I.
Actual
Regression (using
Stata 5.0)
. reg frames vidr occ sys slid pix f4d bub
p2 tnt sli if vidr<24 &
sys>64 & dropid<1
Source | SS df MS
Number of obs = 56
------------+------------------------------ F(
10, 45) = 74.33
Model | 1882.97953 10
188.297953 Prob > F = 0.0000
Residual | 113.991263 45
2.533139 R-squared = 0.9429
------------+------------------------------ Adj R-squared = 0.9302
Total | 1996.97079 55
36.308559 Root MSE = 1.5916
------------------------------------------------------------------------------
frames | Coef. Std. Err. t P>|t| [95% Conf. Interval]
------------+--------------------------------------------------------------------
vidram | .8784342 .198996 4.414 0.000 .4776358
1.279233
occpumhz | .0404642 .0033362 12.129
0.000 .0337446 .0471837
sysram | .0330379 .0084217 3.923 0.000 .0160757
.0500001
slidert | -9.821762 .505197
-19.441 0.000 -10.83928 -8.804243
pixels | -.0162767 .0026963
-6.037 0.000 -.0217072 -.0108461
f4dummy | 2.346115 .7374249 3.181 0.003 .8608646 3.831365
bubble | -.0328498 .0857122 -0.383 0.703 -.205483
.1397835
p2 | 1.338443 .6136032 2.181 0.034 .1025831 2.574304
tnt |
4.363629 .7543733 5.784 0.000 2.844243
5.883015
sli |
3.53395 .7420363 4.763
0.000 2.039412 5.028488
_cons | -6.080369 5.003273 -1.215 0.231 -16.15748 3.996739
------------------------------------------------------------------------
*run for vidram<24, sysram>64, &
id #3 and id#13 excluded as extreme outliers.
II.
Tables
(A)
Variables |
Net Frames Per Second Gained/Lost inPapa Doc Benchmark Tests |
Hardware O (Operational Range Tested)a |
|
|
CPU Mhz (range 225–504Mhz) |
.040 per Mhz
/2.00 per 50 Mhz** |
|
Video Card per
MB (range 4–16MB) |
.878 per MB/7.02
per 8 MBb** |
|
System Ram per
MB (range 65–256MB) |
.033 per MB/2.11
per 64 MB** |
|
TNT Chip/D3D (range no/yes 16MB) |
4.36** if have |
|
SLI (range no/yes) |
3.53** if have |
|
P2 CPU
(range no/yes) |
1.34** if have |
|
|
|
|
F4 (Software) (Range Tested) |
|
|
Gouraud On/All Sliders Right (v. Off/All Left) (On/Off) (On/Off; All Sliders Right/ All Left |
-9.82** |
|
Pixels per 1st Pixel Number* (640x/800x) |
-.016 per pixel/-2.56
switching from 640 to 800** |
|
Bubble 7 (v.
1) (7/1) |
-.033 per tick/-0.20
switching from 1 to 7*** |
|
Version 1.03 (v.
1.00) (1.03/ 1.00) |
+2.34** |
|
|
|
|
|
Number Obs = 56 |
|
|
F(10,45)
= 74.33 |
|
|
Prob > F =0.0000 |
Initial Results 1/18/99 ARHarris; Using Stata 5.0 |
R2 = .943 |
*1st
Number of Pair= 640 v. 480 (product alternative: (640*480=307,200 pixels v.
800*600=480,00 pixels)
**significance
<.001
***not
significant
aTop CPU Mhz Reported (ie, if
Celeron 300 OverClocked to 450, then 450 Mhz Used.)
bEstimated Vidram effects above 16 MB are currently
still uncertain.
(B)
Variables |
Current
Market $$
Price Estimates |
Net Frames Per Second Gained/Lost inPapa Doc Benchmark TestsPer $100 Spent in Current Market Conditions |
Hardware (Range Tested) |
|
|
|
CPU (range 225–504Mhz) |
P2: $0.935/Mhz AMD: $0.383/MhzCeleron: $0.345/Mhz |
P2: $100 Buys 106.95 Mhz @
.040fr/Mhz + P2 CPU Gain of 1.34 4. = 4.28 frames/sec + 1.34 frames/sec = 6.62
frames/sec AMD: $100
Buys 261.10 Mhz @ .040/Mhz = 10.44 frames/sec Celeron: $100 Buys 289.86 Mhz @ .040 fr/Mhz = 11.59 frames/sec |
|
|
|
|
|
Video Card per
MB (range 4–16MB) |
$11.290/MB |
$100 Buys 8.86 MB @ 0.88 fr/MB =
7.80 frames/sec* |
|
|
|
|
|
System Ram per
MB (range 65–256MB) |
$1.630/MB |
$100 Buys
57.24 MB @ 0.033fr/MB =
1.89 frames/sec |
|
|
|
|
|
TNT Chip/D3D (range no/yes
16MB) |
$8.125/MB |
$100 Buys 12.31 MB @ 0.273fr/MB =
3.37 frames/sec** |
|
|
|
|
|
|
|
|
|
F4 (Software) (Range Tested) |
|
|
|
Gouraud Off/All Sliders Left (v. On/All Rightt) (On/Off) (All Sliders Right/All Left Off/On) |
|
$ 0.00 Buys................................
