What CPU war?

Page 3 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Wow

Lots of testosteron in this thread. Decaf anyone?

PS: in case you wondered about my credentials, I'm 13 years old, I failed miserably in primary school since I just couldn't grasp 10 times tables; my biggest personal achievement is my diploma of 25m breastroke. No one else in my family ever achieved that so late in their lives or required more than the 7.5 minutes it took me. I also think its a record in the 50 year existance of our school.

You forgot to tell him about your lesbian twin sister.

I sure hope you just had some spare time on your hands P4 because some things are just not worth working up all the energy for. With allergies you need to know when to take your medicine.

If people start supplying their credentials on forums like these I know it's time to move on to more interesting threads.



BigMac

<A HREF="http://www.p3int.com/product_center_NWO_The_Story.asp" target="_new">New World Order</A>
 
>well wether that makes sense or not i up for debate. for
>one, its obvious amd didnt need HT

So ? It doesn't "need" a trace cache either, but that doesn't make it a bad idea either, but instead one that may well be incorporated into the next AMD core.

> since the id bet amd is looking into other architectures
>that dont focus on extending the pipeline, and there fore
>wont take advantage of HT

You're probably wrong twice: 1) AMD is most likely (if not, then surely) extending the pipeline for K9, just like they did going from K7 to K8 or any other generation before that. Its not because netburst and especially Prescott is bust and went way over the top with hyperlong pipelines that moderately extending the pipe to enable higher clockspeeds is suddenly a bad idea. 2) SMT can make sense on shorter pipelined designs just as well. Have a look at Niagra: only 6 stages, yet 4-way SMT capable.

> plus the other thing is that there would need to be a
>large compelling software need for HT that would push amd
>to do that.

Oh you mean like there wouldn't be a need for threaded software to take advantage of upcoming dual core chips ?

>they are heading for dual core, that is thier answer to HT

no its not. Strange "answer", something that doubles core size, power consumption and theoretical performance in reply to something that costs 2-5% die size for aprox 10% performance gain. Read my posts above, CMT and SMT are absolutely not mutually exclusive, combining them makes perfect sense.

>i dont think that your theory doesnt make sense, its just
>that based on trends now, id say its improbable that will
>happen as soon as you would think.

We'll see. I quite certain it will happen, wether K9 has it or not, I'd say 50% chance, same with Meron. The cores after that, 95% at least. Besides, netburst isn't gong away anytime soon, prescott will be shrunk to 65nm which will extend its live until well after 2006 or 2007. Don't expect dualcore dothans before 2006 either, let alone it would replace prescott by then.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
 
>Lots of testosteron in this thread. Decaf anyone?

Does decaf counter testosterone ? Hmm.. maybe I should stop drinking expresso's when my gf is abroad then :)

>You forgot to tell him about your lesbian twin sister.

Nah, that isnt my own personal achievement. She's is the one to credit for having the guts to undergo that gender change.

>I sure hope you just had some spare time on your hands P4

Plenty.. far too much in fact

>With allergies you need to know when to take your medicine.

I just figured sometimes sneezing can be more healthy than taking the medicine 😛

>If people start supplying their credentials on forums like
>these I know it's time to move on to more interesting
>threads.

Thought I'd just pick up the role of playing the resident fudbuster, since I've not seen a whole lot of more interesting threads.. did I miss perhaps miss anything in the "other" forums ? Can't be bothered skimming through all the pointless rubbish there

= The views stated herein are my personal views, and not necessarily the views of my wife. =
 
ok id like to see how you KNOW im wrong? where is your info that K9 will impliment HT? all your doing is speculating the same as i am, so neither of us can say what conclusion there will be. id like to see some credible proof that k9 would use any form of HT, cuase you know if even the hint of such a thing was out there, it would have spread around by now, since the k9 core will be here next year.

i never said amd wouldnt extend its pipelines, just that HT might not catch on as quickly or give the extra boost you see on a p4 today. i mean why would intel not impliment HT on p-m if it added performance? in fact, intel adds things even when they dont add performance, but just add marketing boosts lol. youd think there was a compeling reason intel wouldnt use it there, since it is all but a trademark now.

you also took my quote incorrectly i said amd is probably looking into architectures thats dont FOCUS on extending the pipeline, i didnt say it might not include extending it, just that that would be of lesser importance. you have to agree that has been thier history, they work on other areas to get performance out of every cycle. now intel yeah, they put more faith in those pipelines for raw clock cycles, well they did anyway..

and agian you took a quote of mine incorrectly. saying dual core is the answer to HT, i didnt mean that comparing the tech, that its compareable, im talking from the marketing aspect. how long has HT been a corner stone for marketing the p4, wlel how does dual core sound for an athlon 64 lol.

id totally agree with you that you cant compare the software tech to the physical tech. but i dont see where you get your iron confidence that amd will adopt HT, id love to see some evidence that caught your interest, its possible ive missed something. and just becuase something is avialable doesnt mean it needs to be used, amd could have put ht into its k8 core, it wouldnt be impossible to pull that off, even waiting to impliment it in s939, but see there are otehr factors beyond just the fact that its there that decide wether they use it or not.

but in the end, amd is doing fine as it is now, even against the dreaded HT enabled p4s lol, they arent neccesairily hard pressed for ways of boosting performance at this time.
 
