Testing HT #2....Updated a bit....

[Duvie]

UPdate:

This part is to clarify somethings that the UT2003demo benchmark erroneously did not represent real world gaming perfromance...IE FPS.....

I played 3 hours of UT2003 (man that killed me!!!) while running 1 instance of SETI@Home work unit (benchmark one).....

What I experienced using fraps benchmark was after 3 hours of benching I experienced an average fps that was 11% less then my average fps with only UT2003 running...I was using 1024x768 high detail but even 1280 res was still playable with min fps above 30.....

At same time I was able to complete the work unit in 2 hours 52 minutes....This represents 45minutes or 35% more then what I can do with one instance on this cpu by itself....

So 89% game play but still do a work unit in about 15 minutes more time then Jeff's Barton 3200+ (2.2ghz) did it....


With tmpgenc I played UT2003 with about a 16% decrease in average fps and the movie took a little less then twice as long to encode with HT enabled in OS and in the program itself....


The difference I see with these apps is that tmpgenc with HT enabled within the program itself takes 100% cpu utilization...whereas both 1 instance of seti and UT2003 only take 50% of cpu utilization with HT on...So neither are HT optimised...This may be why the first one didn't hav as much affet on the main app as well as not as much effect on the 2nd background app....


Like I mentioned below with HT enabled in the OS setting or changing priority manually inthe task manager as little or negative performance effects...can someone verify this as well....Basically HT becomes the controller...As some have mentioned if one could give high priority to UT2003 so that the average fps would be 2-3% less many would gladly take the seti unit taking 50% more then original....









Ok like title states....I don't game but I know many wanted to see HT apps run with not all rendering or multmedia apps like HT testing part #1.....Some guys mentioned multitasking with apps like I show in the testing....

I only have a few games but CSI and Flight Sim 2003 don't really help me in numbers as UT2003demo benchmark does...So I used that...If anyone has HALO or quake3 demo they burn for me I could test that...

PART #2a
Hey I have been doing some gaming + application with HT on and off as a research test....

From what I can see so far is that with HT off my utdemo2003 score (botmatch) plus any other app is usually equal or better with HT off though the 2nd app has blazing performance with HT on...

Basically HT off the absolute priority goes to UT2003demo and after running benchmark 4 times in a row for about 6 minutes work the seti file only logged 70 seconds of cpu time as opposed to 360sec...It only operates 1 out og 5 seconds so it appears HT off and gaming wont share.....

I ran:

UT2003demo by itself (800x600) I have a non gamer card...Radeon 8500 AIW

HT on = Botmatch = 83.35
HT off = Botmatch = 83.47


UT2003demo + TMPgenc file (the one I normally get 2:08 with only it running)

HT on = Botmatch = 63.33 (but tmpgenc file was 68% completed)
Ht off = Botmatch = 82.50 (but tmpgenc file was only 16% completed)***basically will take about 3-4 times longer in the encoding***


UT2003demo + DVD burning ***DVd cpu utilization at 1x DVD-rw only takes like 6-10% max cpu utilization anyways***

HT on = Botmatch = 82.67 (no underruns and 100% buffer)
HT off = Botmatch = 81.76 (no underruns and 100% buffer)

UT2003demo + Pinnacle 8.8 Video Capturing (High quality DV) ***using IEEE1394 interface and download a ~30min xmas day recording***

HT on = Botmatch = 78.8 ( no dropped frames)
HT off = Botmatch = 76.60 ( 10% dropped frames)

UT2003demo + Pinnacle 8.8 Video Capturing (MPeg2 DVD standard encoding OTF) same file as above

HT on = Botmatch = 73.7 ( no dropped frames)
HT off = Botmatch = 73.2 ( no dropped frames)

UT2003demo + SuperPI 32mb

HT on = Botmatch = 58.46 (1 min 37 sec per loop...I usually get 1 min 28 sec) ***remember superpi alone only runs a 50% max cpu***
HT off = Botmatch = 27.82 ( 2 min per loop ) brought the demo to its knees...though superpi was respectable...

UT2003demo + SETi bench I run in my thread

HT on = Botmatch = 60.16 (seti cpu time was virtually identical to time I ran 2 sets of demo 5% complete)
HT off = Botmatch = 82.94 (seti cpu time was 70sec after 4 runs or about 6mins and only 2% done)

Looking at the numbers it will take the HT off one 5 times longer on same work unit...I had to run 4 sets of the demo on the HT off because the state.sah file does not report progress immediately until it gets to like 2%....


What do you think of these numbers...I need to run them on the Barton as well...



EDIT: This is all default priority I did not change anything from default.....I let the sytem decide priority...

 

[THUGSROOK]

i usually dont multitask while gaming.
sometimes ill be dloading while playing.... (cable connection)
so i guess id be interested in those results.



priority might be a concern in these tests tho.
granted most ppl dont set program priorities, but it may be hindering the tests, and the reason why multitasking is difficult w/o HT to begin with?

