Is the
MPX2 designed for overclockers?
Iwill
publicly aim their new dual-AMD motherboard at 'workstation' through to
'entry-level server' markets . . . but:
Anyone with
two brain-cells - or Athlon's - to rub together will see from the specification
&
our comments alongside that the MPX2 was designed to appeal to those wishing to
squeeze every last MHz from their CPU's:
Chipset |
AMD
768/762 |
'B2'-step
southbridge: USB v1.1 works |
PCI-slots |
2
x 64-bit/66MHz; 4 x 32-bit/33MHz |
No
PCI-latency adjust/INT assignment |
Clock-generator |
ICS
94225AF |
100-250MHz
in 1MHz soft-BIOS steps |
Dividers |
1:3
[100MHz FSB] & 1:4 [133FSB] only |
Jumper-set:
not a '333' platform |
CPU's
& Multipliers |
[officially]
AMD MP dual; MP, XP, & Morgan Duron single CPU in primary socket |
Works
dual with uncut L5 XP's; [limited] multiplier adjust for MP's in
BIOS |
AGP |
AGP
Pro, 4x, & 2x |
Largest
cards fit |
Socket
462 |
waterblock-friendly
[large] mount-holes; limited room along lug-axis |
Alpha
8045 [80mm sq.] does not fit; maximum fansinks 68mm along lugs |
VIO |
jumper-settings
for +5% & +10% |
Over-volting
of DDR to 2.75v |
Power
[1] |
EPS12V
or ATX 2.03: at least 200W RMS +3.3v/+5v combined recommended |
Dual-mode
sockets: ATX 20-pin & 8-pin plugs do not latch in place |
Power
[2] |
3-phase
to CPU's |
Mosfets
under 'L'-shaped heatsink |
DDR |
4
x slots: max 3 x 1Gb + 1 x 0.5Gb Registered |
unregistered
DDR works: max 2 x 1Gb |
Onboard |
2
x ATA 100 IDE, floppy, front USB 1.1; IRDA |
|
Backplane |
2
x USB v 1.1; 2 x serial, LPT, 2 x PS2 |
onboard
USB |
Size |
292
x 305mm: ATX-E; 4 rows mount-holes |
very
deep: will not fit many or most midi-cases |
BIOS |
Award/Phoenix
2Mb: versions tested a) MPX20509.bin; b) MPX20523.bin c)
mpx21206.bin |
BIOS-Saviour
RD1-LPC2 fits/works |
The
MPX2's Overclocking
tools:
A |
clockgenerator 'support'
for FSB frequency adjust in 1MHz steps from 100>132 & 133>250MHz |
BIOS |
B |
multiplier adjustment from 5.5>15x in 0.5x steps |
BIOS |
C |
vcore
adjustment up to 1.85v in 0.025v steps |
BIOS |
D |
I/3
[100FSB] & 1/4 [133FSB] FSB/PCI dividers |
jumper |
E |
VIO
[+3.3v & +2.5v] motherboard voltage adjustment - +5% & +10% |
jumper |
Of the
above; C &
E are there to stabilise things
[given adequate cooling] when
pushing CPU's & DDR to or over their comfortable limits.
The others look like a
huge luxury of choices, don't they? . . . 100>250MHz . . . 5.5>15
multipliers . . . .
. . but it all comes down to the PCI-buses: if kept within the usual
limits of +10% of standard
. .
X |
[PCI-bus]
= A x D = 33.3>37
MHz [32-bit] -
or 60-74MHz
[64-bit] |
.
. . So the MPX2 reality of the big number: CPU-frequency . .
. |
Z |
[CPU-frequency]
= A
x B = (X/D)
x B = (33.3>37/D)
x B = |
D1
(jumpered
to 100FSB)x
B = 100>111
x (multiplier[s] available) |
D2
(jumpered to
133FSB)x B = 133>148
x (multiplier[s] available) |
.
. so all that these dreary numbers & symbols demonstrate is that the
overclocking potential of any data-critical system based on the MPX2 is
crucially multiplier-bound.
If
Iwill enable the advertised adjustment of CPU multipliers in 0.5x steps [=
66MHz
steps at standard FSB], then even data-critical systems can perhaps indulge in a little
quiet 'clocking.
How
many MPX2 multipliers?
Iwill's
first publicly released BIOS' [dated 12th June 2002] allows no multiplier
adjustment at all for those with Athlon MP CPU's or L1-unlocked XP's.
The
beta BIOS before this [dated 23rd May 2002] allowed very limited
multiplier adjustment, with the multipliers available strangely
inconsistent from MP model to model: our pair of 1.2GHz MP's had a choice
of the whole-number 7x, 9x [default], & 11x; while a pair of '1800+'
MP's was reported to have a choice between 11x & 11.5x [default] only.
Since
the MPX2 is multiplier-bound [as a 'clocking server-board], we would recommend
waiting to see if Iwill can sort out the MPX2's BIOS before
buying: it's a not entirely trivial problem for Iwill, since the recent extension of
Athlon multipliers to 15x uses an extra L10 'bridge' to sort the pairs.
If
Iwill are unable to provide working multiplier-adjust; the MPX2 is less
desirable as a serious work-platform [with a bit of safe tweaking
allowed]; but will remain a good 'clocking platform.
MPX2
Dividers:
Using a
VICS/Ioss RD2 Pro PCI Bus analyser - its readout of PCI-bus speed is to
the mid-left of the picture below - we confirmed the MPX2 has only the 1:4
divider from 133MHz on up. This device links by cable to a special PCI-card with an
onboard processor, crystal, & Cypress bus control PLD, so gives an
independent hardware reading of PCI-bus frequency.
As you
see; at 137MHz FSB, the 32-bit PCI-bus is at 34.2MHz & the 64-bit bus
at 68.4MHz:

