Graphics Benchmark (was Re: Squeak on SPARClassic, other speed
questions.)
Doug Way
dway at mat.net
Fri Oct 29 06:06:45 UTC 1999
--0-1040572081-941177205=:12120
Content-Type: TEXT/PLAIN; charset=US-ASCII
On Wed, 27 Oct 1999, John Duncan wrote:
> Also, I was wondering. On my P233 ThinkPad, Morphic is a dog,
> especially in 2.6. I heard people discussing the high speed of
> PMacSqueak and the dogginess of IntelSqueak. The 'lassic, being a
> big-endian machine with tons of registers, is supposed to have an
> architectural advantage for Squeak over the Intels, but is this enough
> to make Morphic fast at low clock rates?
Often, if Morphic "feels" slow, the kind of graphics card you have or your
screen depth setting will matter as much or more than the type of CPU.
The benchmarks that come with Squeak measure message send and bytecode
speed, so they won't help you with figuring out graphics card/screen depth
issues.
So, I whipped up a simple benchmark which attempts to measure graphics
performance in Squeak. (See the bottom of this message. Run "Morph
simpleGraphicsBenchmark".)
It's fairly brain-dead... it just opens a workspace at a fixed size, and
scrolls a bunch of text up and down 20 times. (It does temporarily
disable inboard scrollbars, which can be a factor.) The results seem
fairly consistent. Having a much larger or smaller default font makes a
slight difference in the outcome (though not much), so it's best to run
this with the default font settings.
The benchmark mostly exercises the DisplayScreen primitive, plus some
character scanning code.
Fancier benchmarks ("speed tests" might be a better term) could probably
be created which measure the performance of common Morphic operations such
as resizing windows (in non-fastDrag mode)...
Here are some results from some tests I ran. (I ran 2 tests per screen
depth combination.) Mostly, you just want to pay attention to the values
where Squeak's screen depth matches the host OS's screen depth, although
the other values can be interesting. (e.g., on the NT box, you pay a huge
penalty if the two don't match, but it's not as big on deal on the other
platforms.)
Anyway, I find Morphic very usable at a value of 1000, reasonably usable
at around 3000, and pretty tough to use above 7000 or so.
- Doug Way
EAI/Transom Technogies, Ann Arbor, MI
dway at eai.com, dway at mat.net
http://www.transom.com
266 MHz Apple iMac:
8-bit Sqk: 16-bit Sqk: 32-bit Sqk:
Host OS = 8-bit: 1512 1510 2287 2282 3123 3138
Host OS = 16-bit: 2321 2330 2023 2027 3165 3149
Host OS = 24-bit: 3049 3024 2959 2978 3075 3044
300 MHz Apple iBook:
8-bit Sqk: 16-bit Sqk: 32-bit Sqk:
Host OS = 8-bit: 1226 1223 2033 2022 2870 2883
Host OS = 16-bit: 2247 2249 1621 1616 2752 2721
Host OS = 24-bit: 3431 3425 1983 1979 2586 2542
300 MHz VAResearch Pentium II, Windows NT, ATI 3D RAGE IIC AGP card
8-bit Sqk: 16-bit Sqk: 32-bit Sqk:
Host OS = 8-bit: 951 941 5257 5258 5548 5538
Host OS = 16-bit: 931 931 1282 1272 3395 3375
Host OS = 24-bit: 991 982 1682 1672 3315 3355
Host OS = "True Color": 1101 1112 1552 1553 1632 1612
180 MHz Toshiba Pentium I laptop, Windows 95, Chips & Tech. 65554 PCI
board
8-bit Sqk: 16-bit Sqk: 32-bit Sqk:
Host OS = 8-bit: 6829 6762 8574 8924 11489 11521
Host OS = 16-bit: 7432 7441 9282 9265 11523 11428
Host OS = 24-bit: 8162 8128 10184 10143 14288 14305
--0-1040572081-941177205=:12120
Content-Type: text/plain; name="simpleGraphicsBenchmark.st"
; x-mac-type="54455854"
; x-mac-creator="522A6368"
Content-Disposition: attachment; filename="simpleGraphicsBenchmark.st"
Content-Transfer-Encoding: imap_stub
0,664,2,1628,26,
--0-1040572081-941177205=:12120--
More information about the Squeak-dev
mailing list
|