The simple meaning of HT?

[zoonychina]

Good day!Friends.I know what the "superpipeline" and the"superscalar" ,but what is the HT's meaning?I have seen the essays about that before,but I am just a freshman.I cant get the real meaning of it.

 

[ytoledano]

Single CPU systems without HT can't handle more than one thread (something that requires processing power) at a time, so they have to so one thing and then the other. Dual CPU systems can't work on the same thread at the same time so the workload is divided, and if there is only one one task to be done than usually, the gain from having 2 CPUs is very small (that's why they're not recommended for home users).
HT is sort of a cheap compromise, it can handle 2 threads at the same time (something a regular CPU can't) but doesn't process both threads at the same speed as it would have done if they were one at time (a thing a dual CPU system would have been capable of). HT CPUs gain a performance boost when handling 2 tasks but not a performance penalty when handling one (like a dual CPU system would have been crippled to use only one CPU when processing one thread).
You might think "But my CPU just has one thing to do at a time like playing a game or ripping music" but actually many other programs in the background need to use some CPU cycles while they're open (instant messaging software, anti-virus, pop-up blocker, file sharing software (especially file sharing software's), and other processes and services you might not even be aware of) so HT helps more than you'd think.
What I like about HT is that it has virtually no shortcomings (some would say it hurts performance when single-tasking but that's probably less than half a percent). You should note though that the only Windows base OS's that HT works on are WinXP, Server 2003 and Win2K (though not optimized for it). If you have WinME and older, you'd have to turn HT off.

I know what everyone's been wondering - 2^33634943-1 IS NOT PRIME. Now, wasn't that worth running <A HREF="http://www.mersenne.com" target="_new">Prime95</A> for a month to know?

 

[pIII_Man]

win 2k takes a performance hit with HT enabled, best to turn it off...


If it isn't a P6 then it isn't a procesor
110% BX fanboy

 

[samcheetah]

ytoledano that was a very nice answer. i have a question in this regard. is there some type of hybrid dual processor HT configuration. so that both processors will be HT and divide their work efficiently. by efficiently i mean that when there is need for both processors to work on the same thread both will work on it. and when there is need for the processors to work on different thread they will work on different threads. this sounds like utopia but has any microprocessor company attempted.

 

[pIII_Man]

xeons have HT but still that means 1 thread per cpu. As of now there is no solution (at least that i know of) like you stated. However the future will bring dual core processors which should be able to do what you stated.


If it isn't a P6 then it isn't a procesor
110% BX fanboy

 

[ytoledano]

Having 2 processors working on the same thing is not only hard to execute technologically, it kinda contradicts the laws of physics. Processors may be doing great things but they are still making many (many many) small calculations one at a time extremely fast by processing data that is in their cache (which is also why a large and fast cache is important). Einstein proved the world that nothing can move faster than the speed of light, more to the point, you can't compare data moving from the cache to the transistors (or whatever's there, I don't really know) a distance of a few millimeters on circuits that are measured in nano-meters to two CPUs handling the same cache and data moving on the motherboard a distance of sometimes 10 centi-meters on common copper (or even silver) lines. That's why, like P3 man said, Dual-Core is the future.

I know what everyone's been wondering - 2^33634943-1 IS NOT PRIME. Now, wasn't that worth running <A HREF="http://www.mersenne.com" target="_new">Prime95</A> for a month to know?

 

[samcheetah]

xeons have HT but still 1 thread/CPU !!!! if it is HT then there should be two (or more) threads/CPU. am i right

 

[samcheetah]

ytoledano that was just a thought and thanx for correcting it.

 

[ytoledano]

I didn't mean it that way...
Actualy, your thought was right and is implemented in a few aspects today only in a simpler way. Know what are SETI and Prime95? They're projects which require an unthinkable amount of computing power and the workload is split between hundreds of thousands of computers (processors). A main server decides who does what and so there is very little inefficiency (though there is some). You can also find something similar in large corporation servers with multiple CPUs which work on the same project but each take different actions (they might be synced by software or hardware, I really know very little about those million $ and up systems).

I know what everyone's been wondering - 2^33634943-1 IS NOT PRIME. Now, wasn't that worth running <A HREF="http://www.mersenne.com" target="_new">Prime95</A> for a month to know?

 

[pIII_Man]

the ht on xeons is exactly the same as with p4's, one xeon could run 1 or 2 threads at the same time. A dual xeon system could run up to four (2 per cpu) however both of the cpus' work load CANNOT go to a single thread.


If it isn't a P6 then it isn't a procesor
110% BX fanboy

 

[juin]

superpipeligne dont exit and i have nerver hear of that.

Superscalar

Introduce in the pentium 1 that allow the CPU to excute 2 operant at the same time in Parralle.
EX

mov AL,0 put 0 in the register AL
mov EBX,15 put 15 in the register EBX
add ebx,al add Al to EBX store reslu in EBX
mov......................

