very attrative case; it will look good when oriented in both the horizontal and vertical positions and is nice and white with matching drives. A nice case is essential to any fine combination of hardware.

An intel dx4 at 150MHz would be pretty sweet though...

Has anyone tried this yet? That would have made for a great addition to the UMBC, or even at 133 MHz (2x66MHz). I think I'll add these to my "When I'm Bored" list of things to do.

It's unfortunate that my board does not have a removable oscillator to adjust the bus speed. I would have liked to see what would happen with a 45MHz bus.

A solder pump or rework station will make it removable. It should just be a 2-pronged oscillator chip at 14.X MHz or something like that. Then you can put in a socket in its place and have removable oscillators. Actually, this is a great idea in general for customising the bus speed; using discrete removable crystal oscillators to adjust the clock. I don't think this will make my "When I'm Bored" list though since I've been happy with 2x66 MHz operation lately.

So its seems that both UMC and SiS chipsets can be clocked to 66 MHz, even the ancient SiS chips. Does your northbridge get hot? If you place a temerpature element on the northbridge, how hot does it get? A simple test would be to place a SMD NTC thermistor (i.e. 10K) on top of the northbridge and read the resistance to determine temperature.

You mentioned that one of your AMD ADZ chips would not stably operate at 160 MHz. Was it of revision 0494 or 04F4? I find 04F4's to work much better for overclocking.

For your SpeedSys image with "Final results with Cyrix 5x86 120. BTB_EN and FP_FAST". Try setting LSSER = 0 and/or MEM_BYP = 1. LSSER set to 0 has just as much impact as BTB set at 1. Then try re-running GTA1 and SpeedSys. I have now finished my Cyrix 5x86 Special Feature Comparison and charts, but have't had the mood to do the write-up yet.

I'm actually not sure about LVD drives. I assume they might work though, since SCSI is usually pretty good about being backward compatible.

There should be a jumper on the harddrive to force it into Single-ended (SE) mode, at least the Seagate LVD drives have this jumper.

I figure GTA1 must be ALU intensive, because if it were heavily FPU dependant, it would likely run much better on the Cyrix 5x86 than on the AMD chip.

From my tests, I've noticed that DOOM seems to be more ALU intensive, while QUAKE seems much more FPU intensive. GTA1 could be more like DOOM.

Perhaps if I can boost the voltage on my VRM, I could have a shot at 2x66.

If you can, shoot for 3.85V.

Would a 5x86-120 do 66x2?

I'm not sure I understand the question. A Cyrix 5x86-120 should operate with a clock multiplier set at 2X and a FSB of 66 MHz. I find 3.85V to be the sweet spot for Vcore. Refer to the post, So you want a Cyrix 5x86-133 and The World's Fastest 486 for more details on this setup.
So you want a Cyrix 5x86-133?
The World's Fastest 486