Sign in with
Sign up | Sign in
Your question

HT hinders Benchmark scores?

Last response: in CPUs
Share
August 12, 2004 10:26:19 PM

Hi all,

When browsing the MSI Mobo forums i came across a post from someone who was getting poor scores on 3dMark 03.
One of the moderators replied that the p4's HThreading function should be disabled when benchmarking due to it only allocating 50% of the cpu's power when running these tests.
I have never heard this before.Can anyone shed any light on this?

Regards...
August 12, 2004 11:42:53 PM

"One of the moderators replied that the p4's HThreading function should be disabled when benchmarking due to it only allocating 50% of the cpu's power when running these tests."

--That's crap.

_______________
STFU
Related resources
August 13, 2004 12:20:26 AM

--Thats Crap-- Lol well that certainly sheds some light on it hehe..
To be honest i'm still trying to learn more about CPu's and their various functions so i didnt know whether to believe it or not.Is their anyway that this could be true depending on what benchs you run or is having HT'ing enabled always beneficial to the p4's performance?
Thanks for the input by the way..
August 13, 2004 1:29:08 AM

They love to make excuses for poor MSI board performance. "Your power supply is too small" "You have the wrong BIOS" "You have the wrong memory" "You put it together wrong" "Thoughts of Paris Hilton are clouding your mind" And now, "Disable Hyperthreading on your P4, it's slowing your benchmarks on this otherwise unbelievably fast motherboard". It's NEVER MSI's fault! :lol: 

Abit IS7 - 2.8C @ 3.5ghz - Mushkin PC4000 (2 X 512) - Sapphire 9800Pro - TT 420 watt Pure Power
Samsung 120gb ATA-100 - Maxtor 40gb ATA - 100
Sony DRU-510A - Yellowtail Merlot
August 13, 2004 1:37:31 AM

Quote:
One of the moderators replied that the p4's HThreading function should be disabled when benchmarking due to it only allocating 50% of the cpu's power when running these tests.

Wow, that's truly BS. Can't be true.
August 13, 2004 5:45:35 AM

The statement is definately incorrect, but I understand where it comes from; if you launch a single threaded task on a hyperthreading cpu, taksmanager will only show '50% cpu usage', since the second virtual cpu stays unused. However, this means nothing, only that HT doesn't give any benefit if you run only one, single threaded task.

In some cases its true that HT can decrease performance, mostly because the L2 cache can get trashed by two threads running in parallel, but fortunately this is a rather rare occurance. I would suggest keeping HT turned on by default, but you could always benchmark whatever app(s) it is you frequently use to make sure. Either way, the difference is typically rather small (in either direction).

= The views stated herein are my personal views, and not necessarily the views of my wife. =
August 13, 2004 8:02:45 AM

My intention in posting was not to rubbish the mods knowlege or expertise but simply for my own information.
There are many applications that i do not use on this computer where what he says could have made sense.
Many thanks for the brief description of both scenarios and i will indeed run some benchmarks out of curiosity later today.
As a novice all replies and points of view are gratefully appreciated....
August 13, 2004 8:23:47 AM

-Cybercraig- Do i take it then that MSI boards have a poor reputation on these forums?
I must admit that there seemed to be a few problems when thay introduced DOT and MAT on the 865PE boards and i needed a bios flash for mine to work with my Corsair LLatency RAM.
I think it was a bit of a lottery when you bought the board as to whether these features worked ok or not.Not an ideal situation granted.
Thankfully i had no other problems with mine and everthing has worked great since..
August 13, 2004 1:47:14 PM

Quote:
One of the moderators replied that the p4's HThreading function should be disabled when benchmarking due to it only allocating 50% of the cpu's power when running these tests.
I have never heard this before.Can anyone shed any light on this?

While their explanation is obviously based on ignorance (it shows 50% because only one process is comsuming large amounts of resources in a pseudo-dual system) they still hit upon something that is <i>at times</i> correct.

A highly optimized application can be designed to literally use as close to 100% of the CPUs resources as possible per thread. (Typically using a profiler to shift around integer and fp calculations in the algorithms to maximize operation throughput and to minimize wasted CPU cycles between iterations of loops.) While HT multithreads as if it were two processors, it is most definately <i>not</i> a dual core. The result of which is that both threads in HT must share resources, and not just cache, but split up the processing units as well.

What happens when you run two threads of code simultaneously, each highly optimized to fully utilize <i>all</i> of a CPU's potential with a minimum of waste? You get two threads that are fighting for resources so badly that instead of HT running things faster, HT makes it so that these threads are now often having to wait for each other as they pass the weed back and forth before they can take another hit.

HT <i>does</i> increase performance when simultaneous unoptimized threads are in play. However when two threads designed to each use as many resources as possible are in play HT hinders performance, and in extreme cases it can be quite bad.

