AMD rumor

[DrMrLordX]

That's not the way I read it. He was agreeing with MrFanel. When he said "it" he was referring to the rumor that AMD was prepared to roll out "reverse hyperthreading" on existing AM2 silicon. It's just a nitpick, but there's no reason to attack him for agreeing with you and other posters *P

 

[Markbnj]

What about things which cant be effectively done in parrallel and take a lot of processing time, this is where i see this could be usefull.

I think we're saying much the same thing. Whether it's an embedded application like hardware video compression or some other specialized task that can't be parallelized, it's going to be useful in situations where you want to put as much power as possible to work on one thread. Pretty rare situation, but potentially useful. They might be thinking it will open some specialized markets.

 

[Avalon]

Originally posted by: Questar

Originally posted by: dmens
fanboy bs.

QFT

Man, you two just can't wait to jump on an AMD thread. Granted, I think this is also just fanboy fluff, but, whatever.

 

[F1shF4t]

Originally posted by: dmens
Then why'd you say:

LOL untill i see a hell of a lot of evidence supporting it i wont believe it.

to a guy that posted:

This will not happen. Ever.

Indicating you want evidence the technology is not present before believing in such. Whatever, guess someone needs to learn how to read and spell.

And if you're interested in a technical discussion, you might want to start with codeflow instead of "tasks", because such a machine is doing instruction-level speculation.

Yes i was agreeing with him that i DON'T believe every rumor. So ur point being is? LOL if u wish to attack me on my spelling be my guest cause i tell it outright that i cant spell for cr@p, and reading well u might need lessons urself

 

[n7]

I don't really think this is true, but at the same time, i wouldn't be entirely suprised if it did happen eventually.

I love how teh Intel fanbois have been so quick to denounce this since of course this would rape Conroe :laugh:

 

[dmens]

Yes i was agreeing with him that i DON'T believe every rumor

I misread your post. My bad.

I love how teh Intel fanbois have been so quick to denounce this since of course this would rape Conroe

Hah, got news for you kid, I "denounced" (read: dismissed) this desperate fanboy rumor because it is obvious amd (or intel) wouldn't be stupid enough to attempt this stunt given its technical requirements. Since intel has already looked at this idea, the high-level details aren't a mystery to educated folks. As for the performance of this hypothetical machine against its contemporary in say, the 2009 timeframe, that is far from a guaranteed "rape". In fact, on x86, it'll probably suck a lot of ass.

 

[SunnyD]

Originally posted by: dmens

Yes i was agreeing with him that i DON'T believe every rumor

I misread your post. My bad.

I love how teh Intel fanbois have been so quick to denounce this since of course this would rape Conroe

Hah, got news for you kid, I "denounced" (read: dismissed) this desperate fanboy rumor because it is obvious amd (or intel) wouldn't be stupid enough to attempt this stunt given its technical requirements. Since intel has already looked at this idea, the high-level details aren't a mystery to educated folks. As for the performance of this hypothetical machine against its contemporary in say, the 2009 timeframe, that is far from a guaranteed "rape". In fact, on x86, it'll probably suck a lot of ass.

Just like, in fact, Hyper Threading sucked a lot of ass on the Pentium 4's... (yup, providing an ~30% boost on some application performance is sucking ass let me tell ya).

Yeah, in retrospect, most of you Intel "fanbois" will now admit that HT was a dumb waste of an idea, because hindsight is 20/20.

But all in all, any technology that gives ANY performance boost doesn't "suck ass", as long as the end user understands what they are getting... it's a possible theorhetical performance boost on SOME applications with HT. Same for RHT. Multithreaded apps won't benefit at all out of this. SOME singlethreaded apps will get a nice boost. Just like HT did in the opposite manner.

Check your ego at the door and learn about the technology, or, just continue being a fanboy.

 

[Regs]

SunnyD, I owe you a beer.

 

[PetNorth]

Now in

The Inquirer

And

X-bit Labs

 

[Ika]

Originally posted by: PetNorth
Now in

The Inquirer

And

X-bit Labs

Inq is a horrible source. Xbit is not a whole lot better...

You'd think AMD would say more about this, since if it worked it would hurt Conroe a lot more than what AMD is doing now.

 

[darkdemyze]

Originally posted by: guoziming

Originally posted by: PetNorth
Now in

The Inquirer

And

X-bit Labs

Inq is a horrible source. Xbit is not a whole lot better...

You'd think AMD would say more about this, since if it worked it would hurt Conroe a lot more than what AMD is doing now.

My thoughts exactly. Plus AMD hasn't said anything, so nothing is official. Just more rumors and speculation..

 

[the Chase]

Well some more speculation thrown in- Supposedly Intel is also coming in with this same tech very soon....probably different name though.

 

[PetNorth]

yeah... and after that, Intel zealots will say it's possible...

