uzun@pnet01.cts.com (Roger Uzun) (11/17/89)
I switched the j500 jumper on the A2620 board of my A2500 to use 80ns DRAM timing. I only have 90 and 100ns DRAMS on the board. So far no problems in operation and the machine is faster. Will I be relatively safe running the DRAMS at 80ns? If it was going to fail would it fail right away? -Roger UUCP: {hplabs!hp-sdd ucsd nosc}!crash!pnet01!uzun ARPA: crash!pnet01!uzun@nosc.mil INET: uzun@pnet01.cts.com
cmcmanis%pepper@Sun.COM (Chuck McManis) (11/18/89)
In article <720@crash.cts.com> uzun@pnet01.cts.com (Roger Uzun) writes: >I switched the j500 jumper on the A2620 board of my A2500 to >use 80ns DRAM timing. I only have 90 and 100ns DRAMS on the board. >So far no problems in operation and the machine is faster. >Will I be relatively safe running the DRAMS at 80ns? If it was >going to fail would it fail right away? >-Roger Roger, I appreciate your desire to get something for nothing, however you have just set yourself up for a big problem. You see, most chips work well above their rating at STV (Standard Temperature and Voltage). However, there are these things that people in the chip industry call the "corners". These are the bounds for the conditions underwhich the chip must operate. For commercial chips that is usually something like 10% variation in voltage and temperature variations between 0 and 70 degrees C. The reason temp. and voltage are variables is because they have a dramatic effect on the physical and electrical properties of silicon. At Intel the first '386 chips could run at 33Mhz, but only at Room Temperature. Over the full temperature range they could only work reliably at 16Mhz. And chips are like resistors, no two are exactly alike. From the same fab run you may get parts that work at 18Mhz and parts that run at 16.5Mhz, they all qualify as 16Mhz parts but none of them qualify as 20Mhz parts. They all run best at room temperature with 5.0000V power supplies. So in your situation, while the system is at room temp you find everything is working fine. Then one day as you begin debugging an application you find it starts dying in some routine over and over. You begin to track it down but find it never can be isolated to a particular routine or it works in the debugger but not when you run it standalone. All it takes is one memory chip that is now operating at it's rated speed but slower than the design speed of the system. You can spend hours or days or months tracking down bugs like this. Only to find that when the system is jumpered correctly, everything works as expected. It really isn't worth it, especially for someone developing code. It is a lot less expensive than you might imagine to sell your current chips and buy rated chips to replace them with. --Chuck McManis uucp: {anywhere}!sun!cmcmanis BIX: cmcmanis ARPAnet: cmcmanis@Eng.Sun.COM These opinions are my own and no one elses, but you knew that didn't you. "If it didn't have bones in it, it wouldn't be crunchy now would it?!"
uzun@pnet01.cts.com (Roger Uzun) (11/19/89)
Well I will probably be jumpering it back soon just, since like you mentioned it could cause a lot of headaches someday in the summer. For now I have ruin applications/mem diagnostics for 2+ days straight on the machine with no problems, speedup is in the 10-15% range with most applications, slightly less with fp intensive stuff. I really did it just to test things out, see what would happened, I have good reason to believe 80ns DRAMS are not needed, only 90ns parts are really needed. 2M of the RAM on the board is 90 ns 2M more is 100ns. In any case I do not like to desolder that much RAM and I would never do it for 1 10-15% increase in performance. For those more daring types you may want to try it at your own risk. Seems to be working for me. -Roger UUCP: {hplabs!hp-sdd ucsd nosc}!crash!pnet01!uzun ARPA: crash!pnet01!uzun@nosc.mil INET: uzun@pnet01.cts.com