News AMD developing reverse hyperthreading?

http://www.bit-tech.net/news/2006/04/17/amd_reverse_hyperthreading/

 

O_O

Imagine the preformance with the X2 5000+!

Or dual Opterons..

 

Seems a bit against the flow... Everyone is going dual... But in single threaded app's, it would be awsome. Just some on the fly on/off switching would be needed I guess.

So nice idea, but needs a lot of perfecting and fine tuning. But I am confident the people at AMD will pull this off

 

Intel have been working on speculative threading for sometime, so you have to believe that there are serious issues in implementing this - after all how many tasks would truly be able to be split up? Is it a wste of die space? intel worked on speculative threading since the p4 introduced hyperthreading, and its still nowhere in sight. It may be an idea that in theory sounds good, but in practice yields few benefits. Why add x ammount of transistors to yield a small performance increase in a very limited number of tasks where you could use that same transistor count to improve performance accross the board (wider execution paths, more cache etc.)

 

Because it shouldn't be a small performance increase having 2 cores work as 1, it'll be at least 50% boost imo, and having a small execution set add-in for doing that in single-threaded apps would be a great idea, as there are still a hell of a lot of single-threaded apps which need some serious crunching time.

 

migrating single threaded aps to multi threaded where possible should be done by the developers, on a thread level judging what threads can be split (where there are no dependencies etc.) is very difficult. Few tasks could be split, and even those that do get split may face dependency problems and would be a waste of cpu time. The code should be made as parallel as possible, and whilst some single threaded code may be able to be split and see a benefit, most single threaded code would see no performance improvement and possibly a penalty. speculative threading is a stop gap until code is properly optimised, and I just think that instead of spending money implementing it, everyone would be better served if this time and effort is put into actually threading and optimising the code where possible.

 

Why do people keep comparing the Conroe to the AM2.... Conroe is a complete architecture revamp. AM2 is just a socket change, why do people keep thinking that just adding in DDR2 suport for curent generation AMD processors is going to do anythign for performence, when the architecture of the chip isnt designed to take advantage of it.

as far as the "anti-hyperthreading" i agree with whats been said above. in order to make it useful it should be able to switch on the fly, because onlder games will see a benifit. but all the anticipated PC games are going to want to use DUal cores. however, switching on the fly will be pretty useless unless they develop a processor for the processor to calculate which games are using one, or two processors. but it doesnt seem liek somthing that would be able to be done on a harware level. and on a software level. it would be alot better to use the Dual cores as dual cores and not as a single.

 

you need quad cores where you can switch on anti hyperthreading to make it a set of 2 cores running off all 4 ;D

 

surely it can't be too big an expense in terms of die space to have the processor be able to split a thread in two, with shared cache it shouldn't be too big a leap surely? esp with the processors having hyper transport to communicate with, 50% increase in performance would be 25% better than conroe atm... doesnt take a genius to see it has its benefits.

It would also make dual core more attractive in general, just imagine an fx-60 with this, it rules the roost as it is, but with another 50%... wow!! even if it was only 25% it would still be amazing.

splitting threads in the processor takes some of the onus off of programmers, esp with regard to companies that don't have big budgets.

sounds like a really good idea, and with quad core and upwards it really does make a hell of a lot of sense. i mean even if u had one core dedicated to the calculations required for this out of 4 the improvements in single threaded would be immense!!

huzzah for amd!!!

 

I would believe that AMD would try to make the processor only switch to anti-HT, when single threaded operations are made. However, this technology would probably die off once most, if not all programs and games are rewritten or published with multi-threaded support.

However, like 64-bit, I'm guessing that actual consumer adoption of it, will probably be slow (maybe a bit faster, but still slow). Everyone will have the hardware for dual core, but hardware is useless without the proper software to back it up.

 

Super Pi times of 15-20 seconds?

I sure hope so.

 

I like the idea in theory. As its been pointed out, many apps are single threaded and some will prob never be multi-thread as they work better with one. If you can switch an AMD x2 between multi and single-threads then it would be a HUGE performance it. You'd get the best of both worlds assuming its implimented correctly.
And once all the software has migrated to multi-thread then you can just stop the switching.
As many have pointed out, hardware is useless without the proper hardware. But on the other hand, unless the hardware is out then there isnt any point in software.

Chicken and the egg it seems.

 

Yea cause I spend ALL day just using Super Pi. Infact, I have no other programs installed.

 

Hey, it's a good thing for all the Pi lovers out there.

 

Sounds likes a good idea, it really does. However, got some issues regarding 'on the fly' changing between splitting the application up.

Do you guys think this will be like HT where you can see two procs in Task Manager? It just seems more likely that you have to enable it in the BIOS and then you see a single core in windows (or linux or whatever).

Would be a good way to keep tabs on Intel until they get a new architecture involved.


(sorry about the spam...)

Weebl and Bob will be happy then!!!

Despite it being a typo, it's the best spelling of 'only older' I've ever seen!!

 

Some AAPS are imposible or have very little performance advantage from multithreading. This sounds awsome if this works well then AMD will just trample all.

 

I'd stick to dual and quad cores... I just don't see the point of a 10 Ghz core which actually excists out of 2 or 3 or 4 cores when you have 4 cores. 4 cores that could be split with hyper treading and then do 8 different things at the same time instead of an extremely fast single core tht does 1 thing very fast. I say 8 different things at the same time is better then 1 thing very fast.

 

Wow the brilliant explanation we were all waiting for, thank you for enlightening us all.
Great job Einstein

 

I think this is a great idea. I've been wondering why they haven't figured out how to use many cores or processors as one. Mad props to AMD on this one. Not only will this make ALL existing programs faster but programers don't have to program for more than one core, saving time and money.

 

Is it me, or is there just so much not right with this?

Unless you only need to do 1 thing of course, then you'd have a quad system, working at 1/8th of his power.