9.82 f/s(Gour off/all slid left) |
|
Pixels per 1st
Pixel Number* (640x/800x) |
|
$ 0.00 Buys................................. 2.56 f/s (at 640x480) |
|
Bubble 1 (v. 7) (1/7) |
|
$ 0.00 Buys................................. 0.20 f/s (bubble at 1) |
|
Version 1.03 (1.03/ 1.00) |
|
$ Cost of 1.03 Download................
2.34 f/s (version 1.03) |
|
|
|
|
|
|
|
Number
Obs = 56 |
|
|
|
F(10,45)
= 74.33 |
|
|
|
Prob > F = 0.0000 |
Initial Results 1/18/99 ARHarris; Using Stata 5.0 |
|
R2 = .943 |
*NB: Up to 16MB; Gain above 16
MB Gain is Currently Uncertain.
**This number is just for the
purpose of toe-to-toe comparison ; the assumption is you’d buy a 16MB
TNT card.
Software Changes
(1)
First, No Charge, set Gouraud
Off/All Sliders Left if you can stand it.
(2)
Second, No Charge, set Pixels
to 640 x 480 if you can stand it
(3)
Third, Tiny Charge, run
Falcon v. 1.03 if you can stand it..
Now Turn to Hardware:
Hardware
Changes
(4)
First $$$, upgrade to the
fastest Celeron you can afford and overclock it [to a max safe oc speed of
500Mhz+/-]. If you’re at 200 Mhz you can buy $100 worth of Celeron Mhz upgrade
(+12 f/s), you can put Gouraud/Sliders Full On (& still be 2 fps ahead of
where you were).
(5)
Second $$$, make sure you are
upgraded to to 16MB of video RAM. (If you are at 8MB or less.) In going from 8
to 16 MB, by the slimmest of margins, the vote is for TNT/D3D over v2/Glide:
-16 MB
TNT
-> 4.36 f/s (for TNT) + 7.040
f/s for add’l 8MB of Video Ram = 11.40 f/s gain
versus
-16 MB SLI v2 -> 3.53 f/s (for SLI) + 7.040 f/s for add’l 8MB of
Video Ram = 10.57 f/s gain (corrected 1/19/99 ed.)
(6) Next $$$, buy as much
system RAM as you can afford
__________________________________________________________________
*”Buy
Wise Recommendations” are based on current estimates which: assume valid test
reporting; assume linear, additive
effects; may vary from future estimates once more
data are gathered; are not guaranteed
for any specific system. They are also made
ceteris
paribus (“other things being equal”). So, for example, given your
overall
computing needs and uses, buying $100
worth of system RAM may make much more
overall sense than upgrading your CPU.
IV. Summary
Multiple regression (SLS) was used to analyze the data in Rows 1-27 of Papa Doc’s F4 Benchmarks published as of 1/18/99 at: http://home.earthlink.net/~plegrand/Falcon4Benchmarkspart2themeat.htm
Cases # 3 & # 13were dropped as extreme outliers. Cases with 64MB of system Ram and/or 24 MB of video RAM – few in number - -were dropped because they produced complex interactions and nonlinearity in the findings. More cases with low levels of system RAM and high levels of
video Ram are needed to explore this anomalous area. The variance in Frames/Sec explained by the 8 hardware and software variables together was extremely high (about 94%, p<.0000).
Certainty that estimated effects were correctly identified (from highest to lowest certainty) ranged from:
Gouraud/Sliders (Highest
Certainty that Correct).............................Software
CPU Mhz...................................................................................................Hardware
Pixels (Resolution)....................................................................................Software
TNT/D3D..................................................................................................Hardware.
SLIc...........................................................................................................Hardware
Video (Accelerator) RAM.........................................................................Hardware
System
RAM.............................................................................................Hardware
F4 Version.................................................................................................Software
P2 CPU......................................................................................................Hardware
Bubble (Lowest Certainty that
Correct)...............................................Software
cProbably a proxy measure of v2 effects.
Prepared by ARHarris 1/18/99
Email aharris@javanet
with questions, comments,
or requests for data matrix
(Excel, text, or other formats).