CPU Envy

When I first started working for SCO in 1989, Xenix was the ONLY 32 bit operating system available for the Intel 386 platform.   In fact, in 1983, when Linus Torvals was 14 years old, SCO had Xenix running on two platforms:  the Mac Lisa and the Intel 8086.   Yes,  years before Tanenbaum’s Minux was released, SCO had a AT&T System III kernel running in 640k of memory.   And Microsoft even paid them to do it!  But for that story, see wikipedia.

So when I arrived at SCO, their primary focus was still writing OS software and I was working on the X Windows port to the new SCO Unix, the successor to Xenix.  I started with a Zenith 16MHz 386 on my desk, and at the time, that was considered a lot of CPU to dedicate to one user!  But the best part of the job was being one of the first people to work on the 486 in order to get the X server running on the platform.   Of course, I didn’t get it on my desk, I had to drive over the hill to Olivetti for a few weeks (where people smoked in their cubicles!) and work on the machine there because there were so few of them.   Later on, I had the same task on one of the first Pentium systems in order to demo an X Windows application running faster than the Sun Workstations of the time.

Those were the days when it was simple to tell which was the faster Intel Processor.   You just looked at the name: Pentium, Pentium II, Pentium III, and the MHz number and it was obvious.  With the Pentium 4 we were up to 10K MIPS.   And then Intel decided it was better to confuse the market.   So now let me ask a simple question:  Rank the following Intel processors in order of performance:

Intel Core i7 820QM @ 1.73GHz

Intel Core2 Quad @ 2.83GHz

Intel Xeon X5670 @ 2.93GHz

Well you have to figure this is easy,  the Xeon line was introduced in 2001, the Core 2 in 2006 and the core I7 in 2010!   So even given the lower GHz number, the i7 must be the fastest of the bunch, right?    Turns out to be just the opposite.   The i7 is the slowest processor listed, and the Xeon is the fasted, with the Core 2 winding up in the middle.     Of course, the Xeon platform has been redesigned a few times since 2001, and if you look at this handy chart from PassMark, you’ll see that an Intel Core i7 X 980 @ 3.33GHz does now top the list as Intel’s fastest processor, but the models vary widely below that depending on everything from cache size, to bus speed. (And they haven’t listed the Xeon x7350)   It’s been fairly widely recognized now that increasing clock speed no longer gets you the performance returns it once did.

But how does this help you decide which Intel CPU to buy?  Well there’s one more piece of information you need, and that obviously is price.   In this case, the folks at pricewatch are nice enough to give us the latest CPU prices for both Intel and AMD (ignore the PassMark price, it’s already out-of-date).   So now you can see how much each of these cost.   But the thing you really want is a mash-up page (sorry, someone with more time than I needs to build it) that shows you the PassMark/$ figure.    Going back to the three I listed:

Intel Core i7 820QM @ 1.73GHz 3500/$250 -> 14PM/$

Intel Core2 Quad Q9300 @ 2.50GHz 3500/$149.00 -> 23PM/$

Intel Xeon X5670 @ 2.93GHz 9600/$1359 -> 7PM/$

So it’s interesting that with the cheaper processors, you actually get more bang for the buck than the ultra high end.   Another thing that adds to the cost is the power draw.   So I’ve been eyeballing the Q9550S, which is the low power version of the Q9550.  The lower power also means less heat, which means less fan noise, which is always a big win in a HTPC to replace my poor Intel Core2 Duo E8500 @ 3.16GHz.

The E8500 only rates 2400 on the PassMark test, while the Q9550 is at 4300.  But once again we find at $148, the E8500 is getting 16PM/$ while the Q9550S at $350 is only 12.   Now do I really need 2 more processor cores to rip all the old Disney video tapes to disk?   Probably not, but it’s not like you need 4WD on that SUV, but you sure like having it…

Leave a comment

Your email address will not be published. Required fields are marked *