>ok id like to see how you KNOW im wrong? where is your info
>that K9 will impliment HT?

Don't sweat it, lol, I'm not claiming to *know* AMD will implement HT, I just claimed its likely your reasoning to conclude they would not was wrong when you said <i>" id bet amd is looking into other architectures that dont focus on extending the pipeline, and <b>there fore </b>wont take advantage of HT". </i>Just wanted to point out it is more than likely pipeline length will be increased (obviously, that won't be the only change), and even if it weren't, this doesnt indicate it would not feature some form of SMT.

> i mean why would intel not impliment HT on p-m if it added
>performance?

Why didn't they add SSE3 yet ? Or amd64 ? or a trace cache ? 800 MHz FSB ? Even intel doesn't have infinite resources and has to bring out products from time to time; adding SMT isn't a simple operation either, and clearly Banias/Dothan had other primary design goals than extracting every last ounce of possible performance, the focus was clearly not on radically improving the basic PPro core. More to the point, even if intel would have had the time/resources, and even if the required effort would be warranted performance wise considering Dothans (old) design, adding HT might have had a decremental effect on Dothans primary design goal (low power/performance ratio).

>but in the end, amd is doing fine as it is now, even
>against the dreaded HT enabled p4s lol, they arent
>neccesairily hard pressed for ways of boosting performance
>at this time.

Oh.. well, I guess they can just stop working on K9 and K10 then, hmm ? 😉 I still stand by my point: obviously AMD is looking at ways to further increase performance for future chips, and SMT is a very good candidate, especially since, like I said, other unused technolgies for increasing IPC are becoming increasingly rare, and the required design effort aside, there are few downsides to it. I do not *know* K9 or K10 will feature it, but I think its very likely. Same for Meron.

And one more thing: marketing has nothing to do with this. Obviously, marketeers will try and exploit any new feature/technology, like they did with HT (and AMD64,SSE,..hypertransport,.) but there is no denying HT offers some compelling performance advantages under certain circumstances. Are you aware of many other ways to increase performance up to >40% and lets say around 10% on average that are just as cheap in silicon estate ? The P4EE has this huge L3 cache, and it makes about the same difference as HT, arguably even less, think about that... Hyperthreading doesn't give the P4 the edge over the A64, but its a worthwhile technology nevertheless, just like a trace cache for instance.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
 
look guys I don't mean to be a d**k, but I could use some help with a new build and you guys are all in this room arguing over some [-peep-] that is too tedious to read through. Could some of you brainiacs come over here and help me out?

<A HREF="http://forumz.tomshardware.com/hardware/modules.php?name=Forums&file=viewtopic&p=181437#181437" target="_new">here</A>



pretty please??
 
Don't read the post if you're above it all.

I actually had a great laugh when I read that statement. You look down your nose at everyone in this forum (as per your lovely generalizations) and yet have the audacity to turn around and accuse the rest of us of the same?

Spare me.

And please, copying a copyrighted work almost word for word is plagarism. It's not flattery, and the original author shouldn't be pleased that someone else said exactly the same thing. That's why copyright law exists.

<font color=red> If you design software that is fool-proof, only a fool will want to use it. </font color=red>
 
>Power 5 and niagara montecito dont use SMT.There are switch
>on long event.

Power5 definately does SMT. It even adds thread priority. <A HREF="http://www.cs.utexas.edu/users/cart/arch/fall03/KallaSlides.pdf" target="_new"> link</A>

Montecito and Niagra probably do coarse grain multithreading, since AFAIK neither will do OOE. But if you have a link to prove otherwise, be my guest.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
 
now oyu know i didnt mean that amd didnt need to research and produce new tech wiht performance increases. my point is amd vs intle right now shows that HT on a p4 doesnt give it a lead over amd chips in all areas. in fact most areas are still held by the ahtlon 64 line and not p4s. my point is that amd doesnt have to impliment HT becuase everyone else is doing it, or becuase they will need that edge. they showed they could do it wihtout ht this time, and i bet you they will prove they dont need it in the K9.

i should just stop posting since really i have no more info then you and that both of our conclusions arent based on any facts, although id argue my conclusion is more pluasible then yours at the moment, but thats beside the point. neither of us can prove our positions lol.

i dont doubt intel will try to get ht into thier next chip, i mean why wouldnt they, it has been so good to them so far. i think your right about its a good chance intel will keep using it, i just think its not as likely amd will use it.

