Reply 20 of 38, by Anonymous Coward
VLB, PS/2, EISA, POD
Quite erotic actually.
How much L2 cache does it support? It might be time for an upgrade. Does it support write-back and write-through L2 modes? I've love to see a Cyrix 5x86-120 in there and perhaps some EISA SCSI Ultra.
A very nice find. Please post updates.
I'm not sure which chipset the enterprise board uses. Sometimes AMI used OPTi and sometimes SiS. I suspect this is an SiS board, because I think I looked it up one time. I had several opportunities to purchase a board like this, but I was concerned that although it could go up to 256mb it may only be able to cache 64mb due to only having 256kb cache. I do not believe it is possible to upgrade any further.
My MCCI/NICE VL/EISA board also uses SiS chipset. With 1mb of cache installed I believe it is possible to cache all 256mb of RAM with level 2 cache in write back mode. This is due to the board having 3 TAG RAMs, or rather 2 TAG RAMs and one "Dirty bit".
I'm not exactly sure what you'd do with 256mb of RAM on a 486, but at one point I had intended to see if I could get windows XP or 2000 running smoothly on it. I currently "only" have 128mb installed as my board has 16 30-pin SIMM slots, and thus far I haven't had any additional 16mb SIMMs fall into my lap for a reasonable price. Realistically 16mb is probably all you would ever need on a system like this.
One thing I would have really liked is having BIOS support for writeback L1 cache (not just L2). I believe there is a single board out there that can do it. The AIR 486EI. As far as I know, it is the only VL/EISA board that natively supports 3.3V 486 chips.
Believe it or not, there are actually a couple of VL/EISA (no PCI) Pentium motherboards in existence. Acer made at least one model that supported Dual pentium chips on a riser card. I can't remember who made the other one, but it may have been Everex or ALR. In my opinion, that would be something pretty special to have.
The lack of WB for L1 is a pretty big loss, probably 20%.
I'll know for sure pretty soon. I'm planning on running the Ultimate 486 Benchmark ensemble on a Cyrix 5x86 feature-by-feature to determine the impact of each. About a dozen features in total.
Just a quick update. I was at the Albuquerque Balloon Festival last weekend and this weekend I am at a Madrigal Competition with some of my students so I have not had much time, but the cyrix processor came so I thought that I would post a picture or two. (or three or four 😊)
The CPU is a gainberry CPU upgrade so the cpu comes on an adapter that has a voltage regulator on it.
I will try the CPU as soon as I have time, probably next week. The motherboard has an updated bios dated 1994. The bios currently on the board is dated 1992, so I will update the bios before I try the CPU.
I just ordered one of those CPUs as well. I am a bit concerned though, as the latest BIOS for my board is from 92, and it is not working properly with Cyrix chips I have already tried.
Have you tried a Cyrix DX4-100 CPU w/WB cache? How about an Intel DX2/DX4 w/WB cache? Did they work well? And did you confirm that WB was inabled?
For testing a Cyrix 5x86 CPU in motherboards which do not explicitly support them, perhaps you can manually turn off all the next generation enhancements in software, then turn them on one-by-one to determine which is giving you grief. It may very well be that you need to turn off the L1 write-back cache (thereby putting it into write-thru mode).
Most of the fancy features like burst write cycles and linear burst are turned OFF by default, but some others, which are ON by default, may be throwing things off. For example, after some long term testing (going thru feature by feature), I've noticed that some Cyrix 5x86 and motherboard combinations did not like DET_E enabled. The computer would freeze-up after about 24-hours of on-time. Once I turned this off, the computer never frooze again. DET_E is a feature usually ON by default.
You may also want to increase your CPU's I/O timing from 0 clocks to 128 clocks, or any multiple range between. This is controlled in CCR4, bits 0-2.
I'm not sure if a 1994 revision BIOS will help much by way of Cyrix 5x86 support since the specification for the 5x86 came out in 1995. But you may get lucky.
I just wanted to clarify my issues with the Cyrix chip. It is not to say that once I get the chip going that it isn't stable. I have not encountered any crashing or incompatibilities yet...
My main issue is that I cannot get the chip running when jumpered for a 2X or 3X multiplier. My system only boots when I set the multiplier as 2.5X, which puts the chip in 2X mode. Ideally I would want to use the 3X setting, but as of yet I have had no such luck. Using software to change multipliers doesn't look promising either.
I currently have the chip set in WT mode using a socket adapter. I am not 100% sure if my board supports L1 WB cache or not. I have reason to believe it may, but as I don't have the manual for it I can't be too sure. When WB mode is selected, everything seems to work fine except for the floppy drive (a common problem associated with WB L1 cache).
