from http://amethyst.ie.akita-u.ac.jp/~matsuo/axppci+linux.html
and my translation follows.
Noname machine has possibility of hacking upto ~30% up.
if you would like to try becareful with
1. board, linux system may receive unfixable damage.
2. even you do whatever, you can no way beat "cabriolet" or such with
128bit bus, with 21064A
system.
do with your own risk. before you do it, make sure board is cool enough,
also make sure that you have enough cooling system to cpu(it get really hot
that you can't touch).
He puts fan in CPU, and to case, so that they take all the hot air from
case.(double system).
"Noname"$B$OB?>/<j$r2C$($i$l$k$H$3$m$,$"$j!$Ls(B 30%$B%9%T!<%I%"%C%W$G$-$^$9!%;n$7$F(B
$B$_$?$$J}$O!$(B(1)$B%\!<%I!$(Blinux$B%7%9%F%`$KCWL?E*$J%@%a!<%8$rM?$($k$+$b$7$l$J$$!$(B
(2)$B$I(B
$B$s$J$K2~B$$7$F$b(B "Cabriolet"$B$N$h$&$J(B 128bit$B%P%9$r;H$C$?(B 21064A$B%7%9%F%`$K$O$+(B
$B$J$o$J$$(B
$B!$$3$H$r>5CN$N>e$G!$$4<+J,$N@UG$$G!$$d$C$F$_$F$/$@$5$$!%%\!<%I$K<j$r2C$($kA0$K(B
$B!$(BCPU$B$NNd5Q$rK|A4$K$7$F$*$$$F$/$@$5$$!%8=>u$G!$?($l$J$$$0$i$$$KG.$/$J$C$F$$$k(B
$B>l9g!$%9%T!<%I%"%C%W$9$k$H(BCPU$B$,K=Av$9$k2DG=@-$,Bg$-$$$G$9!J;d$N%7%9%F%`$N>l9g(B
$B!$(BCPU$B$K%U%!%s$rIU$1$F6/@)6uNd$7!$6)BN$KGS5$%U%!%s$rDI2C$7$F!$G.$,$3$b$i$J$$$h(B
$B$&$K$7$F$$$k!K!%(B
If you clockup, FPU's result changes(will not work correctly).
in order to fix this priblem, you must clock down a bit(only way)
after modification, check with check_cpuu at ftp.kuamp.kyoto-u.ac.jp
$B%/%m%C%/<~GH?t$r>e$2$k$H!"IbF0>.?tE@$N1i;;7k2L$,68$C$F$/$k%H%i%V%k$,5/$-$k$3$H(B
$B$,$"$j$^$9!%$3$N>l9g!"%/%m%C%/<~GH?t$r2<$2$k$7$+3N<B$K%H%i%V%k$r2sHr$9$kJ}K!$,(B
$B$"$j$^$;$s!%2~B$8e!"(B ftp.kuamp.kyoto-u.ac.jp $B$K$"$k!"(Bcheck_cpu$B$G3NG'$7$F2<$5$$!%(B
Clock up to 284MHz!
AlphaAXPpci33 233mhz has 21066a, this chip is pcibus clock
33.3333mhzx7times, you can change this
jumper to 8 or 9(change mhz to 266, 300)
However, from cpu's pll's vco limit about 284mhz(guess from BogoMIPS) is
top limit.
(each vco limit may change by CPU)
from BogoMIPS, 230.6 to 263.7, 281.0
284mhz is about 20% faster, and so far running at 284mhz has no problem(as
far as any big ones).