Why didn't they add SSE3 yet ? Or amd64 ? or a trace cache ? 800 MHz FSB ? Even intel doesn't have infinite resources and has to bring out products from time to time

well i can tell you why most of those things havent shown up yet. sse3 was probably kept out for the mobile prescotts that intle is now introing. they will probably put sse3 into the next p-m core, along with the greater fsb. as fr as amd64 goes, you know good and well its not just a design issue as to why they havent implimented it. heck theydont evne care to get taht out on prescotts till next year if we are lucky, amd64 is gray area, its like its more political now then anything. intel doesnt want to have to push it out if it doesnt have to. id also argue somehting like adding 800mhz fsb is different from adding HT capabilities.

all im thinking is it stands to reason that if indeed p-m will move to desktop soon, that they would have implimented ht by now on dothan in the first place, its not liek its a new tech for them like sse3. HT has been around a while. I can think of a couple reasons intel might not. Either the performance isnt like youd see on prescott or northwood, or they are saving that to be introed when p-m moves to desktop.


i never was trying to say HT was worthless, its proved its much more then that. its been a saving grace for intel. i was just pointing out its a nice marketing term to be able to push. i think it has a future with intel, although dual core sort of clouds things, one would assume intel would use HT on dual cores, thier architecutre would thrive on that.

but as your well aware, amd and intel work very differently whne it comes to thier design philosophies, just becuase it works for intel and its out there, doesnt mena mad has to use it. yes it probably oculd help in some areas, but amd has plenty of options on things to impliment, they cant do it all, so they hcoose some things over others. its very possible they choose soemthing else over HT.

for the most part you and i are in agreement. the big issue is wethere its something amd will do, which i dont belive will happen with K9, since there is nothing to move the discussion one way or another, we both have to wait lol.
 
Your own paper show that a large single core with local Branch history can be faster that dual smaller core.


Joel S elmer made some paper on that.Alpha team were beliving that a 8 instruction wide 4 way SMT would have been faster that a dual core 4 instruction wide single treads.
SMT is the biggest improvement since superscalar.It allow TLP to be turn as ILP.

i need to change useur name.
 
>After verification you were right on

Well, there is a scoop.

>.I wonder why no one have move to 4 way SMT

Most likely diminishing returns. I imagine with a 2 way SMT implementation, you'd already achieve a rather well utilized pipeline on todays designs, at least as good as other bottleknecks will allow. 3 or 4 way might add little to no performance while increasing complexity further. Could even be a decremental effect FAIK.

= The views stated herein are my personal views, and not necessarily the views of my wife. =
 
All alpha and intel paper on SMT were base on a 4 way SMT.On a 2 way SMT there still time were you are stuck with a dual event or else.A 4 way SMT cannot almost never stall.Also the front end is also not much SMT aware.

i need to change useur name.
 
Sorry P4man,

I don't use SB drivers, as you assume. I have the latest BIOS and if you read my BIO, I use a PC Power and cooling supply. So, I HAVE tried to eradicate the HT instability in my system with the more typical solutions.

No, I wouldn't buy a system from you if you would stand back and insult the customer because you aren't able to resolve the issue for them and assume that THEY are somehow at fault.

You continue your, "I must hate HT" logic. No, it is fine with specific software (most) but it does add a set of circumstances with select applications and under certain conditions that DO cause crashes.

I'm afraid your right, I'm NOIT a scientist and neither are you. I'm and Engineer in a specific field of study. If a set of solutions cause a problem, you have to eliminate one, or the other. I need to use Email, and HT does little for me in real terms. I eliminate HT until which time it's instability is resolved. How does the escape you as "illogical"?

Currently, it may be the driver for my modem. It is the latest version but many vendors will not update to all applications or system configurations. So, I may be stuck until I either get an expensive Internet connection (don't need it) or more compliant modem drivers are found in other modem brands. Yes, it isn't HT specifically and I never said it was. You needed to make that assumption to continue your line of superiority.
 
Please POST the so-called plagiarized source material (with the proper footnote, of course). Seems no one wants to dig it out.

Assumptions, and who's making those?
1.0 I "know" of this source and plagiarized it? No.