The cpu can make the 1,2 memory operation in the same time

HT is the marketing name for SMT.Introduce in the P7 core of intel (p4).That allow the cpu that have 2 tread in the same pipeligne with some stuff must be double cpu arch state IP.......

SO a CPU will fetch operant instuction 2 tread rising the output of the chip.Just think about having the same code twice so you are sure that at lease 2 to 3 operant can be fetch per cycle.There cons to cache pollution fewer resource for critical loop and lower branch prediction as the Branch history hold a mix between the 2 tread in use.

I dont like french test

 

[Mephistopheles]

That is exactly how it works, ytoledano. A main server has to be able to split a program into multiple modules in order for many processors to work efficiently. For instance, for checking primes, you could have one processor checking divisions by odd numbers and another checking divisions by even numbers (somewhat of a simple example, but it works... considering the most stupid way of checking primes), so one doesn't need the result of the other.

This is actually the important part: most programs written in the "conventional" way are linear and require previous results in order to continue operation. For instance, say a program with initial defined values of A and B does:

C=A+B
D=C^(2.5)
E=D*(A+B^4)

...or whatever, just an example. So how could you parallelize this code? Well, you couldn't. How could you evaluate line 3, without having the results of lines 1 and 2? And if you use one processor to calculate these things, then there is no use for a second one - when the first one finishes, it'll be free to do line 3...

So the big issue here is paralellization. This "server" you mentioned - responsible for dividing tasks - must be appropriately programmed, and that's the hardest part. Most of the time, this can't be done without some careful planning.

In that sense, it was very wise of Intel to spread HT before introducing multi-core processors: by the time the dual-core processors are out, many programs will be multithreaded because of Hyperthreading....

<font color=red><b>M</b></font color=red>ephistopheles

 

[samcheetah]

>Know what are SETI and Prime95?

yes i do. i once thought of installing the SETI@home software but then i thought hey im gonna participate in a hunt for aliens lolz. but the prime95 thing is interesting and i visit the mersenne.org site almost everyday.

>A main server decides who does what and so there is very little inefficiency (though there is some)

i recently read an article related to the TeraGyroid project. it was at this year's supercomputing conference. the project required literally billions of calculations for fluid mechanics simulations. the author of the article wrote that scientist dont like the idea that calculations should be divided among home computers (like the SETI and Prime95). the scientists (of the TeraGyroid and the like projects) think that u should only have state of the art supercomputing machines. why is that so??? i mean if the main theme is to divide the work then why only supercomputers??? have any ideas!!

>the ht on xeons is exactly the same as with p4's, one xeon could run 1 or 2 threads at the same time. A dual xeon system could run up to four (2 per cpu) however both of the cpus' work load CANNOT go to a single thread.

ok thanx for the info piii man.

 

[ytoledano]

yes i do. i once thought of installing the SETI@home software but then i thought hey im gonna participate in a hunt for aliens lolz. but the prime95 thing is interesting and i visit the mersenne.org site almost everyday.

I think that's the main reason why people prefer Prime95 over SETI. Prime95 will tell you something like 2^33634943-1 is not prime because it has a factor: XXXXXXXXXXXXXXXXXXXXX, or more oftern: 2^33634943-1 is not prime because Lucas-Lehmer found out it was not.
SETI doesn't tell you crap.

A main server decides who does what and so there is very little inefficiency (though there is some)

Prime works differently than SETI and other software (and your example of the fluid mechanics) because it's not really the workload that get's devided, but the "small projects" (2 months per 10M digit prime) that get devided so each computer completes a project and not just a few calculations. People with slower computers or who don't want 2 months projects also don't have to do parts of pojects, they can factor big yet-to-be-proved primes with small numbers, trying to eliminate the need for the long Lucas Lehmer test.
In SETI on the other hand there is this one big project, and you get no results what-so-ever about (there is no "ET wan't found in the sector you scanned") all users do there is elimintate unlikely sectors, the analogy to Prime95 would have been finding the exponent to check and letting a supercomputer check it.

I know what everyone's been wondering - 2^33634943-1 IS NOT PRIME. Now, wasn't that worth running <A HREF="http://www.mersenne.com" target="_new">Prime95</A> for a month to know?

 

[ytoledano]

Oh, and the inefficiency I was talking about was when computers start factoring a prime ("I'm gonna find that 10M prime with my P3 500 and win that 50,000$!"), get desperate and stop.

I know what everyone's been wondering - 2^33634943-1 IS NOT PRIME. Now, wasn't that worth running <A HREF="http://www.mersenne.com" target="_new">Prime95</A> for a month to know?

 

[samcheetah]

ok thanx ytoledano. the prime thing seems fun. altho i dont think im gonna get into a 2 or 3 months project i might try it out someday.