SMP with different speed processors??? anyone try this?

[isaacmacdonald]

Ok, I'm new to the whole smp thing. I was thumbing through the manual on my mpx board and it indicates that as long as fsb speeds are the same, the multiplier can be different for each processor.

Is there a huge performance loss incurred by using 2 different levels of processors (let's say an xp2000 and a xp1600)? If I run a game, will it automatically prefer the faster processor or do I have to manually assign its process to the 2000? How about encoding divx?

input is appreciated.

thanks

 

[sechs]

First of all, I know of no desktop OS that supports assymetric multiprocessing. Windows is going to expect you to have two identical processors. This is going to muck things up.

Second, given that, one of the processors is going to be the primary. I believe that this is set simply by which socket the processor is in, but I'm not sure, as I've never owned an MP system myself.

Finally, if a program is not multiprocessor-ready, then it's only going to use one processor, therefore negating for that appplication the advantage of two.

 

[isaacmacdonald]

looks like I mistitled this thread "asymmetric smp", which is actually something completely different. has anyone had first hand experience trying to use processors with different speeds, or can anyone point me to some documentation that verifies that windows requires the processors to be the same?

 

[TenaciousPee]

Won't unlocking both processors and setting them to the same multiplier, and by consequence, same FSB settings (cause you can't run 2 processors at 2 different FSB settings period, end of story) be the same as running identical processors?

Sure, you can't run two non-identical Intel processors in SMP mode, but AMD is a whole other story.


But running them at two different multipliers, I don't know how that would effect things. It'd probably work, but it wouldn't be as efficient. Then again, it's not like the two chips talk to eachother while performing calculations. Two processors can't work on the same thread at the same time.
If my dual celeron days serve me right, you do pick a primary processor and then the secondary picks up multi-threaded applications, so Sechs is correct in that the advantage of two processors only if the application exposes that feature.


I know that LameMp3 can used multi-threading; I'm sure there are divx encoders that do so too.

 

[sechs]

Most OSes are going to expect the two processors to have the same abilities. It will want to be able to pass task x to either processor and have it perform the task the same, no matter to which it passes it.

 

[mechBgon]

I believe some BP6 and Tiger100 owners have in fact run different speeds of CPU, and even different TYPES (66MHz-FSB PentiumII + Celeron in a Tiger100, for instance). Still not something I can see any point in trying, when it's so easy to sell one or the other of the mismatched CPUs and get a matching one to take its place.

 

[TenaciousPee]

Sechs,

Most OSes are going to expect the two processors to have the same abilities. It will want to be able to pass task x to either processor and have it perform the task the same, no matter to which it passes it.

And what's the difference between the xp1600, and xp2000? Same abilities, different clock speeds.


Perhaps I'm missing something.

 

[Jeff7]

Originally posted by: TenaciousPee
Sechs,

Most OSes are going to expect the two processors to have the same abilities. It will want to be able to pass task x to either processor and have it perform the task the same, no matter to which it passes it.

And what's the difference between the xp1600, and xp2000? Same abilities, different clock speeds.

Perhaps I'm missing something.

Nope, different abilities. If one person is able to run, and another person can only walk, they don't have the same abilities.

 

[TenaciousPee]

Originally posted by: TenaciousPee
Sechs

Most OSes are going to expect the two processors to have the same abilities. It will want to be able to pass task x to either processor and have it perform the task the same, no matter to which it passes it.

And what's the difference between the xp1600, and xp2000? Same abilities, different clock speeds.


Perhaps I'm missing something.

Nope, different abilities. If one person is able to run, and another person can only walk, they don't have the same abilities.

So, lemme see if I have this correct:

If XP1600 = walk,
and XP2000 = run,
does that mean Jeff7 = teh winnar?

Running and walking aren't necessariliy two different abilities; they're the same ability (traveling in an upwardly mobile, bipedal fashion) performed at two different speeds...


Okay, okay. Your analogy was incorrect, my analogy was just plain bad... the original claim that "Most OSes are going to expect the two processors to have the same abilities. It will want to be able to pass task x to either processor and have it perform the task the same, no matter to which it passes it." was really a poor description of why SMP works in the first place. The OS doesn't see processor speed as an ability at all. It sees instruction sets, registers, pipelines etc. etc. etc. which are exactly the same for BOTH PROCESSORS!

phew! Glad that's out of the way...


Oh, I get it. You were joking.

You were joking, weren't you?

 

[Jeff7]

I guess "abilities" could mean different things here. I was looking at ability as being the overall capacity of a processor to process data, taking speed into account. You were referring to ability as the way that the processor does its work, regardless of speed.
So, no, I wasn't joking. Just "misunderestimating."

 

[sechs]

Maybe you missed that one of them has the ability to speed (that is, process threads a certain rate), while the other cannot meet this speed.

Sure, an OS can see that it has two different processors, but most aren't designed to act on that. To whichever runner it hands the batton, it expects the race to finish at the same time; by having a slow runner, you destabilize the system.

 

[Smilin]

Doesn't sound like a good idea to me.

Generally speaking you should have identical processors that even have the same stepping!

As for the asymetrical multiprocessing comment made by someone earlier: YES your OS supports it. PC's since the origianl IBM PC have used it for many many things. Your keyboard uses asymetrical multiprocessing for instance. I think you may be confused about exactly what it means.

 

[isaacmacdonald]

well, we'll see... on thursday my brand-old 1.3ghz duron comes. I also have an unlocked xp1600, that can prob run fine at 1800+ speeds.

I guess I should clarify the motivation for running two processors at different speeds. Mainly it's to alleviate cpu lockups + hickups when I'm gaming + running other background tasks that randomly demand 100% cpu usage(direct connect at times). For something like this, there's really no need to have a processor that matches the primary processors speed. I suppose it's not a common way of looking at dual processor configurations, but this computer lockup stuff has been pissing me off for years. Hmm. Hopefully the duron works )).