CompuServe Thread

#Graphic Card Review

5 messages in this thread
#21838From: Off BroadwayNov 25, 1991 9:35 AM
Ken… I disagree, speed tests are important. I agree that if you are only doing a series of commands and timing them overall, that does not show much. But if your benchmark breaks down the test by entity type, % of screen size, and a good curve on the results then you might find out something. With GRPERF I found out that many ADI drivers fall all over themselves with 0 width polylines. If I had not done the test then I would not have found this out and gotten programmers to fix it. I asked one programmer of a company that did terrible on 0 width plines how could they put out an ADI driver that did so badly, his response was that his job was programming not drawing. Lame response. But it points out that some companys never really test their drivers before shipping. I've also found that if a company does test it's drivers, it usually uses the San Diego Benchmark test. A major company spent two months testing their new card against the market using BM-ACAD. What a waste. I agree that some cards will not benchmark well. But AutoCAD users, ADI developers and magazine editors <grin> all want speed tests. My two tests are the result of my experience with ADI drivers and 6 months of development between me and TonyT. I feel they do a pretty good job of testing ADI drivers. Robert
#21885From: Evan Yares [WeirdTM]Nov 25, 1991 12:49 PM
Ahem….
#21976From: Kenneth Elliott [CAM TM]Nov 25, 1991 9:42 PM
Robert, I agree that benchmarks such as GRPERF are wonderful to diagnose peformance of drivers and cards. Your 0 width polyline example is a good one. As you noted, you would not have found this problem without your test. Doesn't this reflect how little the user notices such things? I don't view a graphics card a something that *helps* to get things done, but rather something that gets in the way. If I have to wait for *anything* during the design phase, the computer is getting in the way. A good board is one that doesn't cause me to wait on it. So, if the board draws 0 width plines slow, thats ok, as long as it gets finished before I'm ready. The biggest problem I have with benchmarks is they hurt a productive tool which might be 'fast enough' while helping a faster, but poorly implemented product. I'm rather PO'ed at one board maker that promised a slow menu system on a beta I tested was already fixed. Later, I discovered they failed to fix the shipping software. I would have burned them on the review, but now its too late. So, an end user might decide I'm a real jerk because I didn't mention the slow menu system. Or worse. Does this happen to you or did they just single me out….<g> Ken <getting a *little* smarter every day> ///
#22261From: Off BroadwayNov 27, 1991 12:25 PM
Ken… For 0 width plines, some guys' ADIs were way too slow to be acceptable. Until I pointed out the problem they did not know that it existed. Only by getting performance information could the manufacturer go back and improve his work. I disagree–the drafter knows when his driver is slow. His only choice after purchase is to live with it. I agree that if the board finishes a command before you do, that speed is acceptable. Benchmarks are as much for the board manufacturer as the CAD draftsman. Cadalyst has a policy of only reviewing shipping products. Otherwise you can get caught by the difference between beta and shipping product. I've had a few attempts to end run the reviewer myself. Not fun. Robert
#22321From: Kenneth Elliott [CAM TM]Nov 27, 1991 6:13 PM
Robert, One point on which I believe we both agree. If the draftsman perceives the card as slow, its slow. If he's happy, well….<g> Ken /// CAD Masters