Intel IDA vs. rumored AMD reserse HT?

[kknd1967]

Now intel proposes this, first
differnt idea but similar goal

HKEPC

 

[DrMrLordX]

I wonder if this is Intel's mitosis?

 

[F1shF4t]

So they keep one core idle and overclock the other when dealing with single threaded tasks.
Is it just me or is that just stupid
On the other hand u will get cpus which are guaranteed to overclock a certain amount

 

[SexyK]

Originally posted by: Dark Cupcake
So they keep one core idle and overclock the other when dealing with single threaded tasks.
Is it just me or is that just stupid
On the other hand u will get cpus which are guaranteed to overclock a certain amount

Actually it makes perfect sense to me. You get the benefits of multi-core for threaded apps and the benefits of a faster single core on apps that aren't SMP aware. This technology would put an end to the debate "should I get dual-core at 3.0GHz or quad-core at 2.6GHz?" because you could have the benefits of both in one chip. I'll take one, for sure.

 

[Viditor]

Originally posted by: DrMrLordX
I wonder if this is Intel's mitosis?

No...Mitosis is a design that allows the compiler to optimise the single thread code for parallel processing. IDA just shuts off one core and increases the clock of the secnd core.

 

[myocardia]

Originally posted by: SexyK
Actually it makes perfect sense to me. You get the benefits of multi-core for threaded apps and the benefits of a faster single core on apps that aren't SMP aware. This technology would put an end to the debate "should I get dual-core at 3.0GHz or quad-core at 2.6GHz?" because you could have the benefits of both in one chip. I'll take one, for sure.

Unfortunately, I haven't heard of them moving this tech to desktop processors yet. It would definitely make their quad-cores more attractive, I would think. You know, four E6700 cores, while running SMP-enabled apps, and an X6800 while running everything else.

 

[DrMrLordX]

Originally posted by: Viditor

No...Mitosis is a design that allows the compiler to optimise the single thread code for parallel processing. IDA just shuts off one core and increases the clock of the secnd core.

Okay, thought so. Makes me wonder why Intel is even bothering with this tech. I guess it'll work well for non-overclockers. This merely confirms what many of us overclockers already know, though, and that is that most, if not all modern CPUs are essentially shipped underclocked. You can always pick up a few mhz even at stock vcore.

 

[swtethan]

so... buy a t7200 now or later?? OMG!!!!

 

[Furen]

Let me see... instead of shutting down one core to save power it'll shut down one core to overclock the other one (which will increase the power draw of that particular core). Unless the "turbo bin" speed up is significant then I doubt it'll make much of a difference performance-wise, and it won't help power consumption either. Also, I wonder how this will actually work, since windows does seem to like changing CPU cores without prior notice...

 

[Viditor]

Originally posted by: Furen
Let me see... instead of shutting down one core to save power it'll shut down one core to overclock the other one (which will increase the power draw of that particular core). Unless the "turbo bin" speed up is significant then I doubt it'll make much of a difference performance-wise, and it won't help power consumption either. Also, I wonder how this will actually work, since windows does seem to like changing CPU cores without prior notice...

Agreed...I was wondering about that last point myself. I guess it's possible that Intel could write a patch. They can do some pretty slick code writing at Intel, but it sure seems that it would be an awful lot of work for a minimal gain...

 

[myocardia]

Originally posted by: Viditor
Agreed...I was wondering about that last point myself. I guess it's possible that Intel could write a patch. They can do some pretty slick code writing at Intel, but it sure seems that it would be an awful lot of work for a minimal gain...

But it isn't a minimal gain. It's a multi-billion dollar gain, the only kind that matters to any American company.

 

[Avalon]

Who needs IDA when you can do it yourself?
Yes, I realize this is for laptops, I'm just kidding.

 

[Keysplayr]

Originally posted by: Dark Cupcake
So they keep one core idle and overclock the other when dealing with single threaded tasks.
Is it just me or is that just stupid
On the other hand u will get cpus which are guaranteed to overclock a certain amount

Not stupid. Makes perfect sense. For apps that can only use 1 thread, it makes more sense to crank up one of the cores since the app cannot benefit from dual/multi core CPU's.

