#9's Imagine 128 Series 2 vs. IXMicro's Twin Turbo M8
- QuickDraw Primitives tests

By Mike, Published: 8/17/1997
QuickDraw Primitives Testing
In an effort to obtain more detailed performance data for each of these cards, I decided to run the many graphics primitives tests in MacBench. Unlike the standard Graphics and Publishing tests which emulate actual application use ( heaving scrolling, etc) these tests isolate the smallest individual QuickDraw drawing functions, and would provide an additional glimpse into the driver and hardware performance interaction of the two cards at the lowest function level.

The results were both surprising and a little disturbing. In particular I noted some very unusual behavior and scores from the Twin Turbo's version 4.02 driver during the "text" primitives test. Driver version 3.8 did not exhibit this oddity. I have contacted IXMicro and Ziff Davis regarding this issue to see if it is a IXMicro driver/Macbench interaction anomoly (it seems so, as v3.8 was about 30,400 times slower at the text test running on the same card). - or an obscure flaw or bug in MacBench.

Using TT driver v4.02 - MacBench crashed several times after running the "DrawText" portion of the test and showed errors in comparisons between identical test runs on this particular test consistently. When Ziff Davis responds I will provide their answers here. Preliminary responses from IXMicro indicated that they had overflowed MacBench's capability in this test. The problems I have in beliving this individual score is representative of the actual performance is due to A) a factor of 30,400 improvement in one driver revision, B) the fact the same driver does so poorly on the DrawChar and DrawString functions, yet suddenly explodes on DrawText, and C) the difference in the huge scores was purely software version driven. I'll let the reader draw their own conclusions however.

Text Score Oddities:
One of the functions in the QuickDraw primitives test suite is text functions - DrawChar, DrawString, and DrawText. Regardless of driver version of the Twin Turbo, it lagged behind the Imagine 128 in the in the DrawChar and DrawString tests by a large margin, and the DrawText scores for driver v3.8 were also much lower that the Imagine 128 card. Using Twin Turbo driver v4.02 however, resulted in scores for the DrawText test that I find hard to believe (a 30,400 times improvement over v3.8) - as it scored 4 billion pixels/sec - yet the DrawChar and DrawString scores were still far lower than the Imagine card. MacBench crashed once after running this test - and every time when attempting to display a comparison of this score to other saved results - even the when comparing duplicate test scores (running the test with the TT, saving it, running again, loading prev. results). For this reason I feel that the numbers provided are some sort of overflow in MacBench itself.

At this point I was tempted to throw out all the 4.02 driver tests - as my immediate reaction was that the driver may have been tweaked just for MacBench or that MacBench results could not be trusted. The DrawText score just does not make sense.

Another oddity about this combination was that running the test with the TT 4.02, saving it, the running it again resulted in the first test score (still the same 4 billion score) being shown as "0%" and a zero score in comparison to the 2nd test run - as shown here. At this point scrolling in Macbench would usually cause an immediate "Error 1072 - overflow" error crashing MacBench. I can only guess that the number is too large for MacBench to handle (another reason I think the number is not a valid one).

After having lost over an hour of test results due to the Twin Turbo 4.02 driver crash (yes, I should have saved...), I used TT driver v3.8 for the Text tests. Only the text primitives test comparison uses driver version 3.8 - all other test results shown here use the later version 4.02. The Text test results with v3.8 are shown below.

MacBench 4.0 Text Function Scores: 1152x870, millions colors Text Primitives results

For the remainder of the scores TT driver v4.02 will be used, as comparisons of these scores did not generate MacBench overflow errors as did the text test scores.

For the remainder of the test areas I'll just present the results.

MB Arcs tests

MB lines tests

MB Ovals Test

MB Polys

MB Rectanges tests

Round Rectanges:
MB Rnd Rectanges tests

MB Regions tests

The number of tests here are huge - and so the graph image file is very tall. Although it's only a 25k file, due to the height of the image I decided rather than to load the image automatically I'd let the reader chose. If you want to view it click Here.

As you can see, the Twin Turbo did much better in some of the QD primitives tests than the Imagine, but the issue of the 4.02 driver's text scores remain as a dark cloud over this part of the test in my mind. The application, scrolling and zoom tests made me wonder how valid these scores really are, at least in regards to actual application use. The fact that Speedometer and Norton's Video tests showed the Imagine to be significantly faster than the Twin Turbo using similar primitives added to my suspicions. It may simply be that the Twin Turbo 4.02 driver has uncovered a problem in the MacBench 4.0 Quickdraw Text primitives test code.

Primitives Winner: Twin Turbo
(but it's a mixed bag - and the 4.02 text score issue remains unexplained)

on to Video Test Results =>>

Copyright © 1997. All Rights Reserved
Brand/product names mentioned here are properties of their respective companies.
Legal: Users of this web site must read
& are bound by the terms & conditions of use.

For more recent video card reviews - see www.xlr8yourmac.com/video.html.