dyer@harvard.UUCP (Steve Dyer) (01/13/86)
I would like to point out a problem I discovered with using a faster crystal in my PC/AT when I installed SCO XENIX V. I had been running quite successfully for months under DOS with an 18.4 mhz crystal, as opposed to the standard IBM 12 mhz crystal. During the XENIX installation process, and occasionally and unpredictably while running XENIX subsequently, processes would get SIGSEGV (seg violation or memory fault) sent to them. There was no way to predict which processes would be affected or when. Dropping back to 12 mhz stopped the problem. I would like to solicit opinions from knowledgable parties on the cause of this phenomenon and its solution. I suspect that the memory management hardware was stressed lightly, if at all, under DOS, and the higher clock rate can produce glitches in the 6 mhz 286's on-chip memory management hardware. I am thinking about installing a new 286 chip which is rated at 10 mhz (20 mhz xtal) and trying the experiment again. Any other suggestions? Anyone else see this problem? -- /Steve Dyer dyer@harvard.harvard.edu harvard!dyer
abm@ptsfa.UUCP (Al Margolis) (01/15/86)
In article <604@harvard.UUCP>, dyer@harvard.UUCP (Steve Dyer) writes: > ... a problem I discovered with using a faster crystal in my PC/AT > ... I suspect that the memory management hardware was stressed > lightly, if at all, under DOS, and the higher clock rate can produce > glitches in the 6 mhz 286's on-chip memory management hardware. I installed a speedup kit on my PC/XT which ran the system at 7.14Mhz, and the system would not run POST diognostics without indicating memory errors. The solution (arrived at by guessing, not intelligence) was replacing the AMD controler chip with an NEC part. Since one of the DMA channels is used for DRAM refresh, this fix seems reasonable and has been working for several months. I have heard opinions of both (1) the AMD chips are a common source of trouble and (2) I was just lucky with the individual replacementent chip since this is a marginal situation. << preaching >> This is not directly applicable to the AT situation due to the different chip sets, but I think that you can expect problems in several areas when you push a PC to 50% or more above design speed. On the other hand, if you have the time & the nerve to put a soldering iron to your $x000 machine, you can get a lot more out of it -- I'd never go back to 4.77Mhz. << question >> In working through this and a few related problems, I picked up a lot of folk lore concerning the relative quality of various chips (i.e. Intel 8237s are good, AMD's are bad; Fujitsu 41256-15s are better than OKIs; ...) This folk lore did not discriminate between spec and pushed environments. Are there real [significant] differences? I am a bit surprised, especially if true about licensed second source suppliers. I would like to see responses [not flames] from both vendors and users. Please help be out of my ignorance. Al Margolis Pacific Bell {ihnp4, dual}!ptsfa!abm These comments are only my own, and the machine in question is my own PC. I'm not affiliated with anyone who makes semiconductors.
caf@omen.UUCP (Chuck Forsberg WA7KGX) (01/15/86)
I'm running SCO SYS V Xenix on a PC-AT with an 18.1 mHz rock with no apparent problems. I tried a 20 mHz rock which worked fine for DOS but generated various problems for Xenix. So the 20 mHz rock went to the junk box until I come acorss a cheap 20 mHz 286. I will also be looking into rumors that the NEC V60/V70 blows away the 286. -- Chuck Forsberg WA7KGX ...!tektronix!reed!omen!caf CIS:70715,131 Author of Professional-YAM communications Tools for PCDOS and Unix Omen Technology Inc 17505-V NW Sauvie Island Road Portland OR 97231 Voice: 503-621-3406 TeleGodzilla: 621-3746 300/1200 L.sys entry for omen: omen Any ACU 1200 1-503-621-3746 se:--se: link ord: Giznoid in:--in: uucp
doug@terak.UUCP (Doug Pardee) (01/17/86)
> In working through this and a few related problems, I picked up a lot of > folk lore concerning the relative quality of various chips (i.e. Intel > 8237s are good, AMD's are bad; Fujitsu 41256-15s are better than OKIs; ...) > This folk lore did not discriminate between spec and pushed environments. > Are there real [significant] differences? I am a bit surprised, especially > if true about licensed second source suppliers. Yes indeed. Sometimes the differences are "documented" in the spec sheets. I've encountered this with PALs (tm) -- one manufacturer specs theirs as somewhat slower than MMI's "standard". I've seen it with one manufacturer's FAST (tm) logic: one part had much lower input loading