 

[Aluvus]

Originally posted by: darkdemyze
My thoughts exactly. Plus AMD hasn't said anything, so nothing is official. Just more rumors and speculation..

Rumors and speculation are arguably more valuable to them right now.

 

[dmens]

Originally posted by: SunnyD
Just like, in fact, Hyper Threading sucked a lot of ass on the Pentium 4's... (yup, providing an ~30% boost on some application performance is sucking ass let me tell ya).

Yeah, in retrospect, most of you Intel "fanbois" will now admit that HT was a dumb waste of an idea, because hindsight is 20/20.

But all in all, any technology that gives ANY performance boost doesn't "suck ass", as long as the end user understands what they are getting... it's a possible theorhetical performance boost on SOME applications with HT. Same for RHT. Multithreaded apps won't benefit at all out of this. SOME singlethreaded apps will get a nice boost. Just like HT did in the opposite manner.

Check your ego at the door and learn about the technology, or, just continue being a fanboy.

SMT is a "dumb waste of time" to uneducated amateurs, and if that's your personal postmortem on P4's SMT, then you're just completely ignorant on the technical aspects of the industry.

And if you can't fathom the advantages and challenges of something as relatively simple as SMT, there's no way in hell you can understand what this new speculative execution technique entails, and hence you're in no position to tell me about "possible theoretical performance boosts" on "some singlethreaded apps", because that's just a load of bull on your end.

:beer:

 

[Fox5]

This "reverse hyperthreading" seems pointless assuming that devs learn how to properly multithread their code. Of course, PC programmers aren't the most efficient bunch in general.

 

[SunnyD]

Originally posted by: dmens

Originally posted by: SunnyD
Just like, in fact, Hyper Threading sucked a lot of ass on the Pentium 4's... (yup, providing an ~30% boost on some application performance is sucking ass let me tell ya).

Yeah, in retrospect, most of you Intel "fanbois" will now admit that HT was a dumb waste of an idea, because hindsight is 20/20.

But all in all, any technology that gives ANY performance boost doesn't "suck ass", as long as the end user understands what they are getting... it's a possible theorhetical performance boost on SOME applications with HT. Same for RHT. Multithreaded apps won't benefit at all out of this. SOME singlethreaded apps will get a nice boost. Just like HT did in the opposite manner.

Check your ego at the door and learn about the technology, or, just continue being a fanboy.

SMT is a "dumb waste of time" to uneducated amateurs, and if that's your personal postmortem on P4's SMT, then you're just completely ignorant on the technical aspects of the industry.

And if you can't fathom the advantages and challenges of something as relatively simple as SMT, there's no way in hell you can understand what this new speculative execution technique entails, and hence you're in no position to tell me about "possible theoretical performance boosts" on "some singlethreaded apps", because that's just a load of bull on your end.

:beer:

Thanks for not understanding my post. I was basically taking a jab at the "fanboys" that chant marketing gimmicks. I think you misunderstood the line where I said the <snip>Intel "fanbois" admit that HT was ...<snip> It's not what I personally believe, as the tone of my post indicated - ANY technology that produces a performance benefit will always be welcomed, unless it breaks other things to an unrepairable level.

You want MY postmortem on HT? Here:

Intel Marketing back in the day wanted you to believe that HT was the holy grail. Unfortunately, the software market wasn't there for it, and the relatively few applications that could use it were mostly professional in nature until more "common" applications were re-written to take advantage of HT. Unfortunately, it was too little too late, as Intel itself fell off the HT bandwagon it built. Since most HT aware applications perform equally as well or better on a dual processor environment, more resources were spent there. Intel also had to react to competition. It was at this point that Intel Marketing said HT is dead, long live dual-core.

Well, HT isn't dead, it's just hiding until it is needed again (it still exists in all current Intel chips, including Conroe from what I understand - just disabled, but why?).

HT was not a BAD technology. It is something that end users simply did not understand. When these users hear Joe-Marketing Department say "CAN give you 30% performance increase", these users actually comprehend it as "WILL give you a 30% performance increase, out of the box, even with the computer not turned on..."

That second phrase is what most people get stuck on, just look at the bulk of this thread.

 

[Viditor]

After skimming most of these posts, I can see that there hasn't been enough reading done here.
1. Intel's Mitosis is a very similar project...
2. Most of the work for the "reverse HT" is actually done by a new compiler
3. Intel will probably be ahead of AMD in this as they have one of the finest group of compiler writers on the planet...however I suspect that AMD will have the hardware portion already in place as well.

For more info:
Mitosis and Speculative Threading

"Mitosis relies on both hardware and software (compiler) support to work"

BTW, it's not BS and the main hardware change you'd see on the die is the buffereing...

 

[Fox5]

Well, HT isn't dead, it's just hiding until it is needed again (it still exists in all current Intel chips, including Conroe from what I understand - just disabled, but why?).

If it's there but disabled, they either feel they have enough of a performance advantage now where it's not needed or it doesn't work right yet.

