Hyper-threading, what could it bring us?

Intel's website states that it will boost performance up to 30%.

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>

Basically, while one thread is locked up waiting for a memory fetch for some instruction, execution moves on to the next thread. Thus, the processor never has to be idle, except when there really is absolutely nothing for it to do (no threads performing any work.) This basically removes much of the performance hit of higher latency systems when using multithreaded applications, or multiple applications. Each particular thread will still be hit with the latency of a memory fetch, but the processor will be able to move on with the other threads/processes while this is happening.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

Doesn't that mean that it will provide less of a performance boost on higher-end systems?

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>

There are many reasons a processor can stall. One of them is for a memory fetch. Hyperthreading will eliminate idle-time for your processor by moving on to a different thread/process. If you have less memory latency then you will see less of a benefit when memory latency is removed from your list of concerns. However, it will never be entirely removed. The thread that is fetching from memory is still delayed. The benefit is that other threads do not have to wait for that fetch as well. Also remember that there are reasons other than memory fetches that your processor will stall, port i/o for one.

-Raystonn


= The views stated herein are my personal views, and not necessarily the views of my employer. =

Has Intel released any technical docs on HyperThreading?
I can only find PowerPoint-ish presentations at http://developer.intel.com/technology/hyperthread/

Anand just did <A HREF="http://www.anandtech.com/cpu/showdoc.html?i=1576" target="_new">this</A> article on Hyper-Threading.

"Ignorance is bliss, but I tend to get screwed over."

That's basically what I was saying, I just had to go to the bathroom really bad when I said it and didn't want to type it out fully :wink:

<font color=orange>Quarter</font color=orange> <font color=blue>Pounder</font color=blue> <font color=orange>Inside</font color=orange>

au contraire...
on higher end systems it should (note: theorising) provide a greater enhancement because higher end systems are more likely to suffer from less-than-desired bandwidth to RAM, ie. large latencies.
the greater the crippling due to memory latencies, the greater the benefit from removing those latencies.

but the overview on Hyper-threading.. see Anandtech - great article explaining the concepts behind Hyper-Threading. In particular it talks about the way we use applications and that collisions are quite common for 'standard' tasks..
Collisions, that is, within the processor.. ie. 2 threads vying for access to an ALU or an FPU for instance... in this case the management required to sort out the threads and keep everything hunky-dory creates a performance hit...

yes, that's right, it reduces performance... at least that is Anand's conclusion... this after investigating why Intel has disabled hyper-threading on the Current P4 line...

apparently, with a new bios you could potentially enable HyperThreading... and viola.. it would just work... but as stated, maybe there's a performance disadvantage contrary to everything we've been told..

So 30% improvement is possibly a myth. but more likely it's an absolute best case if your concurrent threads are perfectly matched to never conflict for resources within youyr processor.

course, I could be wrong :wink:

I spilled coffee all over my wife's nighty... ...serves me right for wearing it?!?

On aces'forum some guy posted Sandra results with HT enabled. I know, its just Sandra, but the results are pretty damn impressive. Here is the link:
<A HREF="http://www.aceshardware.com/forum?read=65029480" target="_new">http://www.aceshardware.com/forum?read=65029480&lt;/A>

I dont think intel ever said "up to 30%". I think they said anywhere from 0 to 70%. I'll be very interested to find out which aps gain 0%, and which 70% ! Either way, I cant help but love the concept of HT. Its seems so simple and efficient, I really find it hard to understand why no one else has come up with the idea (well, an implementation) many years go. BTW, has AMD ever mentioned HT for any of its forthcomming cpu's ?

= The views stated herein are my personal views, and not necessarily the views of my wife. =

Which is exactly what I said my intel family member told me, hyperthreading only helps in the event of a cache miss.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!

Balzi, the guy in my family i cornered and asked about it(hes a software engineer at intel) said that it did offer a performance benifit, but the rumors were greatly exagerated.

"The Cash Left In My Pocket,The BEST Benchmark"
No Overclock+stock hsf=GOOD!

ahh.. so it's kinda over-rated by some and under-rated by others...
Not that we should have really expected to get too much in the way of proper information BEFORE any testing and benchmarks hey??

so, do you think it makes sense that you could actually lose performance in <i>some</i> instances... even if they are rare???

I spilled coffee all over my wife's nighty... ...serves me right for wearing it?!?

I would say cache miss, branch mispredict, dependency stall and saturated issue ports depending on how intellegent the implementation is.

HyperThreading is tha abilty to work on two threads of instraction at the same time. hence utilizing "blank" excution units hence enabling Hiher IPC (Instracion Per Cycle) on current P4 and Xeons.

However - HyperThreading is far from perfect.
if the two threads try to utiliz thr same rxcution unit (ALU,FPU) it will couse some preformce Hit. in fact enabling HyperThreading today on a desktop PC will couse 0-10% preformace hit.

HyperThreading is good for corperate tasks such as rendering in which you can more or less control whta kind of threads the cpu has to handle - enabling up to 30% preformace Gain.
Also in 2003 p4 will be HyperThreaded for desktop market sence HyperThreading "Aware" programs could benfit around 10-20% via use of hyperthreading. and disabling the use of hyperthreadin if it might couse a preformnce hit.

I wouldn't be suprised if Hyper Threading appeared on Hammer. It probably won't be on Barton, and AMD probably won't have time to do an Athlon MP implementation without taking focus off Hammer.

Bah! A double post! Serves me right for clicking twice...

I doubt it very much since the two systems are based on two totally diffrent architectures and system concepts.

I found the Anandtech artical about Hyperthreading very informative (Anandtechs articals always are :smile: ). Anyway in particular him mentioning that Hyperthreading on a desktop PC will result in poor performance.

<font color=blue>In the majority of cases, if you were to enable Hyper-Threading on a desktop PC you would not see a performance increase, rather a 0 - 10% <b>decrease</b> in performance.

On a workstation there is more potential for Hyper-Threading to result in an overall performance gain, but the term workstation is so broad that it can mean everything from a high-end 3D rendering system to a heavily used desktop PC.

The area where performance gains are the most likely today is under server applications because of the varied nature of the operations sent to the CPU.</font color=blue>

So for all you home users who bought a P4 system under the impression that they will get extra performance through hyperthreading....bad luck. I suppose you will just have to stick to overclocking.

By the way WTF is going on with AMD ? Shouldn't they be sampling thier throughbred processor by now ??? It will be a great shame if it doesnt support 166FSB. Those who thought the P4 was disapointing wait till you see the throughbred. If the rumours are true it will just be a straightforward die shrink...no extra cache, no nothing.

However on the server front Hammer might have a tough time competing with Hyper threaded Xeons. Assumeing Claw/Hammer is released on time. Could someone clear up this licensing issue with SSE2 ? Does AMD need a license to incorporate SSE2 into thier Hammer line processors ?? or is someone just sh*t stirring ?....I forsee some tough times for AMD this year, oh boy its gonna be great to see how this all turns out.

<font color=purple>~* K6-2 @ 333MHz *~
I don't need a 'Gigahertz' chip to surf the web just yet ;-)</font color=purple>

Have you ever even seen a real concept live up to it's projected performance? When they say 30% or 0%-70%, I read that as marketing, not engineering.

<font color=red>God</font color=red> <font color=blue>Bless</font color=blue> <font color=red>America!</font color=red>