than Fairchild's "standard" part, making the input lines susceptible to noise on a board where Fairchild's part worked fine. Sometimes there are undocumented differences. I've seen a problem with a 10K ECL part obtained from a second source which didn't occur with the Motorola MECL part. The second sourced part had problems during initial power-up, and wasn't reliable until each input changed state once. Sometimes it's simply an ability to produce a faster part; AMD is a second-source for the iAPX86 CPU's, but they seem to produce a given speed part a few months before Intel does. And DRAMs! I'm beginning to think that no two DRAM chips are inter- changeable, even from the same manufacturer 1/2 :-) Aw, the spec sheets are nearly identical, but their behavior is something else. [Vision of Wendy's TV ad: "What *kind* of parts?" "Parts is parts."] -- Doug Pardee -- CalComp -- {hardy,savax,seismo,decvax,ihnp4}!terak!doug
glen@intelca.UUCP (Glen Shires) (01/17/86)
> I'm running SCO SYS V Xenix on a PC-AT with an 18.1 mHz rock with no > apparent problems. I tried a 20 mHz rock which worked fine for DOS > but generated various problems for Xenix. So the 20 mHz rock went > to the junk box until I come acorss a cheap 20 mHz 286. I will also > be looking into rumors that the NEC V60/V70 blows away the 286. > Just a clarification: The 286 clock rate is half the crystal, so a 20MHz crystal uses a 10MHz 286. 286's are rated at 6, 8, 10, 12.5 MHz. -- ^ ^ Glen Shires, Intel, Santa Clara, Ca. O O Usenet: {ucbvax!amd,pur-ee,hplabs}!intelca!glen > ARPA: "amd!intelca!glen"@BERKELEY \-/ --- stay mellow
phil@amdcad.UUCP (Phil Ngai) (01/21/86)
In article <604@harvard.UUCP> dyer@harvard.UUCP (Steve Dyer) writes: >I would like to point out a problem I discovered with using a faster >crystal in my PC/AT when I installed SCO XENIX V. >Dropping back to 12 mhz stopped the problem. Does this mean a computer clocked faster than it was designed for won't always work as designed? It must be an IBM conspiracy intended to force people to buy faster and more expensive computers. >glitches in the 6 mhz 286's on-chip memory management hardware. I am thinking >about installing a new 286 chip which is rated at 10 mhz (20 mhz xtal) >and trying the experiment again. There are other components in the PC/AT besides the crystal which hobbyists seem to like to change so much and the 286. For example, there's the RAM, the EPROM, the DMA controllers, the time of day clock, and so on, all of whose timing was presumably carefully thought out by the engineer to work even in worst case conditions. Now you turn the clock up. I wonder if I took my favorite program and removed some lines of code from it, would it run faster? It should, it has less to do. excuse the heavy sarcasm, I had to finally let it out... -- (C) 1986 Joe Random is not a valid copyright. Copyright 1986 Joe Random is. Phil Ngai +1 408 749 5720 UUCP: {ucbvax,decwrl,ihnp4,allegra}!amdcad!phil ARPA: amdcad!phil@decwrl.dec.com
phil@amdcad.UUCP (Phil Ngai) (01/21/86)
In article <187@intelca.UUCP> glen@intelca.UUCP (Glen Shires) writes: > The 286 clock rate is half the crystal, so a 20MHz crystal uses a 10MHz 286. > 286's are rated at 6, 8, 10, 12.5 MHz. There's also a chip in the 286 family called the 82284 which is the actual oscillator device. It too comes in speeds and is therefore another possible reason why a 20 MHz crystal doesn't work in a PC/AT. Note: I don't even know if a PC/AT uses the 82284, although I would be surprised if it didn't, and I don't recommend changing the hardware around if you want it to keep working. -- (C) 1986 Joe Random is not a valid copyright. Copyright 1986 Joe Random is. Phil Ngai +1 408 749 5720 UUCP: {ucbvax,decwrl,ihnp4,allegra}!amdcad!phil ARPA: amdcad!phil@decwrl.dec.com
phil@amdcad.UUCP (Phil Ngai) (01/21/86)
In article <996@terak.UUCP> doug@terak.UUCP (Doug Pardee) writes: >manufacturer's FAST (tm) logic: one part had much lower input loading >than Fairchild's "standard" part, making the input lines susceptible to >noise on a board where Fairchild's part worked fine. Lower input loading sounds like an advantage to me. If it causes problems, is it possible the device was not used correctly? Remember most input loadings are specified as maximums and even the original manufacturer may improve the part and reduce the loading. >Sometimes there are undocumented differences. I've seen a problem with >a 10K ECL part obtained from a second source which didn't occur with the >Motorola MECL part. The second sourced part had problems during initial >power-up, and wasn't reliable until each input changed state