I've read somewhere the IBM chips might be rated more conservatively, and thus are supposedly more overclockable. Nice overclock there feipoa 😉

Jesus Christ. Are your cache chips stable at 66MHz?

I was thinking to do something like this several years ago with a UMC PCI board. I heard some of them have 66MHz FSB and PCI 1/2 divider. You should see if you can find a board with a 1/2 divider so that your graphics adapter isn't running so badly out of spec.

feipoa wrote:

Skipped frames?

quake doesn't skip in a timedemo

With the increasing number of people obtaining this 120 MHz Cyrix gainbery processor, maybe somebody will have luck with 133 MHz in Windows.

I ordered one of those IBM 5x86C chips too. I should have in within a week. My VL/EISA motherboard actually supports a 66MHz bus, but I haven't been able to get it going with any of my CPUs yet. I don't know how well my VLB Hercules card would like that though. It works well at 50MHz. Perhaps with a wait state I can manage something.

I can confim, it is a Cyrix clone, Texas, IT's, and IBM, all made them, because cyrix did not have its own waffer plant, and made loose contracts with hes supliers. So they "stole" the schematics 😁

I've heard that IBM was a true believer in the 5x86, and the driving force behind the production. Apparently they were the ones that proposed a 32-bit version of 6x86 to Cyrix. Perhaps that would explain why IBM continued to produce the 5x86c long after everyone else had given up.

You've indirectly raised an interesting question in my mind... does an IBM-branded 5x86-133 exist, even as an engineering sample?

I suppose it's possible something like that exists. But it seems that IBM was the first to realise that design wasn't very stable beyond 100MHz. As they didn't even really produce 120MHz parts (okay, a few of them are out there), I have my doubts they would have bothered pursuing 133MHz.

What I'd really like to get my hands on is a very late model 5x86C. I believe they made these chips until at least early 1997. (apparently IBM produced the 5x86 until it's contract expired in 1998)
I remember reading somewhere (it may have been CPU-world) that there is actually a core later than stepping 1 rev 3. I have so far failed to dig up the post that mentioned this, but I think the key point was that only IBM produced it.

I'd certainly settle for a phantom revision if an IBM 133 MHz part was an impossibility. A 120 MHz Cyrix 5x86 with fully-functioning branch prediction; w00talicious!

Has everybody checked their IBM Cyrix's? Both of my IBM 120's are Stepping 0, Revision 5.

One of mine has

IBM9314 P40049
50H6281 PQ

Not game to remove the other cpu from the voltage regulator/adapter.

As far as I know, you need to read the stepping/revision from the ID register of the chip, DIR1. An easy way to check it is with ChkCpu16 in DOS.

For example, ChkCpu16 returns,
DIR1: 05h

The 'h' just means the number is a hexidecimal integer. Convert the '05' to binary, 8 bits. The first 4 bits are the revision, the second 4 bits are the stepping. Convert those two 4 bit segments to decimal. So for this example, the stepping is decimal 0 (0000 in binary) and the revision is decimal 5 (0101 in binary).

I'm not sure if there is a method to determine the stepping/revision from the CPU's bottom printed markings.

Wait, you actually have 5x86C 120s?

I found a way to get my 5x86 into 3x mode. After booting up at 2x, I set the CPU into 1X mode, then change the CLKMUL pin to 3X, and then set the cpu into 3X mode using ctchip.

I plan to rig up a jumper cable to my vrm adapter to make this practical.

How is your progress on the CLKMUL jumper switches?

Do you actually have your Cyrix working at 120 MHz on an EISA board as confirmed by benchmark programs and other utilities?

When you say that you set the CPU in 1X mode, I assume this is also via ctchip? I have never heard of ctchip. What is it, and how does it differ from the Peter Moss and other similar Cyrix 5x86 register mod utilities? I'm figuring it is some kind of software-based clock changing utility, and not the Connecticut Child Identification Program.

I do not have 5x86C's.

You don't have 5x86Cs? You posted a photo of it at the top of this thread!

CTCHIP is a pretty interesting program. It isn't just for 5x86 chips. It supports various types of CPUs, and motherboard chipsets. I guess for the most part it does the same thing as the Peter Moss utility, but you should check it out and make sure you aren't missing anything.