Some software takes this into account and won't multithread when HT is in use. Other software just sees HT as two CPUs and happily runs two threads.

So sometimes, for some applications, it <i>is</i> beneficial to disable HT. I don't know enough about the inner workings of 3dMark03 to verify if this is the case for this program or not. I'm just going by what I've seen in my own software department.

<pre><b><font color=red>"Build a man a fire and he's warm for the rest of the evening.
Set a man on fire and he's warm for the rest of his life." - Steve Taylor</font color=red></b></pre><p>
August 13, 2004 2:37:29 PM

Excellent.A very informative post.
3DMark2003 aside then i can see where this could be a help or a hinderance even though a havent come across this myself or i may have done and not even noticed it.
Is there then a great increase in performance under those conditions you descirbe when HT is beneficial? as was posted earlier when video editing or is the increase still marginal?
Also do developers have to program a piece of software to take advantage of this feature or that of any other chip or do some of these features just work better on some applications than others by nature of the way they were designed?
Many thanks for the post slvr_Phoenix,it makes interesting reaing....
August 13, 2004 2:43:31 PM

Quote:
HT does increase performance when simultaneous unoptimized threads are in play. However when two threads designed to each use as many resources as possible are in play HT hinders performance, and in extreme cases it can be quite bad.


games would be a good example, no?

-------
<A HREF="http://www.albinoblacksheep.com/flash/you.html" target="_new">please dont click here! </A>

Brand name whores are stupid!
August 13, 2004 2:47:18 PM

Do you mean that Games which are highly optimised and process a lot of data simultaneously will mean that chips using HT will suffer?
August 13, 2004 2:55:01 PM

no not really. oops i quoted in a confusing way lol!

Aquamark3 is made to benifit from HT, and it does quite nicely.


i mean that I have heard that most games suffer from HT

-------
<A HREF="http://www.albinoblacksheep.com/flash/you.html" target="_new">please dont click here! </A>

Brand name whores are stupid!
August 13, 2004 3:00:21 PM

Ah i see.Thanks for the update.
Also you have answered my other question regarding software being written to take advantage of these features.(ie)Aquamark3.Thanks for the reply....
August 13, 2004 3:11:44 PM

Quote:
havent come across this myself or i may have done and not even noticed it.

Most of the time you won't. Most of the performance gains (or losses) from HT are so small that you could just as easily see as much of a change from slower timings in your RAM. It's usually a small thing. It takes pretty extreme situations to make it enough to really notice.

Quote:
Is there then a great increase in performance under those conditions you descirbe when HT is beneficial?

Not usually, no. For some people every little bit helps though. It is especially beneficial in heavily used servers that run many small threads, which is why it was implemented in Xeons first. Most people don't run their PCs that way though. So for most people the only real benefit is that Windows XP runs so many little threads in the background that it does make Windows feel a little more 'smooth'. (Though this is a highly subjective observation.)

Quote:
as was posted earlier when video editing or is the increase still marginal?

HT really helps the most when running small threads. A bunch of small threads running together place much more nicely. In the case of Windows XP with its gaggle of background processes, this can also seem to boost the performance of single-process applications that heavily press the CPU. (Beause switching between cycles of the small threads while running the heavy thread would normally waste a lot more time without HT than it would with HT enabled.)

Quote:
Also do developers have to program a piece of software to take advantage of this feature or that of any other chip or do some of these features just work better on some applications than others by nature of the way they were designed?

Like many things in life, that answer is rather complex. Developers <i>could</i> design their software to use HT for running multiple threads in a very optimized way. Multithreading a program however is usually a great pain in the arse and not worth the small benefit that you could gain. Mostly the only developers that use multi-threading are those that either have sections of code that go dormant for extended periods of time because they're waiting for something else (like a USB device to respond) or because they know that many of their users will have computers with multiple processors.

More often however a programmer that wants to optimize their code will keep their code single-threaded instead and use a profiler to tell them how to better organize their code to maximize their usage of the CPU for a single-threaded application. The plus side of that is that it's quick, easy, and yields good results. The down side is that if you take profiling to an extreme you've just optimized for a specific processor instead of making your code run well on both Intel and AMD CPUs.

The programs that gain the most from HT are generally the programs that are single-threaded, but not optimized to extremes using a profiler. In theory a developer could specifically code for HT and gain even more, but I haven't heard of anyone bothering to do this yet.

The programs that gain the least from HT are the programs that have been extremely optimized with a profiler, because they were written to work best when only their thread running. This can go to extremes of a performance hit when programs like this have been multi-threaded <i>and</i> highly optimized with a profiler to theoretically take full advantage of every processor in the box simultaneously. When these programs see two CPUs available (and don't check to see if one of those is a logical CPU instead of a physical CPU because of HT) and run two heavy threads, expecting them to be executed on two physical CPUs, these threads fight each other for resources when run on the same physical CPU thanks to HT.

