Different mulitiplier/clocks per core, is this a bad idea?

[Allio]

I have an unlocked X2 550 which I'm overclocking with PhenomMsrTweaker. It lets you set individual multipliers per core, but I've always kept them the same because that just seems like the thing to do. In fact I think I assumed that setting asynchronous clocks would probably cause Windows to barf. Then I tried it, nothing broke, and benchmarks seem to see an improvement.

Any reason not to do this? It seems pretty helpful, especially for those with unlocked cores (which are usually the weak point of the processor). In my case it looks like I can probably set core 0 and 1 to 3.8ghz+ while leaving my weaker "free" cores 2 and 3 at 3.4ghz. And in general, you could maximise an overclock like this by extracting every last bit of potential out of your quad.

Any horror stories?

 

[soccerballtux]

No reason not to. You might find some cores can go higher than others at a given voltage. Cores 3 and 4 error out faster at a given frequency than 1 and 2 do for a given voltage. It definitely did help my benchmarks a lot.

 

[tweakboy]

That is not a good idea at all. Youll get instability and crashes and it's not going to be a jolly fun experience. steer clear. Just OC your CPU to how far you can take it then leave it alone.. gl

 

[soccerballtux]

That is not a good idea at all. Youll get instability and crashes and it's not going to be a jolly fun experience. steer clear. Just OC your CPU to how far you can take it then leave it alone.. gl

Perhaps you care to explain why these instability and crashes don't happen with either my chip, or any Intel chips with Turbo mode? The cores are running at a different frequency.

OP-- there's no difference to Windows if cores are running at a different frequency, as long as the data gets calculated without error.

 

[heyheybooboo]

That is not a good idea at all. Youll get instability and crashes and it's not going to be a jolly fun experience. steer clear. Just OC your CPU to how far you can take it then leave it alone.. gl

I'm not sure I understand your reasoning.

The OS spikes different cores at varying levels on an almost constant basis.

If the OP clocks cores at varying rates and passes whatever stress tests he cares to employ, I don't think it will matter at all.

IIRC, AMD has separate power planes for cores and the memory controller, so whatever implied or perceived instability from varying cores would most likely be negated ...





--

 

[lopri]

If the delta between cores is big enough, then there can be a problem. The board's BIOS also need to cooperate with the CPU's P-states. I remember when I first got my 955 BE, my board wasn't ready for it and ran one core at 3.2 GHz and the rest at 800 MHz. The system didn't crash, but it would lag even on the desktop as Windows hops around threads.

I suggest the OP to test it with games and see if there is any unusual behavior. I don't think there will be an inherent problem, but if there is it will manifest itself.

There is a recent AT article that probe a similar topic, from which I've learned of another advantage of 'inclusive' L3 implemented in Nehalem architecture. And it's something AMD should definitely consider for its Bulldozer architecture, IMO.

That is also why "Smart Fetch" dumps the L1 and L2 caches in the L3 cache. This avoids not only waking the idle core up too soon, but it also avoids the performance hit associated with snooping a "napping" core. Intel's CPUs do not have this problem: the inclusive nature of the L3 cache means that if data cannot be found in the L3 cache, you will not find that data in any core's L1 or L2 caches.

http://it.anandtech.com/IT/showdoc.aspx?i=3722&p=5

 

[Allio]

Interesting! I guess the best thing to do is to experiment. Getting a few percent speedup in benchmarks would easily be outweighed by a lack of perceived smoothness due to P-state skullduggery. Theoretically at worst it shouldn't be any slower, as the slowest core is running at the same speed it was at my max "all core" overclock, but we'll see.

I'll give this a go and report my findings back for anyone interested

 

[RavenGuard]

That is not a good idea at all. Youll get instability and crashes and it's not going to be a jolly fun experience. steer clear. Just OC your CPU to how far you can take it then leave it alone.. gl

I'm sorry, but this is wrong. I can see how it's remotely possible with huge clock differences, but I don't see where you did your research for this.

 

[lopri]

Thread hopping in action.

http://www.youtube.com/watch?v=8mztqOxsUik&feature=player_embedded#

I laughed while watching the demo. Poor guy.

 

[CurseTheSky]

Thread hopping in action.

http://www.youtube.com/watch?v=8mztqOxsUik&feature=player_embedded#

I laughed while watching the demo. Poor guy.

Ouch. All I could think while watching that was the Windows 7 demo looked like crap. He said there were fewer horses in the Vista demo compared to 7, but it looked like the opposite on several occasions. To top it off, the Windows 7 machine was stuttering every 5-10 seconds.

I'm sure it was a fluke, but man that sucks.

 

[Cogman]

I'm sorry, but this is wrong. I can see how it's remotely possible with huge clock differences, but I don't see where you did your research for this.

Everything I know about hardware engineering says that this could potentially cause some major issues. If AMD has designed their chips/memory interfaces to work even though different parts of the chips are running at different clocks, then good on them, it wouldn't be a simple task to accomplish this (I kind of wonder what benefit they were going for as well. Collisions at the memory controller would be a major issue, there is a reason things are clocked in the first place, collision avoidance is a big one)

*edit* now that I think about it, I remember mention of this feature a while ago. It was not, however, for what you might think. AMD designed their CPUs to be able to run on different clocks so that one core could go into a power saving mode (Cool'n'Quiet) while another core could be fully pegged. Allowing for a lower power consumption even though one core was maxed out.

 

[Schmide]

They're all running off same the base clock, any variation in execution frequency will be abstracted from the core through cache latencies. (i.e. higher multipliers) The buses all run at the same speed.

 

[soccerballtux]

If the delta between cores is big enough, then there can be a problem. The board's BIOS also need to cooperate with the CPU's P-states. I remember when I first got my 955 BE, my board wasn't ready for it and ran one core at 3.2 GHz and the rest at 800 MHz. The system didn't crash, but it would lag even on the desktop as Windows hops around threads.

I suggest the OP to test it with games and see if there is any unusual behavior. I don't think there will be an inherent problem, but if there is it will manifest itself.

There is a recent AT article that probe a similar topic, from which I've learned of another advantage of 'inclusive' L3 implemented in Nehalem architecture. And it's something AMD should definitely consider for its Bulldozer architecture, IMO.

http://it.anandtech.com/IT/showdoc.aspx?i=3722&p=5

You know when mine was idling at 800Mhz it was lagging in simple things like Firefox; that's why I have it idle at 2.2Ghz now (only takes 1.07v to do so).

 

[Nathelion]

You know when mine was idling at 800Mhz it was lagging in simple things like Firefox; that's why I have it idle at 2.2Ghz now (only takes 1.07v to do so).

This is a well known issue in Windows. Supposedly 7 is better at scheduling, but at least up through Vista, the scheduler bounces threads around and essentially loads each core an equal amount. Individual core clocking naturally plays havoc with this scheme.