AMD and HyperThreading...

A

absolut][krypkn0t

Weaksauce
Joined
Mar 22, 2002
Messages
64
Ok... so if these guys can copy eachothers technology and not have to worry about patent infringement... why hasn't amd done this... it's would give them that final edge they need (IMO) to finally trump those damn other guys. I just don't understand it... has there been talk? I don't follow this too closely so please forgive me if there has been tons of talk about this issue already... and if there has... just point me in the direction i need to go to read up on it... or do they have something else up their sleeve they are just waiting to pull out when the timing is right?
 
My guess is, maybe trying to put some sort of HT technology into their CPUs will drive the prices of their CPUs up? And well AMD is known for being cheaper. I wouldn't want them to trump over intel (the other guys). Maybe just pull ahead alittle in the industry, but not thump them.. this great competitoin we see between them benefits us all.. wether you're an AMD fan boy.. which you sound like... or and Intel Fanboy.. or you just go with the CPU that fits your needs and wants, like me.

I'm not bashing AMD for I use AMD. 2500+ barton :)

That, or maybe AMD doesn't think HT will be taken advantaged of by programs and stuff.. so why waste the money now?
 
I'd say that the new 939's are gonna do enough damage to intel as it is. yay! :p
 
Adding HT would require an entire re-design of the entire execution pipeline, and it's not worth it. AMD has already shown damned fine performance at massive clock defecits. Yes, the responsiveness would be nice, but if you want SMP get SMP, don't play with SMT.
 
AMD wouldn't benefit much from HT, intel benefits mainly because of the long pipelines on the p4. besides, both companies are planning to go dual cores next year, so HT becomes something of a moot point. (although it is likely that intel will have 2 ht cores on their dual core chips= 4 virtual procs. that chip would redefine bandwidth hungry)
 
mwarps said:
Adding HT would require an entire re-design of the entire execution pipeline, and it's not worth it. AMD has already shown damned fine performance at massive clock defecits. Yes, the responsiveness would be nice, but if you want SMP get SMP, don't play with SMT.
Click to expand...

true... i hear ya there... good thoughts guys... and your right rancid the competition is very good... for EVERYONE... by trump them i didn't necessarily mean take over the #1 spot for size of company and market share... i don't think that will ever happen... besides being second makes you work harder to be number one anyways... and i'd be happy just knowing that all the other eejits out there just "think" they have the best processor...

time for a new sticker "common sense inside" haha
 
absolut][krypkn0t said:
soo... is there a way to run dual FX's on an opteron board? hehe
Click to expand...
Not without somehow unlocking the missing HyperTransport links needed for chip-to-chip communication, and communication between the memory pools.
 
i asked AMD guys this at the tech tour. basically they see HT as just a gimmick that not only gives only a little benefit to SMP-enabled apps, but it also makes single-processor apps SLOWER.
 
kronchev said:
i asked AMD guys this at the tech tour. basically they see HT as just a gimmick that not only gives only a little benefit to SMP-enabled apps, but it also makes single-processor apps SLOWER.
Click to expand...

which is why some overclockers rather use a 2.4a instead of a 2.4c because the 2.4a overclocks much higher but doesnt have HT.
 
Yeah.. very little reason for AMD to waste time with HT when they've got dual-core CPUs on the drawing board. I'm really waiting for a dual-core consumer CPU...

Hyperthreading is really more of a way for Intel to undo some of the damage from their long-ass pipelines & get more execution units working simultaniously.
 
acascianelli said:
which is why some overclockers rather use a 2.4a instead of a 2.4c because the 2.4a overclocks much higher but doesnt have HT.
Click to expand...
the reason has nothing to do with HT, it is because the 2.4A is a prescott with a 133mhz fsb and an 18 multiplier. these give it rediculous 1:1 overclocking overhead. (200mhz fsb is 3.6 ghz, 4 ghz is easily in reach w/o stretching the nb or ram too much, plus the prescott overclocks like mad once you get decent cooling on it) too bad it still performs worse than northwood. oh and btw, HT is an optional feature, you only need to use it if you want to, and it does make the machine multi-task more smoothly IMHO.
 
absolut][krypkn0t said:
Ok... so if these guys can copy eachothers technology and not have to worry about patent infringement... why hasn't amd done this... it's would give them that final edge they need (IMO) to finally trump those damn other guys. I just don't understand it... has there been talk? I don't follow this too closely so please forgive me if there has been tons of talk about this issue already... and if there has... just point me in the direction i need to go to read up on it... or do they have something else up their sleeve they are just waiting to pull out when the timing is right?
Click to expand...

