AMD Working on Reverse Hyperthreading

[lil_jimmy_norton]

This would be awesome if true. May be enough to top Conroe.

http://www.geek.com/news/geeknews/2006Apr/bch20060417035855.htm

Pretty much tricking the system to use both cores as one. That way singlethreaded apps would be able to utilize both cores.

If this is true it could possibly crush Conroe...

For more info just google "Reverse Hyperthreadin" and you'll get a ton of stuff.

[imported_Mrt1989]

Re: AMD Working on Reverse Hyperthreading

This is good news :) Put they need to trop prices also. new conroes are cheap and they clock like crazy.

[lil_jimmy_norton]

Re: AMD Working on Reverse Hyperthreading

This is good news :) Put they need to trop prices also. new conroes are cheap and they clock like crazy.

They are dropping prices.

http://www.totalbf2.com/forums/showthread.php?t=79046

There is a rumor (which I am highly sceptical about) saying that this revers hyperthreading is already in place in current AM2 processors. But like I said, I highly doubt that.

More on the RUMOR here.

http://www.xtremesystems.org/forums/showthread.php?t=103636

[Blak_Phoenix]

Re: AMD Working on Reverse Hyperthreading

Very clever.. and logical..

[Sir. MatDef]

Re: AMD Working on Reverse Hyperthreading

I read about this a while ago, and it's one of the best things I've heard recently. This definately could help amd retain the performance crown (If they can get it to market fast enough).

[imported_Wreck]

Re: AMD Working on Reverse Hyperthreading

I read about this a while ago, and it's one of the best things I've heard recently. This definately could help amd retain the performance crown (If they can get it to market fast enough).

Yeah, the rumor has been around a while, but the real question is how much would this effect performance in the real world.

[TokenBrit]

Re: AMD Working on Reverse Hyperthreading

I dont see how this could possibly work so until somebody with a background in proc architecture writes an informed discussion on it I am calling BS.
Edit:
In particular the thought of this appearing on K8 seems even more ridiculous.

[Sir. MatDef]

Re: AMD Working on Reverse Hyperthreading

I dont see how this could possibly work so until somebody with a background in proc architecture writes an informed discussion on it I am calling BS.

Call it what you want, but it is possible and amd is working on it, the only question is when it's going to come out and whether or not amd already has the tech built into their am2 procs.

[lil_jimmy_norton]

Re: AMD Working on Reverse Hyperthreading

I dont see how this could possibly work so until somebody with a background in proc architecture writes an informed discussion on it I am calling BS.
Edit:
In particular the thought of this appearing on K8 seems even more ridiculous.

Why can't it work. If you can split 1 physical into two logical ones (ala. HyperThreadin) there should be no reason you can't do the reverse.

[i made you]

Re: AMD Working on Reverse Hyperthreading

This will definitely soften the blow and buy AMD the time it needs for K8L etc release. For the time being, it will be very useful for current applications. It's a very good move by AMD because they took advantage of Intel's blind arrogance.

[TokenBrit]

Re: AMD Working on Reverse Hyperthreading

The reason this will not be on AM2 is it requires a change in architecture while K8 remains unchanged apart from a new memory controller.
Apart from the fact that, theoretically it makes no sense. An x86 instruction is a discrete object that cannot be blindly seperated into constituent parts. Typically it involves the linear processing of a set of instructions i.e one must follow the previous one. An instruction cannot be processed on two procs.
An x86 thread is again a linear sequence of instructions. Typically one must be complete before the next can be processed. However a sufficiently intelligent software compiler can realise certain instructions are independent and split them into multiple threads. This is the definition of multi-threading. It is possible to force this with hardware but it effectively means a new core must be added, merely to figure out just how to split the thread, but this requires MAJOR ARCHITECTURAL CHANGES and even then it won't be anywhere near as efficient as proper multi-threaded software, and it STILL wont be RHT, it will just be some bizarre form of multi-threading.
The time spent figuring out just how to split the thread would probably be better served just processing it.
There are other possibilities, it is possible for in a sense a proc to guess what instructions may be required and to advance process them for when the thread needs them, but again this is not efficient.