My board does not have a flash BIOS, but perhaps there is a later BIOS available. My board sold as late as 1994/1995.
"Using software to change multipliers doesn't look promising either."
What approach did you take here?
I have had success (in software) switching the Cyrix 5x86's core frequency via the PMR register using bits 0 & 1. Through these two bits, one can change the internal core frequency between 1x, 2x, 3x, and 4x.
Originally, I was hoping to change some 120's to 133/4x, but Cyrix disabled this for 80-120 MHz 3x parts, but it works for 100/4x and 133/4x parts. I have been able to switch from 3x to 1x to 2x.
The literature mentions that if you are running your Cyrix at 2x, you first need to switch down to 1x, then you can go up to 3x. Always drop to 1x before increasing. Did you try this approach?
I have read about this tool, but I have not yet tried it as I was discouraged from what I read on newsgroups. I was under the impression that you could drop from 2X or 3X down to 1X, but not the other way around. If you have actually done this however, then I would like to give it a shot.
Do you have a link to the tool?
I did a little test with my Cyrix 5x86-133 to verify the CLK register selection. I used chkcpu (16-bit) to verify the core frequency and Maxspeed as a secondary backup.
I was able to switch from 4x to 2x to 1x, then back to 2x, 1x and 4x, however, when I set it from 1x to 3x, the frequency actually only went to 2x according to chkcpu.
The wording of the cyrix literature is a little ambiguous, but it that seems after you boot-up and if you start out at 4x, for example, you can switch between 1x, 2x, and 4x. If you start out at 3x, you can switch between 1x and 3x. If you start out at 2x you can switch between 2x and 1x. If you start out at 1x you can transition to either 2x or 3x, but once you go up to 3x, you can only go back down to 1x, and if you decide initially to bump up to 2x, you cannot go to 3x until you reboot with 1x.
FYI, chkcpu tends to thinks the multiplier is still at 4x when I dropped the multiplier 2x and incorrectly reports, for example, 67 Mhz, 4x, 16.7 FSB.
For your case then, you'd want to start out at 1x, and once booted, set the register to 3x. The problem you may run into is that your BIOS/motherboard may only have a 1-bit CLKMUL, meaning that if you set it to 1x, the Cyrix will want to run at 3x, and if set to 2x, the Cyrix will run at 2x.
Have you been able to boot with a 1x setting at 33 MHz?
Any idea on how to do a BIOS mod? This would be a pretty good mod for everyone with a Cyrix 5x86, because if you could start in 1x mode, then you would be free to choose any multiplier your heart desires. There is a guy who reprogrammed a bunch of pentium BIOSes to support K6+ chips, so this would seem to be along the same lines.
Unfortunately, I haven't yet gotten involved with BIOS mods. There are BIOS kits out there though. If you have some extra time, you could read up on it. Unfortunately, I've been too busy with other non-motherboard related electrical design work to invest in a BIOS kit.
Jans did modify my socket 7 BIOS to support the K6-III+. It worked great, but that seems to be his little niche. I'm not sure if it would be asking too much for him to modify a 486 BIOS.
I was thinking, if your motherboard doesn't boot with the Cyrix on 3x mode, then maybe even if you were able to boot on 1x mode, and switched it to 3X mode, it may still hang. It seems really odd that everything works fine on 2x mode w/WT L1 but not on 3x mode.
Are you using the Gainbery CPU adapter? Is your motherboard CPU voltage jumpered to 5 V? What voltage output do you measure from the Gainbery's voltage regulator?
Have you tried using software to switch the Cyrix to WB mode after you've booted? Maybe you're floppy will still work in this situation.
I'm not using a gainberry adapter. I have 3 different VRMs here though. My best one has jumpers for multiplier (2X, 2.5X, 3X), AMD 2X mode, and voltage 3.3V or 3.45V. I use this in conjunction with an interposer to force the CPU into WT mode. As it turns out, the Cyrix 5x86 CPU does not need this bracket, as it automatically configures itself to WT in my board. Switching to WB mode with software causes a lockup.
As I stated before, the system does not boot up when the CPU is jumpered for 2X mode. It only boots when the 2.5X jumper is set (which then gives me 2X mode). It boggles the mind.
After reading the Cyrix 5x86 BIOS guide, I am doubtful that a BIOS mod for 1X mode would help at all. It sounds like if you attempt to switch to a higher multiplier, the CPU will verify the status of the CLKMUL pin before taking action.
What I would really like to know is what the unmarked bits in the PMR register do.