2. Most of the work for the "reverse HT" is actually done by a new compiler

If an application can be reverse HTed, shouldn't it also be possible to multithread it? Or does it do something more abstract than just seperating 1 thread into 2, such as having the 2nd cpu perform probability analysis for the chance that certain code will be encountered on the first cpu?

 

[SunnyD]

Originally posted by: Fox5

2. Most of the work for the "reverse HT" is actually done by a new compiler

If an application can be reverse HTed, shouldn't it also be possible to multithread it? Or does it do something more abstract than just seperating 1 thread into 2, such as having the 2nd cpu perform probability analysis for the chance that certain code will be encountered on the first cpu?

Reverse Hyperthreading is not taking a Hyperthreaded application and removing the HT-ness from it.

RHT cannot be compared to HT in any way. What RHT will do is abstract the CPU hardware of two or more individual CPU's (processor, core, CPU) into a single virtual processor. Then, as software dispatches instructions to the CPU, the underlying virtual layer will take the instructions and dispatch them to each individual processor. What this does is allow for instructions to happen in parallel. What does this buy you? In single-threaded applications which only need and only use ever one CPU, this allows the "virtual" processor to do "twice" the work of a single physical processor by literally doubling resources.

The word "twice" is in quotes because due to overhead, management and synchronization issues, you will never get a full 100% increase in benefit out of the virtual processor. Expect to see significant gains, but like HT, I wouldn't plan on more than 30% in best case situations.

Where does RHT shine? Single threaded linear applications, and possibly single-threaded applications with excessive needs for branch-prediction, where the virtual processor could literally use one processor to process one branch path fully while at the same time processing another branch path fully, then choosing the correct path with the result waiting in half the time. Of course the latter example can be done in a multithreaded environment just as easily, if not better.

 

[VisionxOrb]

I belive the Alpha EV8 Have SMT technology and AMD has most of the alpha engineers. Therefore Id imagine AMD will have it in there CPU before intel does but from what ive heard it requires a groundup new proc so wont be seen yet, but who knows.

 

[dmens]

Originally posted by: SunnyD
Reverse Hyperthreading is not taking a Hyperthreaded application and removing the HT-ness from it.

Dude, thanks for your post-mortem of P4 HT, but you're totally confused on the technical details. There's no such thing as a "hyperthreaded application", you don't need to rewrite software to take advantage of SMT, and marketing never took part in its inception, implementation and deployment. Not to mention SMT not even mutex with multicore, you can have both.

And you're confused with this new concept as well, which I assume to be intel's mitosis and the amd equivalent which is rumored to exist on AM2 and/or K8L (but certainly does not). I am so confused by your post I can't be sure what you are referring to. There's no software involved in intel's mitosis other than the fact that static optimizations would greatly benefit such a hypothetical machine, and the aforementioned machine is just a single processor to begin with, not multiple cores.

Please read Vidtor's link before further speculation, thanks.

 

[SunnyD]

Originally posted by: dmens

Originally posted by: SunnyD
Reverse Hyperthreading is not taking a Hyperthreaded application and removing the HT-ness from it.

Dude, thanks for your post-mortem of P4 HT, but you're totally confused on the technical details. There's no such thing as a "hyperthreaded application", you don't need to rewrite software to take advantage of SMT, and marketing never took part in its inception, implementation and deployment. Not to mention SMT not even mutex with multicore, you can have both.

And you're confused with this new concept as well, which I assume to be intel's mitosis and the amd equivalent which is rumored to exist on AM2 and/or K8L (but certainly does not). I am so confused by your post I can't be sure what you are referring to. There's no software involved in intel's mitosis other than the fact that static optimizations would greatly benefit such a hypothetical machine, and the aforementioned machine is just a single processor to begin with, not multiple cores.

Please read Vidtor's link before further speculation, thanks.

Whatever you say dude. You're always right, and the rest of the world is totally whacked. :thumbsup: :roll:

 

[dmens]

Originally posted by: SunnyD
Whatever you say dude. You're always right, and the rest of the world is totally whacked. :thumbsup: :roll:

LOL, in this case, heck yeah I'm right, and *you* are whacked. I only do this stuff for a living...

 

[SunnyD]

Originally posted by: dmens

Originally posted by: SunnyD
Whatever you say dude. You're always right, and the rest of the world is totally whacked. :thumbsup: :roll:

LOL, in this case, heck yeah I'm right, and *you* are whacked. I only do this stuff for a living...

As do I... your point? Besides, do you exactly know how AMD's "Reverse Hyperthreading" even works? And also, if you'd ever bother to even comprehend (since supposedly you should already know how to read) you'd realize that my description has already encompassed everything Mitosis does.

So, with that said, why don't you stop talking out of your ass and say something useful for once.

I'm done here. Once you enter a thread, that thread becomes worthless from that point forward.