Using CTCHIP I was able to get my 5x86 chip into 3X mode. I booted the system at 2X, changed the VRM jumper over to the 3X setting, and then changed the registers using CTCHIP.

I benchmarked it using speedsys 4.78. It outperformed my am5x86 at 160MHz by 10% or so, but I have found that in actual use it feels slower.

You don't have 5x86Cs? You posted a photo of it at the top of this thread!

Well, apparently I didn't look at my own photo very carefully. I guess I have two 5x86C's. I figured all the Gainbery/IBM 5x86-120's were the same, 5x86C. Are they not all with the C on them?

I benchmarked it using speedsys 4.78. It outperformed my am5x86 at 160MHz by 10% or so, but I have found that in actual use it feels slower.

When you say "actual use", are you refering to DOS games, or Windows-ing around? Turn on the next generation enhancements of the 5x86-120 and see if it still feels slower than an X5-160. If you turn on RSTK_EN and LOOP_EN you should get a boost in ALU performance. If you turn on FP_FAST, you'll get a big jump in FPU performance. Unfortunately, I still haven't had time to run a feature by feature performance comparison for this 5x86-120. A 5x86-120 with enhancements on should "feel" faster than an X5-160. It does for me in NT 4.0, but my mainboard fully supports these added features.

For the most part:

5x86 = Cyrix
5x86C = IBM
st5x86 = SGS Thomspon

They are mostly similar, except that the IBM chips had better quality control, and the ST chips normally had 4X multiplier.

I suppose that the gainberry units may only use the IBM chips. But I am not using a Gainberry kit. I have two Cyrix 5x86s (one 0.5, and one 1.3) and one IBM 5x86C (0.5). I wanted to know if either of your IBM 5x86C chips were stepping 1 rev 3s.

I have been testing real world results using PC DOS 2000. I primarily rely on Grand Theft Auto 1, as it really taxes the system, especially in 24-bit colour mode. The only two enhancements I currently have on are Branch Prediction and FAST_FP. Were you saying in another post that Branch Prediction isn't completely stable with other enhancements enabled like LOOP_EN and RSTK)EN?

My ST 5x86-100's don't have a 4X multiplier option; they were manufactured in week 2, 1996. They are also S0,R5.

I haven't seen any 120/133 MHz Cyrix 5x86 with S1,R3. Does anyone out there have one? The only S1,R3's I have are Cyrix 5x86-80 (Week 35-39 1995) and Cyrix 5x86-100 (Week 44, 1995). In fact, even my Cyrix 5x86-100/4X from Week 7 of 1996 is S0,R5.

For a S1,R3 Cyrix 5x86-100, Windows 98SE and Windows NT 4.0 were only long-term stable with BTB_EN if, and only if,

BWRT = 0
and
LOOP_EN = 0
and
RSTK_EN = 0

Note that S1,R3 does not seem to support BWRT in any configuration.

For a S0,R5 Cyrix 5x86-120/133, Win98SE was unstable with BTB_EN even if LOOP_EN and RSTK_EN = 0. The BWRT setting seemed to have no effect on stability with S0,R5.

For a S0,R5 Cyrix 5x86-120/133, WinNT 4.0 was pseudostable with BTB_EN ON if, and only if,

LOOP_EN = 0
and
RSTK_EN = 0.

The BWRT setting seemed to have no effect on stability with S0,R5.

At any rate, these were the results I jotted down. Your mileage may vary. Stability in DOS exceeds that of Windows.

From the U4BC's pdf, refer to RAW column D vs. column E as this compares the effects of BTB_EN vs. LOOP_EN/RSTK_EN. BTB_EN has about a 4% incease in performance over a CPU with both LOOP_EN and RSTK_EN enabled. Perhaps if you run a S1,R3 Cyrix 5x86-100 at 120 MHz, this would equate to the fastest possible Cyrix 5x86-120. I personally haven't tested my S1,R3 Cyrix 5x86-100 at 120 MHz, but if I did, I'd probably want to run it at 3.6V or 3.7V, and not the stock 3.45V.