On Tuesday 19 December 2006 15:49, John Andersen wrote:
On Tuesday 19 December 2006 14:28, Randall R Schulz wrote:
Why on earth must you be so rude?
Pot/Kettle.
Compounding the offense by accusing me of being rude.
But it wasn't my intent.
What, then?
And how, exactly, is the kernel that's running my system just fine, "wrong?"
Its not the native kernel for that hardware.
The hardware is equally i386 and x86_64. A 32-bit kernel is every bit as "native" as a 64-bit one.
As for your theory, a quick look at the clock times will reveal that the 64bit registers load in the same number of clocks as the 32. And the 64bit address registers, pointer registers, etc also load in the same number of clocks as the 32 bit registers. So moving stuff around in memory is faster doing so in 64bit chunks.
It's not about the processor, OK? It's about how much demand 64-bit everything places on the memory subsystem. Why fetch and store so many 64-bit quantities that don't make any use of the upper 32 bits? It's all a big waste of memory bandwidth.
The reason for 64bit processors is NOT simply address space.
If one has no need for 64 bit anything, it's all a waste.
Drive controllers move data to memory at what ever speed and chunk size they are designed to use, but your access to that data in 32bit chunks that take just as long to load as 64bit chunks is slower because it takes more clocks.
Driver controllers are DMA. The processor is not involved in each individual data word transfered.
Look, I've done this exact same thing, loaded the i386 kernel on a Pentium D (also a 64bit dual well processor, in spite of the Pentium name).
Why would results for an ancient Pentium-D imply anything for a Core 2 Duo, which is an entirely different implementation of the x86 architecture?
Since it was a new server with no mission data installed I re-installed using the correct kernel, and it was significantly faster with the 64bit kernel.
Which leaves us with the original accusation--that I made a 32/64 choice. I just stuck the disc in and installed.
I did not bench long running applications, or time long compiles, just running yast, kde, and some database loading. Since it runs at init 3 most of the time the 32bit would have worked, and might have gone un-noticed. But it was faster launching applications in 64bit, and the time to load the sql database from raw files was also faster, 37 minutes compared to 52.
I don't care about those things. I care about CPU-intensive symbolic computations for which 64-bit words, including every single memory address, are dragging around 32 bits of useless zeroes. OK? RRS -- To unsubscribe, e-mail: opensuse+unsubscribe@opensuse.org For additional commands, e-mail: opensuse+help@opensuse.org