Date: Thu, 18 May 1995 13:34:55 -0700 (PDT) From: "Rodney W. Grimes" <rgrimes@gndrsh.aac.dev.com> To: james@miller.cs.uwm.edu (Jim Lowe) Cc: FreeBSD-hackers@FreeBSD.org (FreeBSD hackers) Subject: Re: Help! Message-ID: <199505182034.NAA15099@gndrsh.aac.dev.com> In-Reply-To: <199505181958.OAA24762@miller.cs.uwm.edu> from "Jim Lowe" at May 18, 95 02:58:04 pm
next in thread | previous in thread | raw e-mail | index | archive | help
[Moved to hackers list, please don't double post!!!] > > > related but I have no idea. I have now disabled the external cache and I > > > am trying the make world again... It is a lot slower... > > > > If this is not the ASUS board, ignore me!! > > > > I have qualified and done extensive testing on the ASUS Triton board, it > > does work. To this day I have probably completed over 100 make worlds > > on these boards. I see you are running a 100Mhz CPU chip, are you also > > running 60nS DRAM as required by the motherboard book. Have you left > > the BIOS settings on the default values, or have you tried to speed > > things up by tweaking them?? > > I am running 32 Meg of 70ns EDO DRAM (this was all that was available a > month ago) with 12ns async cache with a SuperMicro mother board and AMI bios. > I havn't tweaked much of anything. I think the problem might be the external > cache speed in my case. My system has been running for almost 2.5 hrs now with > the external cache disabled and make world is just about finished. Unless SuperMicro has pulled some time out of the memory setup in the chip set you *MUST* run 60nS Simms, I don't care if they are EDO or not, I was able to get 60nS EDO a month ago, but not in 16MB parts. I have rumor that Samsung is now making EDO in 50, 60 and 70nS parts. I have never been very impressed with SuperMicro boards, I own one of them and wish I didn't. This is the older P5-60/VLB/ISA board. Using it as a guide the book clearly says you must run 60nS simms with a 66Mhz CPU. I have found that almost *every* motherboard running an external clock frequency of 66Mhz require 60nS DRAMS. The clone makers might get away with stuffing 70nS in there and running DOS/Windows, but it will *not* work for unix which is known to use memory much more agressivly than DOS. Disabling the external cache has disabled memory read burst cycles most likely as they are usually caused by the external cache requesting a 32 byte line fill, thus you are now not using main memory near as hard. Or it could be a bad or slow cache ram causing you fits. > I talked with some people down in the Electronics shop and they suggested > that the cache speed must be 10ns with a 100 Mhz processor and 8ns with > a 120Mhz processor (1/Processor Mhz). Do you know if this is correct? It is not correct, they have not accounted for the fact that the external bus clock is different than the internal CPU speed. The 90MHz and 120MHz parts uses a 60MHz bus, the 100Mhz and 133Mhz part use a 66Mhz external bus. They have completely failed to take into account any bus delays, loading delay, gate delay external to the CPU. In general you need 15nS cache chips for 66Mhz externally clocked systems, but can cut 1 cycle out of the burst by going to 12nS or faster parts (1/66Mhz is 15nS, thus with 15nS parts you need to run at a 3-2-2-2 cycle typically, but going to 12nS parts of the pipeline burst type you cut this to 2-1-1-1). > > Also, do you know if there is anyway of causing a core dump if the system > hangs? Ah, compile in the kernel debugger ``ddb'', see LINT file in conf directory. Then ctrl-alt-esc should put you in the debugger, ``call diediedie'' should cause a panic of ``Becuase you told me to'', and do a crash dump. -- Rod Grimes rgrimes@gndrsh.aac.dev.com Accurate Automation Company Custom computers for FreeBSD
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199505182034.NAA15099>