To the best of my knowledge HT is not included in any of the cross-liscnse agreements. Those only encompase the SIMD extensions.
I believe AMD has some intellectual property patents for SMT (as does practically every major chip manufacturer...), so they don't need to buy the idea from Intel, their implementation however is propriatary.

But anyway, as people has said, it's not a simple thing to implement. It didn't take 20 years to go from concept to silicon because it was a bad idea (thoiugh obviously you do need a fully pipelined and/or superscalar core to use it).
While HT does provide a benifit to true multithreaded apps, those benifits are less (how much less is certainly a matter of debate, and of code) on a shallow and wdie pipeline like AMD has.
 
Some real good points brought up in this thread and here's something else to ponder on while waiting for Intel's future dual core proc; the present Dothan proc, which is what the future dual core proc is supposed to be based on, also doesn't support HT either. And I imagine that is becasue it uses a much shorter pipeline like AXP or A64 than a long ass pipeline like P4. If Intel wants to include HT into their future dual core proc, then they will have to totally rework the present Dothan core to do it like mwarps has previously pointed out, talking about AMD procs.
 
AFAIK the cross licensing agreements encompass the x86 architecture and all extensions of it, that is how Intel can use x86-64 and AMD can use x86 at all, so HT would be crosslicensed, but there is no reason for AMD to implement it.
also we are gonna see dual prescotts before we see dual dothans
 
bobsaget said:
I'd say that the new 939's are gonna do enough damage to intel as it is. yay! :p
Click to expand...

they will, once the rest of us can afford them. once could build an entire system for the price of just one chip.
 
You're right, I mispoke. To my knowlegde AMD hasn't liscensed HT to date, but under the long standing agreements, they would be allowed to, providing they pay fair market value.
People remember AMD got SSE(1 and 2) and Intel got 3D Now Enhanced and seem to think they just share what the invert freely, for better or worse that's not the case, 3D Now was payment for SSE and vice versa.
Intel picked up the x86-64 license back in (i think) the last big patent swap in late 2001.
I have not heard of AMD picking up HT, so I'm pretty certain if they wanted to use anything under the relevant patents from Intel they will have to cough up some cash, or some patents of equal value.
 
At the AMD Tech Tour I went too, they said they will never impliment Hyper Threading because it's virtual dual core...they're doing the real thing, they're gonna create cpu's with actual dual cores...
 
Merlin45 said:
the reason has nothing to do with HT, it is because the 2.4A is a prescott with a 133mhz fsb and an 18 multiplier. these give it rediculous 1:1 overclocking overhead. (200mhz fsb is 3.6 ghz, 4 ghz is easily in reach w/o stretching the nb or ram too much, plus the prescott overclocks like mad once you get decent cooling on it) too bad it still performs worse than northwood. oh and btw, HT is an optional feature, you only need to use it if you want to, and it does make the machine multi-task more smoothly IMHO.
Click to expand...

Yes...D-stepping 2.8A currently holds the world's highest oc...somewhere around 5.5GHz. But doesn't HT require more transistors so wouldn't that lead to a bit more heat?
 
Who cares, ht sucks. Next thing we'll know Intel decides to add "extreme multiplier pro and a half", that lets the processor use multipliers by increments of .25. Go figure why it's useful.
 
DaveX said:
Yes...D-stepping 2.8A currently holds the world's highest oc...somewhere around 5.5GHz. But doesn't HT require more transistors so wouldn't that lead to a bit more heat?
Click to expand...
Ht is present in all northwood, willamette, and prescott procs, the only difference is whether or not it is in use. it can be disabled hence having a chip with it disabled for you is no advantage. the reason that the A prescotts OC so well is their high multipliers combined with the fact that once heat is removed as a constraint, they OC like crazy.
 