21066a's core is 21064a(275mhz), so 284mhz would be no problem.
to find place of jumper, setup, check DEC's ftpsite at
ftp.digital.com go to design guide's cpu_ref.txt
AlphaAXPpci33 233MHz$B%P!<%8%g%s$O!$(B21066A$B$r@Q$s$G$$$^$9!%$3$N%A%C%W$O!$(BPCI$B%P%9(B
$B$N%/%m%C%/(B 33.3333MHz$B$N(B7$BG\$G!$F0$$$F$$$^$9!%$3$N!V(B7$B!W$H$$$&CM$O%8%c%s%Q$K$h$C$F(B
$B@_Dj$5$l$F$$$F!$!V(B8$B!W$^$?$O!V(B9$B!W$KJQ99$G$-$^$9!%(B CPU$B$N%/%m%C%/<~GH?t$O$=$l$>$l(B
$B!$(B266MHz$B!$(B300MHz$B$K$J$j$^$9!%<B:]$K$O!"(BCPU$BFb$N(B PLL$B$N(B VCO$B$N8B3&$+$i!"Ls(B
284MHz ( BogoMIPS$B$+$i?dDj(B)$B$,>e8B$K$J$j$^$9!J(BVCO$B$N8B3&$K$O!$8DBN:9$,$"$C$F(B
$B!$(BCPU$B$K$h$C$FJQ$o$k$+$b$7$l$J$$!K!%(B BogoMIPS$B$O!$(B230.6$B$+$i(B 263.7$B!$(B281.0$B$K$"$,$j(B
$B$^$9!%(B 284MHz$B$G$O(B233MHz$B$N$H$-$N!$Ls(B20%$BB.$/$J$j$^$9!%(B 284MHz$B$K$"$2$F$b!$$$$^$N(B
$B$H$3$m!$CWL?E*$JLdBj$O=P$F$^$;$s!%(B 21066A$B$N%3%"$,(B 21064A$B!J(B275MHz$B!K$J$N$G(B
$B!$(B284MHz$B$J$i==J,M>M5$,$"$k$N$G$7$g$&!%%8%c%s%Q$N0LCV!$@_Dj$O!$(BDEC$B$N(B ftp$B%5%$%H(B
ftp.digital.com$B$K$"$k(B Design Guide$B!$(Bcpu_ref.txt$B$r$4Mw$/$@$5$$!%(B
If you wanna go further
$B$b$&>/$7!$2~B$$9$k(B
for 256k cache, you can change speed to 12ns(look for SROM select jumpers,
cpu_ref.txt)
its hard to get hold of 12ns cache memory at this moment, but you can use
15ns, it has no problem so far.
256K$B$N>l9g!"%-%c%C%7%e%a%b%j$N%9%T!<%I$r(B 12nS$B$K@_Dj$9$k$3$H$,$G$-$^$9(B ( SROM
select jumpers$B!$(B cpu_ref.txt $B$r;2>H(B)$B!%8=:_!$(B 12nS$B$N%-%c%C%7%e%a%b%j$O!$<j$K(B
$BF~$j$K$/$$(B
$B$h$&$G$9(B($B<{MW$,>/$J$$$+$i!$;EF~$l$k$N$r$d$a$?$H8@$o$l$?(B)$B$,!$(B 15nS$B$N%a%b%j$G$b!$(B
$B:#$N$H$3$mCWL?E*$JLdBj$O=P$F$^$;$s!%(B
go even further.
$B$b$C$H!$2~B$$9$k!J$A$g$C$HLLE]!K(B
this is kind of kinky(...)
change pci bus clock up!
