Running Snow Leopard in 64-bits

Question

This topic seems to come up again regularly, but I still see no definitive information.
I have a very recent MBP (MacBookPro5,3), so its firmware is 64-bit capable (EFI64). However, if I try to boot it in 64-bit mode by holding down the 6 and 4 keys while powering off and restarting, it still always comes up in 32 bit mode(i.e. profiler shows "64-bit Kernel and Extensions: No").
How can I boot a 64-bit kernel?

Answer 1

rippo wrote:
FYI- I just forced it into 64-bit mode using the "6 & 4" startup. I didn't see any noticeable speed increases, and USB Overdrive (32-bit) didn't work properly.
After you raised this issue, I did some more tests. The tests indicate why you didn't notice a speed increase. Specifically, no performance benefit accrues in relation to specific software from starting in 64-bit mode. However, if you had compared 32-bit and 64-bit software of the same type, then, given sufficient RAM, you might have seen the sorts of results I show below--no matter which kernel mode you used for the two pieces of software (i.e., mix and match however you like). The point is that your machine will run software correctly no matter what.
Anyway, I have both Geekbench 32-bit and 64-bit. I made five benchmark passes for each of the four possible combinations of Geekbench “bitness” and Sno kernel mode (e.g., 32-bit Geekbench and 64-bit kernel mode is one such combination). Below are the Geekbench averages for each combination.
(1) 64-bit Geekbench run in 64-bit mode--4,268 avg of 5
(2) 64-bit Geekbench run in 32-bit mode--4,264 avg of 5
(3) 32-bit Geekbench run in 64-bit mode--3,910 avg of 5
(4) 32-bit Geekbench run in 32-bit mode--3,907 avg of 5
My results suggest the following.
(a) The existence of essentially zero kernel mode effects—which are the effects of moving to a higher kernel mode and holding the Geekbench bitness constant.
For (1) and (2) above, the effect is (4268 – 4264) / 3908 = .00094
For (3) and (4) above, the effect is (3910 – 3907) / 3908 = .00077
(b) The existence of nice Geekbench bitness effects—which are the effects of moving to a higher Geekbench bitness level and holding the kernel mode constant.
For (1) and (3) above, the effect is (4268 – 3910) / 3910 = .092
For (2) and (4) above, the effect is (4264 – 3907) / 3907 = .091
These results are very sensible to me.* First, they suggest that a Mac with an Intel 64-bit processor will run 64-bit software equally well whether started in 32- or 64-bit kernel mode. Next, the same type of observation applies to 32-bit software assuming it will run in 64-bit kernel mode. Third, from an overall perspective, these observations suggest that there is no performance benefit related to specific software from booting into 64-bit mode. Finally, my results suggest that appreciable benefits from running 64-bit software accrue if one is running as little as 4 GB of RAM--as I am.
*These results contradict some results I presented earlier—some of which did not make sense to me. The prior results were done quickly and I had only one pass for each combination. The results could have been driven by this factor. Alternatively, I could have made a mistake in recording my results—which I doubt that I did, but maybe.