Definition: Plagiarism is knowingly using the ideas or words of another person and claiming that they are your ideas or words.
http://www.science.smith.edu/~mmarcotr/Hortwebpage-fall/termpaper/plagerismlesson.htm

I unknowingly use a similar "concept", not "exactly" the same thing, to describe a set of processes for clarity. Where do, someone said hot-dogs, parallel my size of milk glasses? The volume of milk poured? The height above the glass? Using "food" isn't really even that similar except for it was indeed "food" that was the analogy or simile.

2.0 I "hate" hyper thread? No.
Please posts my "hate" will you?

A few individuals that run around like news group Nazi's and spill THEIR assumption and try to FORCE their ideals is what you should be spared from. Making up illogical assumptions on posts, as well as who does or doesn't like what, in order to support a single minded and "invented" campaign (HT never has crashes, who "knowingly" plagiarized what, ETC) is of no use to me or any one else. We have some swagger that is out of hand, and many private posts to me agree, on this web site by certain people. Sorry if I can't be chased off by all of this. Many will read the posts, but too few will participate with the criticism and plain mean spirited self-grandeousing comments made to inflate egos.

It would be all too easy to fall into that trap, and I apologize to anyone who feels I have. The news group is about helping solve problems. We all have knowledge to share. Crass comments and put-downs are not about knowledge but insecure self-proclamations.

Someone asked about my background and age. I did. I didn't say it made me better than them (I was insulted in an Email that said I was chicken to NOT give the information!).
I have no idea what use it really is, but you asked. I'm not ashamed of it and yet I am not a PC software or hardware expert either. And it was all correctly mentioned most education is of little practical application unless it improves problem solving, different than problem BLAMING, ability. Do I believe that Sound Blaster products are horrible because certain key individuals say so simply because they interfere with HT? Do I even know if this is true? It is true that the products won PC Worlds and Maximum PC best in class awards.

For those of you who enjoy slander, cursing, and self-grandeoused put-downs to get your point across I don't say good riddance but simply, go ahead and hate me because I won't waste too much time hating you. Ants are at all picnics.
 
This should really help servers. SCSI types multiple instruction tasks being fitted to IDE drives is another example. I'm not so sure where desktop PC's are in all this just yet. But heavily used servers need to be as fast as they can be.

I agree that HT is a needed technology for really hard-hitting applications. The CPU is just sitting there between tasks? Give it some more if they are there!
 
seems you got kind of late to the thread ehre as it has mostly died off lol

anyway, you remidn me of a good parrelle we can draw now. look at the debut of ncq (native command queing) to desktop hard drives. it was harelded as the enxt big thing for hard drives since hyperthreading for p4's. on paper it looked promising, it allows the hard drive to decide which tasks to complete first to make it more efficient. but now that we actaully see such a drive , maxtor maxline III, using a proper ncq enabled chipset , intel 915/925, there is no real beneift at all to users other then those that started using the process in the first place, servers. so while ncq wasa hit in the scsi world, i have a feeling it wont be as useful in the desktop world, where more cache, larger capacities, and to a lesser degree rpms and connections lend more of a boost. I can give you plenty of examples of that.

hyperthreading does work wonders on a p4, but that doesnt neccesarily mean it would work wonders on an athlon 64. somthing that may look great on paper or show great things in one place, can behave very differently in another.
 
since when did i call that low? did you even read my entire post?

i even said the opposite, HT has done wonders for the p4, did you miss that line? seesh i wasnt putting down HT at all, dont get so defensive

You should buy need glass

im sorry but that is giberish, what were you trying to say?
 
<font color=blue>"I can make a cappucino faster with a prescott beat that!"</font color=blue>

You could also heat the cappucino (to boiling) with the Prescott as well...

<font color=blue>This is a Forum, not a playground. Treat it with Respect.</font color=blue>
 
Sorry i dont have read the last few ligne.

The cpu that gain the most from SMT will be VLIW and in order cpu like Tukwila and Niagara.

i need to change useur name.
 
ok, thats fine then.


and yes i can agree that VLIW and the like will take advantage of SMT for sure. it looks like ibm will be doing so when it makes the triple core chip for the xbox 2, you can see an article about it here:

<A HREF="http://www.xbitlabs.com/news/mmedia/display/20040627030133.html" target="_new">http://www.xbitlabs.com/news/mmedia/display/20040627030133.html</A>
 
I wonder how small are those core to reach 3.5 ghz.Dual instruction wide single FPU?

I would like that intel can with a very simple 4 SMT core.There is some much possibility with IA64 and SMT combine.

i need to change useur name.<P ID="edit"><FONT SIZE=-1><EM>Edited by juin on 06/28/04 01:29 PM.</EM></FONT></P>