 

[DrMrLordX]

Originally posted by: keysplayr2003


Not stupid. Makes perfect sense. For apps that can only use 1 thread, it makes more sense to crank up one of the cores since the app cannot benefit from dual/multi core CPU's.

At least until they get mitosis working right?

 

[myocardia]

Originally posted by: DrMrLordX

Originally posted by: keysplayr2003


Not stupid. Makes perfect sense. For apps that can only use 1 thread, it makes more sense to crank up one of the cores since the app cannot benefit from dual/multi core CPU's.

At least until they get mitosis working right?

Which is still in the pipedream category, as far as I know. Not that it wouldn't be very, very nice to get close to SMP performance from single-threaded apps.

 

[yanman]

Given a laptop's tightly controlled thermal output, slowing down one core and speeding up the other seems, while pretty simplistic, a fairly good way of getting better single thread performance but for the desktop sphere it seems pretty cludgy. Take an overclocked system - both cores are already being pushed. Declocking one core and clocking up the other one even further may not even be possible if they've been pushed to the limit of their stable operating speed.

Bring on Reverse-hyperthreading and hopefully Intel will be able to copy it

 

[Cooler]

The real question is it A FSB increase or a exta unlocked multiplier. If its unlocked it would basicly give you a higher speed grade with simple bios change.

 

[myocardia]

Originally posted by: Cooler
The real question is it A FSB increase or a exta unlocked multiplier. If its unlocked it would basicly give you a higher speed grade with simple bios change.

Glad you caught that. It had slipped by me. Yeah, there's no way they'd use FSB. It would have to be either a +1 or +2 multiplier increase.

 

[aka1nas]

Wouldn't this hugely increase the amount of validation and testing they would have to do for these parts? Also, how do you determine which core gets overclocked? It's not always the case that both cores can hit the same clockspeed at a given voltage, etc. If every core can hit these speeds at acceptable TDP, than why not just bin them higher? Having independent power management of the cores makes a lot of sense, especially in a mobile chip. The overclocking one core part seems like it is unneccessarily complicated.

 

[Roguestar]

Originally posted by: Furen
Let me see... instead of shutting down one core to save power it'll shut down one core to overclock the other one (which will increase the power draw of that particular core). Unless the "turbo bin" speed up is significant then I doubt it'll make much of a difference performance-wise, and it won't help power consumption either. Also, I wonder how this will actually work, since windows does seem to like changing CPU cores without prior notice...

It makes sense to me. And windows hasn't had a problem with speedstep and other performance-trade systems before, surely this is just the reverse of turning down clockspeed to save power, and windows has no problem with that.

 

[zephyrprime]

Originally posted by: Furen
Let me see... instead of shutting down one core to save power it'll shut down one core to overclock the other one (which will increase the power draw of that particular core). Unless the "turbo bin" speed up is significant then I doubt it'll make much of a difference performance-wise, and it won't help power consumption either. Also, I wonder how this will actually work, since windows does seem to like changing CPU cores without prior notice...

This is for notebooks which have a large overclocking headroom because they run at low clock speeds to begin with.

 

[MDme]

one problem with this approach is for us who OC. since we all OC via the FSB (except for the FX, X6800). here's why:

1) when we OC, we usually OC near the limit for the particular chip (unless we are FSB limited).
2) if we are already running at the near max clock and then this IDA increases the cpu multi, it may get to a clockspeed that is unstable.
3) if intel did this via fsb overclocking, then same thing applies since we OC via the FSB, any further increase in FSB might just make it unstable.

again this is for us who overclock our CPUs (essentially unlocking the hidden headroom in the chips). this may work though for those who don't OC.

 

[myocardia]

Originally posted by: aka1nas
Wouldn't this hugely increase the amount of validation and testing they would have to do for these parts? Also, how do you determine which core gets overclocked? It's not always the case that both cores can hit the same clockspeed at a given voltage, etc. If every core can hit these speeds at acceptable TDP, than why not just bin them higher? Having independent power management of the cores makes a lot of sense, especially in a mobile chip. The overclocking one core part seems like it is unneccessarily complicated.

Not if they give it slightly more vcore, when it happens. This would be extremely easy to do, since it's all controlled by the BIOS, just like the downclocking that all new chips do.

 

[ForumMaster]

hopefuly this won't be needed. it's an interesting idea, but since multicore is the future, hopefully all software will be multithreaded.