Ive never noticed any difference from my 3.0c to my 3400+ with and without HT. I dont do photoshop or anything taxing while doing any other apps at the same time. But I have never noticed a difference personally.
 
HT = Half Ass form of SMP

Dual Core = Defines the meaning of SMP

Forget HT all, the next significant change in processor design is multi-core. Dual Core Athlon 64FX = Drool.
 
intel does not have a cross license for 3dnow. this was clarified in an article during the opteron roll out phase. as for hyperthreading making a system slower-sometimes, a dual can exhibt the same problem. driver coding for 2 processors is no easy task. also there have been known issues with windows xp scheduler and has been patched, but supposedly will require another patch for some better performance (sp2). as i have read and seen hyperthreading on linux is awesome because of its robust scheduler-so do not think that it is the hardware, software can easily be the issue.

hyperthreading is not useless if anything it started the push towards dual cores and coding for dual cpu's. you have to supply hardware before software for it is made. i am really glad that amd has finished dual cores, but the only problem i see with this is the fact that it will be an opteron before we see it a non enterprise solution.

noticing is rediculous becuase it will be relative to the person. benchmarking can be the same-honestly i have never seen a perfect benchmark. although, it does provide an idea of what to expect and probable problems, they never seem to be up to date for the hardware nor test it in its specified state.
 
Those who are hoping for HT in Intel's next CPU I wouldn't bet on it. They are probably going to strip that feature out. Because A) Microsoft isn't going to give home users 4P SMP capability. B) It will save space on the die which will be critical when going to a dual processor core.

Also, having owned a P4 and an Athlon 64 side by side. When the computer gets bogged down from programs opening and closing HT does help smooth things out. HT helps durring boot up when all the Appz loaded in the taskbar. But the other 95% of the time it makes little diffrence.

Also people keep asking about SMP on the FX. I went to the Boston Tech Tour. Asus and Tyan are hinting we might be seeing some SMP on the lower end A64 CPUs. Like the good old days of the dual Celerons. :) I don't know anymore about it so don't ask.
 
M4d-K10wN said:
Who cares, ht sucks. Next thing we'll know Intel decides to add "extreme multiplier pro and a half", that lets the processor use multipliers by increments of .25. Go figure why it's useful.
Click to expand...

What the hell are you talking about? Hyperthreading is successful technology that doesn't take much money for Intel to implement. It shows increased performance in many apps. I don't think you should post at all if you do not understand what we are discussing. And I still don't understand what you are trying to say with the multiplier comment.
 
Anyway, annouced today 2H 2005 Dually core Opterons and desktop chips.

As we know, HT is an effect of a long pipeline in the P4 architecture. Intel got lucky that some clever clogs figured a way to fill it up and boost the chips performance. Or maybe it was totally deliberate. Either way the idea has made intel loads of dosh and the mainstream lapped it up without even knowing what it is.

I always chuckle at the PC world (UK) ads on TV. "Featuring a three point zero gighurts processor with HT technology" - I mean come on, WTF is HT technology when its at home? Amazingly it seems to sell.... :confused: (speaking from a general public point of view - I know what it is :p )
 
Mad_Pyro said:
What the hell are you talking about? Hyperthreading is successful technology that doesn't take much money for Intel to implement. It shows increased performance in many apps. I don't think you should post at all if you do not understand what we are discussing. And I still don't understand what you are trying to say with the multiplier comment.
Click to expand...

no it doesnt. HT is a gimmick made by intel to attempt to make their processors on par with AMD's performance:megahertz ratio, when in fact its PROVEN to make single threaded apps slower.
 
It increaces double threading performance by about 18%, now? and decreaces single threading performance by about 6%. VERY USEFUL.
 
Hypertheading is quite useful, as shown by the numerous people that burn a CD while playing ut2k4 (/sarcasm?)

AMD plans to implement a technology similar to hyperthreading by going straight to dual core, and AFAIK, anandtech announced that they finished the design and dualcore opterons will be available next year.
 
Fryguy8 said:
Hypertheading is quite useful, as shown by the numerous people that burn a CD while playing ut2k4 (/sarcasm?)