This table
shows how things might pan out
on the MPX2 for a pair of MP CPU's being kept 24/7 stable at around 1200 MHz;
with the user maximising memory bandwidth by increasing the FSB-frequency: |
multiplier |
speed |
FSB |
Divider |
32-bit/33MHz PCI |
64-bit/66 PCI |
PCI-spec % |
12 |
1200 |
100 |
1:3 |
33.3 |
66.7 |
+0 |
11.5 |
1207 |
105 |
1:3 |
35 |
70 |
+5 |
11 |
1199 |
109 |
1:3 |
36.3 |
72.7 |
+9 |
10.5 |
1197 |
114 |
1:3 |
38 |
76 |
+14 |
10 |
1200 |
120 |
1:3 |
40 |
80 |
+20 |
9.5 |
1197 |
126 |
1:3 |
42 |
84 |
+26 |
9 |
1188 |
132 |
1:3 |
44 |
88 |
+32 |
9 |
1200 |
133 |
1:4 |
33.3 |
66.7 |
+0 |
8.5 |
1198.5 |
141 |
1:4 |
35.25 |
70.5 |
+6 |
8 |
1200 |
150 |
1:4 |
37.5 |
75 |
+12.5 |
7.5 |
1200 |
160 |
1:4 |
40 |
80 |
+20 |
7 |
1197 |
171 |
1:4 |
42.75 |
85.5 |
+28 |
6.5 |
1202.5 |
185 |
1:4 |
46.25 |
92.5 |
+39 |
6 |
1200 |
200 |
1:4 |
50 |
100 |
+50 |
5.5 |
1199 |
218 |
1:4 |
54.5 |
109 |
+63.5 |
The obvious
problem with the MPX2 is that it's missing one vital tool: a 1:5 [or even a 1:6]
divider. With these, the table might read . . . .
multiplier |
speed |
fsb |
divider |
32-bit/33MHz |
64-bit/66MHz |
PCI-spec
% |
7.5 |
1200 |
160 |
1:5 |
32 |
64 |
-4 |
7 |
1197 |
171 |
1:5 |
34.2 |
68.4 |
+2.5 |
6.5 |
1202.5 |
185 |
1:6 |
30.8 |
61.6 |
-7.5 |
6 |
1200 |
200 |
1:6 |
33.3 |
66.7 |
+0 |
5.5 |
1199 |
218 |
1:6 |
36.3 |
72.7 |
+9 |
Without
dividers higher than 1:4, a serious workstation or entry-level server based on the MPX2
is limited by sanity & data-security to an absolute maximum FSB below 150Mhz - it simply doesn't matter whether or not the thing will run at FSB's
higher than this.
Balls-out
overclockers, & users of low-end machines with known-durable IDE hard disks [with
no valuable data on 'em] might amuse themselves with trying for FSB's of 160MHz
or greater; but the markets Iwill publicly target for the MPX2 have at most
17Mhz of the advertised '133>250' to play with.
We found the
MPX2 to be an easy 'clocker up to the low 170'sMHz FSB - at 167FSB, as you see,
the PCI-bus is at 41.7MHz:

Does
the XP work?
Yes,
unlocked XP's work fine, & are recognised & work as MP's.
We have read an
unconfirmed report that 'L5-locked' XP's will work SMP after a
warm reboot - or a restart from the OS if they've already
registered as MP's: the knack is to cold-boot, then immediately
before the first POST screen to do a hard reboot [ie press the
'reset' button]
Otherwise, on
a normal cold boot, only the primary CPU will be recognised [as an
XP]. |
Wottle
she do then?
The MPX2
is a superb overclocking motherboard: stable, simple, & with
fine-grained adjustment through combinations of the 1MHz FSB-stepping,
& [theoretically] the 0.5x multiplier adjustments.
Our
'workstation' system - with 2 x '1700+' XP's - ran easily up to 155MHz,
& is 24/7 stable at 11x 149FSB [1639MHz]:

Our
'server' system - with 2 x 1.2GHz MP's - settled down to a dull, dreary,
responsible, 24/7 stable overclock of 11x 137FSB [1507MHz].
We
stripped out one system [the 'server' - funnily enough] down to a
64-bit/66MHz SCSI-HBA in a 32-bit slot controlling a single 10K rpm U3W SCSI-HD -
we knew this card would tolerate PCI-speeds up to to 78MHz.
This
system ran Windows 2000 with every appearance of stability at 167MHz FSB;
where we took a range of benchmarks. It continued to run calmly up into
the low 170's; but would not run stably at 175MHz FSB. We didn't max out
the VIO &/or vcore settings to press on further.
There
seems no reason to doubt that a dedicated overclockers' MPX2, with active
cooling, could attempt 200MHz FSB.
This
performance is without rival among MPX-chipset motherboards, &
is a tribute to Iwill's skill in design, careful selection of quality
components, & quality-control in manufacture.
'stable' is a
vague & overused term: we have made available for download
encrypted certificates issued through use of 'CPU Stability Test
v5' for some speeds we hold to be 24/7 stable. Interestingly, our
'workstation' system - certified 'stable' at 150MHz, would not
complete the 'Sciencemark_MP' benchmark at this speed - though it
would do so repeatedly at 149FSB/1639MHz. You may have higher - or
lower - standards for stability & will very properly treat our
opinions with care. |
If you overdo
things through the soft-BIOS so the MPX2 will not POST, an
undocumented escape to force standard [133MHz] FSB is to hold down
the 'insert' key while powering-on. This avoids having to clear
the CMOS. |
|