Intel's Penryn Benchmarked
Steve Kerrison writes "Intel's keen to show off its up-coming 45nm Penryn Core 2 CPU. HEXUS had some hands on time with the new processor to get an idea of how well it will perform once its released: 'Intel's new 45nm Penryn core adds more than just a clock and FSB hike, so much so that even a dual-core Penryn is able to beat out a quad-core QX6800 under certain circumstances.'"
If your app benefits from SSE4 optimizations, the gains compared to the current Core 2 can be giganormous (DivX encoder: +85% at equal clock). Otherwise, expect a per clock advantage of about 10%.
And you would ofcause first need to add auto SSE4 support to your compiler.
Or just use the AdBlock Plus extension, which blocks all the advert scripts without disabling any of the other scripts on the site.
It also has the advantage of blocking gif image ads on other sites that NoScript misses.
Um, ia32 could never address more than 4GB per process (and often it was 2GB). Even though with PAE you could put segments anywhere in a 36-bit address space. Most modern C compilers have no idea about "far pointers" [think back to the 16-bit days] so you're still stuck to at most 32-bits of address.
As for pointers being twice the size, yeah that's a pain. You can code around that if you know you'll be indexing something smaller than 4GB in size (hint: x86_64 can still efficiently use 32-bit registers). But if you're poking around a 10GB mmap object you just need 64-bit pointers so there is no getting around that anyways.
Since x86_64 can run 32-bit apps in long mode, you can just re-compile your app for 32-bit mode if it's absolutely not going to take advantage of the memory space or register size.
Tom
Someday, I'll have a real sig.
Wikipedia is wrong.
Disclaimer: I am an employee of Intel, but I do not speak for Intel. This post reflects my opinions and not those of Intel Corporation.
Jonathan Pearce jonathan@pearce.name
3EAAFB2A http://www.jonathan.pearce.name/