I'd guess mostly likely of programs to feel HT pains are older versions of 3D rendering softwares. (3DSM, Maya, etc.) This might also be for some games. Id is known for multithreading their engines.

And most to feel a HT benefit are database engines and file servers, but also to a lesser extent programs run under Windows that use one heavy single thread like a lot of games do, and probably video/audio encoding apps.

In theory small programs that don't max out a CPU also gain a lot from HT, but since they aren't limited by the CPU anyway, who really cares that they run faster? :) 

<pre><b><font color=red>"Build a man a fire and he's warm for the rest of the evening.
Set a man on fire and he's warm for the rest of his life." - Steve Taylor</font color=red></b></pre><p>
August 13, 2004 3:23:28 PM

Quote:
games would be a good example, no?

Very much so. When Id started multithreading their engines they had enthusiasts with dualie boxes in mind. That one day Intel released HT was a considerable detriment to these engines. (Their only saving grace was that their code was optimized for a more generic P3/Athlon useage than the highly-specialized P4, so there was still a bit for HT to boost in some ways, even if at the same time HT was hindering it in other ways.) And unfortunately many game developers license Id's engines to save themselves development time and just work on producing their own content. So it wasn't just Id's games that suffered from HT.

That actually reminds me. I haven't yet looked into Doom3's results. I wonder if Id is still multithreading and if so if they're trying to handle HT well. There's considerable potential there if you use HT well. There's also a considerable pitfall if you pretend that HT can just be ignored.

<pre><b><font color=red>"Build a man a fire and he's warm for the rest of the evening.
Set a man on fire and he's warm for the rest of his life." - Steve Taylor</font color=red></b></pre><p>
August 13, 2004 5:41:22 PM

Get down on your knees and thank the Lord that you have the 865 Neo and not the 875 Neo. MSI is trying hard to become the king of RMA's. They already are the kings of the beta BIOS. Doesn't work? "What me worry!" Sending three bad boards in a row to Anandtech was just icing on the cake for me. Why take the time to test the boards? Just ship'em Barney! Who reads Anandtech reviews anyway? Nice to know that Anantech gets the same treatment everyone else does. Screw the customer!

Abit IS7 - 2.8C @ 3.5ghz - Mushkin PC4000 (2 X 512) - Sapphire 9800Pro - TT 420 watt Pure Power
Samsung 120gb ATA-100 - Maxtor 40gb ATA - 100
Sony DRU-510A - Yellowtail Merlot
August 13, 2004 6:57:39 PM

ive always hated MSI... their video cards look cheap. there will be chips off centre, messy soldering and whatnot.

-------
<A HREF="http://www.albinoblacksheep.com/flash/you.html" target="_new">please dont click here! </A>

Brand name whores are stupid!
August 13, 2004 7:13:20 PM

My experience with MSI has been that their quality control is very poor. There is a reason that their products are so cheap. <i>However</i>, that is not to say that <i>all</i> of their products are crap. Some are quite good. It's just that MSI seems to leave it up to the consumer to decide if a part's quality is bad or not, instead of checking it at the factory. This makes MSI an extreme (and sometimes costly) gamble.

I absolutely love my MSI video card. I've seen plenty from the same batch that had fans falling off, or worse. When I ordered mine however I was just willing to use epoxy and/or thermal epoxy to fix anything broken when I bought mine. The price was cheap, the performance looked good (and has proven itself since), and the thorough software bundle was the icing on the cake. When it arrived it turned out to be just fine. No home repairs were necessary. I've spoken to others with less luck than that.

To this day I just look at MSI as a gamble. And if I consider their product I <i>always</i> hit the tech support forums and such to see what problems other people are having with that product. Sometimes it's something that I could fix or live with. Sometimes not.

<pre><b><font color=red>"Build a man a fire and he's warm for the rest of the evening.
Set a man on fire and he's warm for the rest of his life." - Steve Taylor</font color=red></b></pre><p>
August 13, 2004 10:54:03 PM

Thanks very much everyone for the feedback given to my original question.You have also answered a lot of other questions i had about HT.The advantage i'am finding with these boards is that many people who frequent it are not all gamers and hence appear to give a balanced view on the merits of certain hardware.
The above posts on this have been very informative and will be invaluable in helping me begin to learn a little more about the hardware i have and what i can expect from it.
On the subject of MSI i got the impression i had been lucky when i started to use their boards 1 year ago when i had the compatability problem with the Corsair RAM.On reading their data sheet for the board i discovered that the Mobo had only been tested with Kingston RAM and a bios flash was required to improve compatability.
As i recall one of the main problems apart from the one people were having with DOT and MAT was the liveupdate feature for drivers and bios.There were a hell of a lot of people who killed their board on trying this.I tried to flash my bios yesterday using their latest version of Liveupdate and it still failed,luckily it worked on trying it from a floppy.
Many thanks again for your time and patience in answering my queries..
!