change 33.3333mhz to 40.000mhz(borrow the power) and from this change main
memory bandwidth
go to peak 80MB/sec to about 100MB/sec. It has big effect on large program
which won't go to 2nd level cache.(to change this pin1 is not switch, so if
you can't fix it, DON'T DO IT)
$BY]GK$j$G$9$,!$(B PCI$B%P%9$N%/%m%C%/$r$"$2$F$_$^$7$?!%(B 33.3333MHz$B$N%*%7%l!<%?$r$O$:(B
$B$7$F!$(B 40.000MHz$B$N%*%7%l!<%?$+$i!$(B 40MHz$B$N=PNO$rGR<Z$7$^$9!%$3$l$K$h$C$F!$%a%$(B
$B%s%a%b%j$N%P%s%II}$,(B peak 80MB/sec$B$+$iLs(B100MB/sec$B$K>e$,$k$N$G!$#2<!%-%c%C%7(B
$B%e$K:\(B
$B$j$-$i$J$$$h$&$J%W%m%0%i%`$G$O8z2L$,Bg$-$$$G$9!%(B ($B$3$N%\!<%I$G;H$o$l$F$$$k%*%7%l(B
$B!<%?$O!$(B pin1$B$,%9%$%C%A$K$J$C$F$$$J$$$N$G!$2~B$$N$?$a$K$O30$5$J$1$l$P$J$j$^$;$s(B
$B!%%@%a$@$C$?$H$-$K!$85$KLa$;$k<+?.$,$J$$$J$i!$<j$r=P$9$Y$-$G$O$"$j$^$;$s(B)
and on his caluclator, 56mhz bus is working.but not too much of change.
$B$5$i$K!$(B 56.0000MHz$B$G$b(B($B;d$N7W;;5!$G$O(B)$B0BDj$7$FF0$$$F$$$^$9!%(B40.000MHz$B$KBP$7$F(B
$B!$%j%K%"$K9bB.2=$9$k$o$1$G$b$J$$$G$9$,!$$=$l$J$j$KB.$/$J$j$^$9!%(B
Clock up to 297MHz!
once before, he wrote max is 284mhz, but that's when you are using 3.3v, if
your cpu is cool enough, you can change it.
When you do this, actually system get more stable at faster cpu speed as well.
check behind board's R2 and R3(they are regulating R2's both side to 1.25V)
to parallel to R2, put 470$B&8(B, it boost up to about 3.6V(this is regulators m
ax).
in this case BogoMIPS=293.6$B!$(Bclock speed approx. 297MHz$B!J(Bguess from BogoMIPS
, and depends on cpu)
at this speed it gets REALLY hot, be careful with cooling.
$BA0$K!$(BVCO$B$N8B3&$,(B 284MHz$B$G$"$k$H=q$-$^$7$?$,!$$3$l$OEE8;EE05$,(B 3.3V$B$N$H$-$NOC(B
$B$G$9!%(B CPU$B$NNd5Q$KM>M5$,$"$k$h$&$J$i!$EE8;EE05$r>e$2$k$3$H$G!$H/?6<~GH?t$N>e8B(B
$B$r>e$2$k$3$H$,$G$-$^$9(B($B$=$l$@$1$G$J$/!$9bB.2=$KBP$9$k0BDj@-$,>e$,$k8z2L$b$"(B
$B$k(B)$B!%(B
$B%\!<%II=$N!$(BCPU$B2#$N%l%.%e%l!<%?$G!$(B3.3V$B$r:n$C$F$$$^$9$,!$$3$N(B 3.3V$B$r7hDj$7$F$$(B
$B$k$N$,!$%\!<%IN"$NDq93(B R2$B!$(BR3$B$G$9!J(BR2$B$NN>C<$NEE05$,(B 1.25V$B$K$J$k$h$&$K%l%.%e%l!<(B
$B%H$7$F$$$k!K!%(BR2$B$KJBNs$K(B470$B&8$rF~$l$F$d$k$H!$Ls(B 3.6V$B$KEE05$,>e$,$j$^$9!J$3$l$,!$(B
$B$3$N%l%.%e%l!<%?$N8B3&$J$N$G!$$3$l0J>e$K$O>e$2$i$l$J$$!K!%$3$N>uBV$G(B
$B!$(BBogoMIPS=293.6$B!$%/%m%C%/<~GH?tLs(B 297MHz$B!J(BBogoMIPS$B$+$i?dDj!$(BCPU$B$K$h$j8DBN:9(B
$B$,$"$k!K$K$J$j$^$9!%(BBogoMIPS$B>e>:J,DxEY%9%T!<%I%"%C%W$7$^$9!%(B CPU$B$OH/G.NL$,$5$i(B
$B$KBg$-$/$J$j$^$9$+$i!$Nd5Q$K5$$rIU$1$F2<$5$$!%(B
about 1mb cache memory.
