My HyperThreading test. weird results.

[trueblack]

alright, recent talks of HT got me testing.
while I don't use much application using HT, I know 3dMark uses lots of it. so I did a test using 3dMark:

here are the darnest thing I got:

my set up
CPU 4.8Ghz
GTX 670 SLI

with HT off:
physics score lower. (-300)
graphics score higher. (+1000)
total score:: 25802 (avg of 3 rounds)

with HT on:
physics score higher.
graphics score lower. (????????)
total score:: 25765.3 (avg of 3 rounds)




Overall score with HT is higher, but nearly ignorable.
however, how come my Graphics score is 'consistently higher' with HT off? I thought I must be alone, or I did some mistake, so I googled it, and guess what, I am in fact not alone, I found another post from EVGA forums of a person seeing the same strange result on his GTX 690...

http://www.evga.com/forums/tm.aspx?m=1780061&mpage=1&print=true


So question here is.. does anyone know why? OS issues?
or does both cases above escape some logical loopholes?

 

[Janus67]

Try with SLI off?

 

[EarthDog]

So, which 3DMark is this (11?)?
What other benchmarks show this same response?
Have you tried this with a single card?

More testing needed...!

Also, your scores do not match the other guys INCREASES from enabling HT. He increased nearly 2k points with enabling HT. You went down but within the margin of error (contrary to what you post currently says).

 

[trueblack]

yes, I realize his 'overall' score shows HT on better.

I am mostly talking about the graphics score, I apologize if I was unclear.
which both cases see a 'significant' graphics score boost with HT off.

I will try single card later tonight and see how it relates.

upfront, i can think of a few potential reasons:
- SLI doesnt sync well with HT on?
- might HT be bottlenecking some data to GPU?

I do not know if this happens with AMD crossfire, if anyone with that setup could check, that will be even more informative.

 

[bmwbaxter]

Subbed

 

[trueblack]

while I am prepping my system, I am web-researching the topic.

got another interesting read of people with similar issues:
http://www.evga.com/forums/tm.aspx?m=1325010&mpage=1&print=true

Gaming/Benchmarking issues have long been a problem with hyper-threading. It's best just to turn it off, along with C1E, and anything else related to energy saving. This has been the case with all achitectures, not just the i series. All the way back to P4 with HT, I suppose.

http://hardforum.com/showpost.php?p=1037945973&postcount=135

- this above shows BF3 performing with issues when HT is on.
- this above has no SLI.


- strangely, all reporting this issue, has an i7 processor

 

[EarthDog]

I7 is the more common cpu with ht...then I think there is an I3 with ht? Not sure off the top of my head. But that would make sense its I7 you see as I5 doesnt have ht.

Vantage and 11 respond well to ht. I think you will find that regardless of gpu score, overall will be higher. The questions in my head is the gaming side of it. I recall some testing in the past where some rare games were hurt by ht and some it didn't respond at all while another small amount did respond well. Not sure if that testing was under SLI/cfx/single card though.

 

[trueblack]

I7 is the more common cpu with ht...then I think there is an I3 with ht? Not sure off the top of my head. But that would make sense its I7 you see as I5 doesnt have ht.

Vantage and 11 respond well to ht. I think you will find that regardless of gpu score, overall will be higher. The questions in my head is the gaming side of it. I recall some testing in the past where some rare games were hurt by ht and some it didn't respond at all while another small amount did respond well. Not sure if that testing was under SLI/cfx/single card though.

You are right, not all i5 has HT, so that observation you made there makes perfect sense. thanks for pointing it out.

and yes, right now my concern is on gaming too, cause if HT actually hurts the graphics score, (not sure if SLI/CXF related this moment.), then for i7 owners that use their computer for gaming, turning off HT for a slightly higher clock will make more sense.

 

[EarthDog]

Cool looking forward to seeing how games actually respond!

 

[trueblack]

ok. results I got:

I used bioshock as gaming reference.
differences are smaller than I thought.. but I think clear:

with HT on, SLI on
FPS 130-145
FPS appears choopy

with HT off, SLI on
FPS 139-156
FPS appears less choopy

with HT on, SLI off
FPS 85-98
FPS appears smooth

with HT off, SLI ooff
FPS 88-104
FPS appears smooth

this is just one game, hard to draw conclusions.
but initial results seems to indicate:

- HT doesn't help graphics performance
- SLI more affected than non-SLI

 

[EarthDog]

Cool. Though awfully hard to come to any conclusion with one sample... More games please!

Also, does thaw game have a canned benchmark or are these manual run throughs? Canned of course would be best for accurate and repeatable results.

 

[trueblack]

Cool. Though awfully hard to come to any conclusion with one sample... More games please!

yes, hardly conclusive, since I tested 1 game only.
could as well be game driver not good for HT !!!

but nonetheless.. initial findings.. I will test the next few games next:

CoD:BO
FarCry
AC-3
BF-3

 

[EarthDog]

Can you do this in a canned benchmark? Seems kind of questionable, a manual run through, ya know? Depending, that difference could be in the runs as its only a few fps difference and all. I also assume your testing is in single player mode correct?

Avp has a free benchmark and tool. Batman, civ v, dirt3 have built in benchmarks fit example.

 

[dejo]

most accept that an HT core is 30% efficient compared to a true core. if a game or app can take full advantage of as many cores as are offered by the cpu, then you should theoretically see a gain with HT enabled. If said game or app can only use (x) cores, then there is the chance that the game or app will get parked on HT cores rather than a true core- which would/could hamper performance.

 

[EarthDog]

I knew this was tested ad nauseum before.. here is one link I found while googling...

http://www.overclock.net/t/671977/hyperthreading-in-games

and another... http://forums.anandtech.com/showthread.php?t=2274887

MEH here is the google result... its all there: https://www.google.com/search?biw=1...dmQ&emsg=NCSR&noj=1&ei=t7HBUcG9HKrD0QHjsoCQDA

 

[False Christian]

I think hyper-threading will be useful in future games and applications once developers know how to use it properly.

 

[trueblack]

most accept that an HT core is 30% efficient compared to a true core. if a game or app can take full advantage of as many cores as are offered by the cpu, then you should theoretically see a gain with HT enabled. If said game or app can only use (x) cores, then there is the chance that the game or app will get parked on HT cores rather than a true core- which would/could hamper performance.

I did not know that.

and that is good information, makes sense.

therefore as current games mostly don't know how to use more than 4 cores, turning HT off will in fact be more efficient for games. This is also inline with my findings.

the only game that seems to be 'not affected' is BF3, HT on or off, sees no FPS change. All the rest of the games I tested sees 'similar' drift of 5-15% less in FPS when HT is ON.

sorry to disappoint Earthdog, I didn't use any benchmark tests, but I tried 5 games in total, the result is conclusive enough for me..

if anyone would like to test this on official benchmarks, that would be wonderful. I also would like to know if AMD CFX compares differently to Nvidia SLI too.