I got about thirty responses from those who tried the Radeon 9500 modification. About two-thirds reported “success,” a third reported failures.
Can’t say getting a particular card from a particular place did any good; for instance, we got reported failures and successes from cards bought at Newegg.
Frankly, I can’t say I would take this evidence and testify in court with it.
Data also came in rather lopsidedly. When I first put up the article, I got mostly failures. After updating the front page to indicate initial results, I got a bunch of successes. Then it sort of evened out.
A few people said, “I have artifacts, but it was a success.”
Well . . . .
On the other hand, it seems that some of those who try this try to crank these cards up to overclocked Radeon 9700 Pro levels, and consider it a failure if they don’t make it.
The memory chips on 9500s are slower than those found in 9700s, on average, they aren’t going to do as well.
Most used the software mods (even those who initially made hardware mods), and results didn’t seem to be affected by the type of mod made, so there’s no point in making the hardware mod.
Those who failed didn’t have problems getting the card back to regular 9500 status, but they sure weren’t too happy with the stock performance they got for their $160. They didn’t find it much better than cards like the Radeon 8500 it was meant to replace.
I just don’t think the quantity and quality of the data is good enough for any precise measurement.
However, just because data isn’t quite good enough for an exact measurement, that doesn’t mean it’s completely useless.
Given the skewing factors mentioned above; the actual success rate is probably not as low as 30% nor as high as 70%. It’s probably somewhere inbetween.
You also have to keep in mind what the data will be used for: deciding whether to risk a successful software mod or not. This is not like comparing two pieces of equipment. Whether the failure rate is 45% or 55% isn’t (or at least shouldn’t) make any real difference in the decision.
Probably the best way to assess the risk here is to assume it’s a coin-flip: heads it will, tails it won’t.
If $160 means something to you, those kinds of odds mean you ought to know before you buy what you’re going to do with the card if it can’t handle the mod.
Now if the chairman of EBay has you on his Christmas card list, you already have an exit strategy.:) If you can sell the card to Cousin Joe, you do, too.
If you’re replacing an antique with this card, you may well find even a failure a big improvement, and consider it only a $50 gamble you just happened to lose.
However, if you don’t fit any of the above, $160 means a lot to you, and this is meant to replace a Radeon 8500/Ti4200 class card, maybe you don’t want to do this. You’ll get little gain for your $160 in case of failure, for you, it’s gambling $160 on a bet that half the time will get you a $230 card, half the time essentially nothing.
It would be wiser to save up a bit more and get a Radeon 9700 instead (which is $230, and probably will at least modestly dip in price as time goes by in the near future), or at least see if these 9500s go down in price.
Personally, if I were in that boat, I wouldn’t do it. If it were an $80 or $100 card, I would, but at $160, the risks exceed the rewards.