What is faster? 366@550 or 500@600?

Snugglebear · Post by **Snugglebear** » Sat Dec 07, 2002 1:15 am

For many things a 20mhz bus increase won't do much (i.e. MS Word, Access, a few newer games, etc.). However, when any app is dealing with a large dataset such as in video encoding, you want as fast a bus speed as possible.

As simply as I can put it, x86 processor caches are designed for small, desktop-sized datasets (again, Word, IE, etc.) and hence only have a few hundred KB. The celerons have only 128KB of cache, and thus plow through it very rapidly when dealing with any significant working set. Once a cache miss occurs the chip has to fetch the needed data from somewhere else in memory, and that takes an much longer time compared to a cache hit. We're talking dozens of clock cycles compared to a handful when data is in cache. To make matters worse, you're using two small-cached celerons, both running through their datasets very quickly. They both have to contend for the same bus to access memory and the delay goes up yet again.

So, in your situation you're using two 466s on a 80mhz FSB. This means the worst-case scenario is where both are hitting the bus every chance they get and dropping the effective bus to 40mhz for each chip. Running two 366s @ 550 on a 100mhz bus means worst-case is 50mhz for each chip. That extra 10mhz makes a big difference.

Lastly, let me mention something called processor affinity. With admin rights in Win2k you can set a certain application/process to run on only one processor via the task manager. Doing this allows your Divx encoding to work on only one CPU, leaving the other to do whatever else needs doing. The advantage here is that Win2k isn't constantly shifting the encoding back and forth between CPUs in an attempt to keep the system responsive. That shifting behavior causes many more cache misses, loading the bus and making things far slower.

tomatoeboy · Post by **tomatoeboy** » Mon Dec 09, 2002 10:50 am

That would appear to make sense as if you're telling the OS that there's another CPU it will try & designate some work to that CPU, Only to find that there's nothing there & then have to do it on the original chip.

lungster · Post by **lungster** » Fri Dec 13, 2002 10:11 am

An MP kernel running on an SP system will run slower because the MP kernel is compiled with all of the MP locks in place. These are the locks that would normally be used to prevent contention between the 2 processors. In an SP environment, each CPU has its own IRQL to handle preemption and interruptability. With more than one processor, there is no global IRQL scheme and so locks must be used.

As for the original question, I'd take 366@550 over 500@600. The increase in CPU speed of 50 MHz (9 % increase from 550 to 600MHz) is more than made up by the increase in bus speed of 25% (80 to 100 MHz) under most situations. Another thing to consider is that with FSB=80MHz, the AGP and PCI bus will both be quite a bit over spec and will tax anything plugged in. 366@550 is entirely within spec and much less likely to lead to bad behavior over overclocked I/O devices.

RRLedford · Post by **RRLedford** » Fri Dec 13, 2002 8:41 pm

If a MP system is converted to UP (Dual 366@500 => Single PIII/1100E) and it still shows the MP Kernel is running for WinXP Pro, does the OS have to be reinintalled to get rid of the MP kernel??

Post by **Derek** » Fri Dec 13, 2002 8:42 pm

RRLedford wrote:does the OS have to be reinintalled to get rid of the MP kernel??

I wouldn't do it any other way.

lungster · Post by **lungster** » Fri Dec 13, 2002 9:39 pm

RRLedford wrote:If a MP system is converted to UP (Dual 366@500 => Single PIII/1100E) and it still shows the MP Kernel is running for WinXP Pro, does the OS have to be reinintalled to get rid of the MP kernel??

With W2K there is a way to "downgrade" that involves copying and renaming some core system files. I'm not sure if this is possible under XP since they instituted system file protection. But like Derek says, even if it were possible, I wouldn't do it unless I was absolutely sure I can get away with it ! One wrong move and it's toast !

If it were me, I'd take a good hard look at how much work it would be to resintall all my apps and user data. Chances are, the performance difference isn't going to be all that great. If for some reason you feel the need to squeeze that last drop of speed, then you may want to look at other parts of your system as well - especially the disk drives and display card.

my $0.02