once. Yes, parts are not parts. Some second sources are complete or partial redesigns. Differences in behavior, especially in unspecified modes of operation are not uncommon. Some second sources are true second sources where masks are exchanged. There is much less room for differences in the latter case, probably no more chance than there is for a difference between lots from the same manufacturer. >Sometimes it's simply an ability to produce a faster part; AMD is a >second-source for the iAPX86 CPU's, but they seem to produce a given >speed part a few months before Intel does. Actually, we tend to be VERY conservative in our testing. But we have some very hot processes too. >And DRAMs! I'm beginning to think that no two DRAM chips are inter- >changeable, even from the same manufacturer 1/2 :-) Aw, the spec >sheets are nearly identical, but their behavior is something else. I can't say I've had much trouble. What exactly are you referring to? There are some things you have to get right, and a lot of parameters you have to satisfy, but once you've done so DRAMs seem to work quite reliably for me. DRAMs are getting horribly complicated these days. Page mode, nibble mode, static column addressing, read-modify-write cycles, self-refreshing, and hardly anyone implements the same set of features or in the same way. What a mess. But I have personally used DRAMs in a half a dozen circuits with no trouble. In production, too. All boards were tested over extreme temperature ranges with varying manufacturers. -- (C) 1986 Joe Random is not a valid copyright. Copyright 1986 Joe Random is. Phil Ngai +1 408 749 5720 UUCP: {ucbvax,decwrl,ihnp4,allegra}!amdcad!phil ARPA: amdcad!phil@decwrl.dec.com
john@hp-pcd.UUCP (john) (01/22/86)
<<<< < Does this mean a computer clocked faster than it was designed for < won't always work as designed? It must be an IBM conspiracy intended < to force people to buy faster and more expensive computers. < But what frequency was it designed for? The tech ref shows a 12 Mhz crystal but some of the test points are marked -16Mhz. Its like the design was for a faster part but was slowed down later for some unknown reason. The "Speed up your AT" movement reminds me of all those backyard mechanics that have been squeezing more performance out the family auto than Detroit ever thought they were putting in. John Eaton !hplabs!hp-pcd!john
doug@terak.UUCP (01/22/86)
>> manufacturer's FAST (tm) logic: one part had much lower input loading >> than Fairchild's "standard" part, making the input lines susceptible to >> noise on a board where Fairchild's part worked fine. > > Lower input loading sounds like an advantage to me. If it causes > problems, is it possible the device was not used correctly? Remember > most input loadings are specified as maximums and even the original > manufacturer may improve the part and reduce the loading. True, the specs don't give a minimum loading figure, but I don't know anybody who terminates every etch on a board in a low impedance just in case some chip manufacturer should decide to reduce the input loading to a few microamps. FAST family parts are, well, fast. They tend to be used in applications with very high frequency signals. FAST also generates some pretty sharp edges. It's nearly impossible to keep signals from cross-coupling on the board; you have to rely on the low impedances to swamp them out. > >And DRAMs! I'm beginning to think that no two DRAM chips are inter- > >changeable, even from the same manufacturer 1/2 :-) Aw, the spec > >sheets are nearly identical, but their behavior is something else. > > I can't say I've had much trouble. What exactly are you referring to? I should've been more explicit. Make that 256K DRAMs. In simple applications with short traces, few chips and non-critical timing, they probably aren't a problem. But when you try to put a few hundred together to make a large memory subsystem (this necessarily results in lengthy traces) and then try to run them close to rated speed, good luck. Some chips work, others don't. Even though the specs are met, some just seem more sensitive to things like noise and edge rates. -- Doug Pardee -- CalComp -- {hardy,savax,seismo,decvax,ihnp4}!terak!doug
gpw@ihlpf.UUCP (Wilkin) (01/27/86)
Subject: Re: Fast AT xtals? How about fast XT xtals? In-Reply-To: your article <11541@ucbvax.berkeley.edu.BERKELEY.EDU> Look at your video board. What speed does it expect! Most boards clock off the mother board clock, soooo maybe adding a clock to the video board (color only I think) will help. -- George Wilkin AT&T Network Systems usenet ihnp4!ihlpf!gpw 312-979-6593 work