The reason HT works but RHT wont is simple and it is entwined in the discrete nature of processor instructions.
Say you have two kids(instructions) and both want to ride on your shoulders(proc).
You can carry one, then the other (single thread).
You can carry one, and get a friend to carry the other (multi-thread).
You cant carry half of one each, then the same again (RHT).

Single thread applications are not designed to be split and while to brute force them with AN ENTIRE EXTRA CORE is possible it is not really efficient, and it is certainly not a part of K8. Indeed by the time anybody got the hang of this, even if it were possible to make it efficient you would expect multi-thread apps to be common anyway.

[i made you]

Re: AMD Working on Reverse Hyperthreading

I never knew you could relate RHT to two guys carrying a kid. Awesome logic! This is almost as good as the "I don't know if you AMD guys knew this, but you can overclock a processor to infinity" argument.

[wootoow]

Re: AMD Working on Reverse Hyperthreading

The reason HT works but RHT wont is simple and it is entwined in the discrete nature of processor instructions.
Say you have two kids(instructions) and both want to ride on your shoulders(proc).
You can carry one, then the other (single thread).
You can carry one, and get a friend to carry the other (multi-thread).
You cant carry half of one each, then the same again (RHT).

Where's the option of carrying both kids on seperate shoulders?

[Flanker]

Re: AMD Working on Reverse Hyperthreading

TokenBrit, that has to be the stupidest example I've ever seen.

RHT doesn't divide up the instruction. It works by sending every other instruction to the second core. So let me remake your example. For one, there are never just two kids waiting. There are millions, even billions. Also, in a multicore system, both of you would not be carrying a kid, unless there were two lines of kids (two threads). 90% of our apps, and 99% of games, are single-threaded, so I'll stick to that. Now, in a multicore system, one of you would be standing idle while the other carried the line of kids across. It is in RHT that both of you would be carrying kids across. You would take one kid, and your friend would take the next kid in line, then you would take another, and so on and so forth. You would each be carrying every other kid.

Jummy, its very possible that its already built in. You don't need very special hardware to make it happen. At most, it will require a Windows update and a BIOS flash.

Here is the same thing on The Inquirer.

Wreck: Performance won't double, thats for sure. The general concensus from all major forums is that the performance increase would be 66%. It should be very close to what we see when there is a Dual Core processor in a Dual Core-supporting game. There have been such benchmarks, and if you saw them, you would know that a X2 3800+ performed like (roughly) a 3.2 to 3.4Ghz Single Core A64.

--------------------

This would be great if Intel didn't have the exact same (or VERY similar) thing. Like I said, its a very basic and easy to implement concept, and it is almost guaranteed that it is built into Conroe. Intel's "BadAxe" motherboard has an option to enable "Core Multiplexing" in it's BIOS. Core Multiplexing is almost exactly what Reverse Hyperthreading is, and it has been in the works for years (obviously). Since it is in the BadAxe BIOS, thats all the confirmation I need.

Where's the option of carrying both kids on seperate shoulders?

I was wondering that too. Technically, its already there with the 3- and 4-issue cores. Meaning that you can, metaphorically, carry 3 (A64) and 4 (Conroe) kids on your shoulders. With RHT, AMD would be ale to carry 6 kids on it's shoulders, lol. With Core Multiplexing, Conroe would be carrying 8. Hope those kids don't weigh much, lol. :p

[TokenBrit]

Re: AMD Working on Reverse Hyperthreading

You think you can just give alternate instructions to different procs just as easy as that?
:laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh: :laugh:

What you actaully seem to be talking about is speculative threading which is something quite different, and has been worked on for some time by many companies/universities and certainly not exclusively AMD. Importantly it CERTAINLY requires a change in architecture. It seems impossibly unlikely that any K8 chip will display this property.