Forum Home
Press F1
 
Thread ID: 45095 2004-05-10 00:36:00 HyperThreading Technology Performance Questions Chilling_Silence (9) Press F1
Post ID Timestamp Content User
235462 2004-05-10 00:36:00 Greetings,

Ive just been a bit curious about Hyperthreading, its effects, how it works, and does it work....?
Does it simply have twice the number of registers?
Does it do much performance-wise seeing as its still only a 3.4 (or 3.2) Ghz processor doing two things at once?

Comments? Ideas? Links I can visit?

Cheers


Chill.
Chilling_Silence (9)
235463 2004-05-10 00:48:00 for most things it is almsot as good as a duel CPU system.

as far as i know its two CPS but only onw cache and one instruction decoder and micro code cache.

have a read of this.

www4.tomshardware.com
robsonde (120)
235464 2004-05-10 01:01:00 Here's an article that compares performance with HT enabled/disabled.

accelenation.com
metla (154)
235465 2004-05-10 09:22:00 There is some good info at arstechnica (arstechnica.com).

My summary would be: it sounds great in theory, but in reality most apps are still single threaded, there are extra overheads running a SMP/T OS, and it can mess with the processor cache which the P4 really needs to stop it turning into a celeron.

Things will get more interesting when dual-core chips go mainstream.
bmason (508)
235466 2004-05-10 15:30:00 Needs the correct software to maximise performance otherwise you can actually end up with slower performance wait for a few years before jumping onto it kiwibeat (304)
235467 2004-05-10 19:02:00 I know nothing of the technicalities of it, but in MHO it does exactly nothing to increase speed. My comp tells me it is multi-tasking. So what?
Jack
JJJJJ (528)
235468 2004-05-11 03:04:00 > MHO it does exactly nothing to increase speed. My
> comp tells me it is multi-tasking. So what?


That was my thoughts too... Its still only a 3.4Ghz CPU, still capable of the same number of CPU Cycles per second. For over-all number-crunching (Such as encoding an MP3) it doesnt appear to offer any benifit because the CPU is already maxed out trying to do the encoding.

OTOH, if you were watching TV in a window while chatting online it might offer a marginal increase from what I can see, but neither is likely to need to Max the CPU, so the difference would go largely un-noticed.

Personally I dont really see the point.. Its still crunching the same amount of numbers right?
Its still by and large 3.4Ghz regardless of if its a Hyperthreading P4 or two non-HT 1.7Ghz CPU's right? IMHO if you need to squeeze out that much life, buy a SMP system, say two 2.8Ghz CPU's for the same price as one of the EE HT P4 3.4Ghz CPU's!

Or have I missed the rail completely here???


Chill.
Chilling_Silence (9)
235469 2004-05-11 07:36:00 www4.tomshardware.com elf_to_ht
The last paragraph somes it up i think.
mark.p (383)
235470 2004-05-11 11:06:00 > That was my thoughts too... Its still only a 3.4Ghz
> CPU, still capable of the same number of CPU Cycles
> per second. For over-all number-crunching (Such as
> encoding an MP3) it doesnt appear to offer any
> benifit because the CPU is already maxed out trying
> to do the encoding.

You are assuming that the CPU can be kept busy 100%. But what tends to happen is it gets stuck waiting for something requested from memory which is much much slower than a register or cache.

The idea of HT is to give the processor something to do while the other thread is blocked.

Things like a big processor cache and variuos forms of prediction are already used to reduce the effect, but the high clock rate of the P4s makes the waits worse.


But as I said about it sounds great in theory.....
bmason (508)
235471 2004-05-11 13:59:00 > > That was my thoughts too... Its still only a
> 3.4Ghz
> > CPU, still capable of the same number of CPU
> Cycles
> > per second. For over-all number-crunching (Such as
> > encoding an MP3) it doesnt appear to offer any
> > benifit because the CPU is already maxed out
> trying
> > to do the encoding.
>
> You are assuming that the CPU can be kept busy 100%.
> But what tends to happen is it gets stuck waiting for
> something requested from memory which is much much
> slower than a register or cache.

Okay, so taking it from a slightly different angle... I can currently surf the web and encode an XviD avi at the same time because I know that encoding the avi isnt maxing out the CPU. Therefore surely the OS or CPU (Wichever) would be clever enough to move over to something such as the web-browser window which needs a change in information, correct?

This is all new ground to me...


Chill.
Chilling_Silence (9)
1 2