AMD plans to implement a technology similar to hyperthreading by going straight to dual core, and AFAIK, anandtech announced that they finished the design and dualcore opterons will be available next year.
Click to expand...

as will dual intels i believe, surprisingly WITHOUT HT
 
kronchev said:
as will dual intels i believe, surprisingly WITHOUT HT
Click to expand...
Where have you heard that they plan to drop HyperThreading with the two-core processors?
 
M4d-K10wN said:
Who cares, ht sucks. Next thing we'll know Intel decides to add "extreme multiplier pro and a half", that lets the processor use multipliers by increments of .25. Go figure why it's useful.
Click to expand...

ever use HT? I can tell a difference on my 2600+ where HT would come in handy. Take this example (which is the main thing i notice without HT):

I use the program objectdock (it's like the Apple icon thing where you hover over an icon with your mouse and it enlarges) and i put all my icons on that so when i open something i click on it. Now with my AMD system when i click on something on that bar, there's a pause before the program opens (the whole computer pauses for a second or two)...it's kinda annoying but nothing i can't live with.

Now with my P4 system i had before my AMD system there was absolutely no pause when clicking on that bar and opening the program. And when i ran my mouse up and down the bar with the P4 system it uses roughly 50-60% of the cpu while it's 100% with the AMD. So it does have it's uses.

If you go from a P4 with HT to an AMD XP system you will be able to tell where HT comes in handy...it's when trying to open multiple things at once is where you benefit from it the most.

But when it comes to opening one single task such as games or stuff like that my AMD system was faster than the P4.
 
xonik said:
Where have you heard that they plan to drop HyperThreading with the two-core processors?
Click to expand...

I think that there won't be HT in Intel's planned dual-core Dothan for desktops. But, if they make a dual-core out of Prescotts, then you will see HT. I doubt, though, that we will ever see dual-core Prescotts because of the insane amount of heat that they would produce. Correct me if I'm wrong, but with two 100W cores, the cooling system would have to dissapate 200W! That is not economically feasible, so Intel is taking their Banias/Dothan design and throwing it into desktops, possibly in dual-core form.

I already know that the Banias chips run in the realm of 20-40W, but does anyone have info about the heat put off by the upcoming 90nm AMD offerings? Their latest Athlon 64s run pretty hot (80-90W?) and putting two of those together in a dual-core packaging would result in 150W+. This also seems sort of impractical, so I would presume that their 90nm parts run cooler if they are planning on packaging them in dual-core form.
 
Well, considering I never multitask, and when I do, that might mean the dastardly act of running Internet exploere AND Msn messenger at the same time ( :eek:!!! ), I can say that 94% perforamnce is no thnx.
 
mantastic said:
ITheir latest Athlon 64s run pretty hot (80-90W?) and putting two of those together in a dual-core packaging would result in 150W+. This also seems sort of impractical, so I would presume that their 90nm parts run cooler if they are planning on packaging them in dual-core form.
Click to expand...
That can't be true. I've played with desktop Athlon 64s and a Cool 'n Quiet compatible motherboard, and it runs so cool the motherboard shuts off the fan during light CPU loads. I see temperatures in the 30s and 40s degree Celcius during games and encoding type work, with the stock heatsink/fan unit, mind you. I have serious doubts that they actually dissipate that much. Also, read my post in this thread:

http://www.hardforum.com/showpost.php?p=1026175971&postcount=6
 
xonik said:
That can't be true. I've played with desktop Athlon 64s and a Cool 'n Quiet compatible motherboard, and it runs so cool the motherboard shuts off the fan during light CPU loads. I see temperatures in the 30s and 40s degree Celcius during games and encoding type work, with the stock heatsink/fan unit, mind you. I have serious doubts that they actually dissipate that much. Also, read my post in this thread:

http://www.hardforum.com/showpost.php?p=1026175971&postcount=6
Click to expand...

they do dissapate around 80watts but they are good at power management. however dual cores will NOT mean "two cpus on one package each putting out 80 watts"
 
No, I'm quite sure that they are designed to use a heatsink that can handle ~80 watts, not that they actually dissipate 80 watts. If you were right, then how come all the desktop, Socket 754 Athlon 64s all "dissipate" the same amount? Look at the datasheets if you don't believe me.
 
You must log in or register to reply here.
Back
Top