1MB$B%-%c%C%7%e%a%b%j$K$D$$$F(B
if you change to 1mb, it speeds up about 20%. you need 400mil width(wider
than 256k chip) 128k x 8 bit chip. dont' get confused with 300mil width
(same width with 256k).
you need to weld(spell) cache sockets' wide side' 31, 32 pin to narrow side
of 31, 32 pin.
then you can use 300 mil chip. becareful to the pin between(pattern going
thru).
$B%A%C%W$rF~$l49$($k;v$G!$%-%c%C%7%e%a%b%j$r(B 1MB$B$K$9$k$3$H$,$G$-!$(B20%$BDxEY9bB.2=(B
$B$5$l$^$9!%(B 400mil$BI}(B( 256k$B%A%C%W$r$h$j9-$$(B )$B$N(B 128K x 8bit$B$N%A%C%W$,I,MW$G$9(B
$B$,!$J6$i(B
$B$o$7$$$3$H$K(B 300mil$BI}(B ( 256k$B%A%C%W$HF1$8I}(B )$B$N$b$N$bGd$i$l$F$$$^$9!%$A$g$C$H(B
$BLLE]$G(B
$B$9$,!$H>EDIU$1$G!$%-%c%C%7%e%a%b%j$N%=%1%C%H$NI}$N9-$$J}$N(B 31$B!$(B32 pin$B$r69$$J}$N(B
31$B!$(B32 pin$B$K$=$l$>$l$D$J$$$G$d$k$H!$(B300mil $B$N%A%C%W$G$b;H$($^$9(B($B%T%s4V$K%Q%?(B
$B!<%s$,(B
$BDL$C$F$$$k$N$GCm0U(B)$B!%%-%c%C%7%e%a%b%j$rM_$7$$J}$O!$$W$i$C$H%[!<%`$KCmJ8$9$k$H!$(B
$BLdBj$,>/$J$$$H;W$$$^$9(B($B%]%T%e%i$JItIJ$G$O$J$$$N$G!$BT$?$5$l$k$+$b$7$l$J$$(B)$B!%(B
1mb cache has problem, that correctable ECC error occurs occasionally.
even if you have 15ns, you need to lower cache mem. speed.
from 12ns, 15ns to 20ns. if you are boosting up the cpu's voltage
regulator change to 15ns.
demerit for this is that you need to lower the speed of cache, but since
you have 1mb, this merit
is over the other.
1MB$B%-%c%C%7%e$K$OLdBj$,$"$C$F!$(B correctable ECC error$B$,5/$-$d$9$/$J$k$?$a!$(B
$B$?$H$((B
15ns$B$N%A%C%W$G$"$C$F$b!$%-%c%C%7%e%a%b%j$N%9%T!<%I$r2<$2$J$1$l$P$J$j$^$;$s(B ($BDL>o(B
12ns$B!$(B15ns$B$+$i(B 20ns$B$X!$(BCPU$B$N6!5kEE05$r>e$2$F$$$k>l9g$K$O(B 15ns$B$X(B )$B!%%-%c%C%7(B
$B%e%a%b(B
$B%j$N%9%T!<%I$r2<$2$J$1$l$P$J$i$J$$%G%a%j%C%H$,$"$j$^$9$,!$%-%c%C%7%e$r(B 1MB$B$K$9(B
$B$k$3$H$K$h$k8z2L$NJ}$,Bg$-$/!$==J,%a%j%C%H$,$"$j$^$9!%(B
Let me how your machine goes, but I DON'T RECOMMEND THIS AT ALL, TAKE YOUR
OWN CHANCE.
NO ONE, BUT YOU ARE RELIABLE ON THIS MATTERR, THIS IS JUST INFORMATION.
-- To unsubscribe: mail -s unsubscribe axp-list-request@redhat.com < /dev/null
Copyright © 1995-1997 Red Hat Software. Legal notices