(i can set priorities in w98 thru reg, no gui)

 

[myocardia]

Well, I guess that seals the deal, my next system will be a P4C. Thanks Duvie, you do more for us than our parents did!

 

[Duvie]

That is what I was wondering...on the priority that is....That is also why I tried not to muck around with it since I rarely do with my own system....I was thinking about priority in the HT on system to see if it helps in the gaming app and still give you a bit of a perofrmance increase on the 2nd app....


The real question is when gaming is the 2nd app merely a non time issue??? Does one care about how fast it ges done like seti or superpi (number crunching type things)???

I see the one real benefit in the testing was the gaming + the capture and the gaming + DVD buring...


Any other things you would like to see thugs???


I need some more gaming test but no way am I paying 40-50 for something I wont play competitively...Purely testing only....

 

[THUGSROOK]

i would think that anyone multitasking while gaming, would be more interested in the game running well. thats prety much the main reason not to multitask while gaming. (especially with newer DX9 games)

id also be worried about having errors in the TMPG encode or the DVD burn.

what does UT bench w/o any multitasking for you?
you didnt post those results

does UT in of itself support HT?

 

[myocardia]

Duvie, I for one play only FPS's and flight simulators, so I would like to see how FS2002 runs while folding proteins, specifically.

 

[Thor86]

I play BF1942/Diablo2LOD while crunching SoB (2 instances) at the same time, and still do not get any system hiccups.

 

[Duvie]

Originally posted by: THUGSROOK
i would think that anyone multitasking while gaming, would be more interested in the game running well. thats prety much the main reason not to multitask while gaming. (especially with newer DX9 games)

id also be worried about having errors in the TMPG encode or the DVD burn.

what does UT bench w/o any multitasking for you?
you didnt post those results

does UT in of itself support HT?

It is posted there...It is the first set of numbers....

 

[Duvie]

Originally posted by: Thor86
I play BF1942/Diablo2LOD while crunching SoB (2 instances) at the same time, and still do not get any system hiccups.

Other then superpi with HT off and the fact that in most of the intense apps the game usually slows the botmatch to 70-75% of original with HT on...However none of the demo seemed laggy with 70-75% of botmatch score......NO hiccups either with HT on....

I have the game and I played bombing run for an hour while crunching seti and when I was done the seti program was more then half way done....I ran 2 instances of seti alone and I got 2 units of the bencmark done in 3 hours versus 1 unit in 2hours and 7 minutes with one instance....


too bad the game doesn't have fps counter somewhere on the screen for me to watch as I play...Does it???

 

[Technonut]

too bad the game doesn't have fps counter somewhere on the screen for me to watch as I play...Does it???

Fraps

 

[Confused]

Originally posted by: Duvie
too bad the game doesn't have fps counter somewhere on the screen for me to watch as I play...Does it???

FRAPS


Confused

 

[Duvie]

I am currently reruning some of the test using that new program fraps..not using the demo benchmark but actually playing death matches and capture the flag....I started the testing and then realized I would rather use F11 key and have it calc the average frame rate.....The first 2 test I just looked that the fps up in the corner and would look for the lowest fps....


I switched to 1024x768 with default detail.......lowest frames per sec was about 54 but averaged around 80-90....I got the encode done in 4:25 or slightly more then twice the time if it was running as a single app...

At 800x600...lowest fps was 60 and averaged around the same 80-90...I got encode done in 4:10 or slightly less then twice the time.....

At default in both instances the lowest fps with ut2003 alone was about the same or withn 1-3fps as the mulitaskin operation....

In my condition the vid card is the limitation obviously......


But remember when you look at the data it is about comparable with HT on and off.....If the knew testing above holds to old testing then I expect to have similar fps at 1024 but have the clip take 10-15 minutes or 5 to 7x its time ran alone....


Edit: Also I only have one machine and I would only game or run flight simulator to feel the time during and encode and to me doing that by sacrificing time is not worth it with the HT off...My average vbr 2 pass encode may take 4 hours so 6 hours isn't too bad if I game for say 2-3 hours of it.....At least after gaming for 2-3 hours I know it is at least close to half done with HT off it appears I would only be maybe a qaurter done....

I would also as a single pc user would feel more inclined to run some seti or FH if it didn't effect my usage as greatly....So I will be running some more testing based on seti as a full time background user....

 

[mechBgon]

Suggestion: Gaming benchie with a full no-holds-barred antivirus scan running in the background. I mean a virus scan with heuristics on, compressed files scanned to the max, ALL files, no exceptions, full CPU priority if the AV software has a setting for it... hardball, not Nerf.

IMO this is THE definitive multitasking benchmark... systems all over the corporate world do it on a daily basis, or they should anyway (the virus scan, not the gaming ). You haven't seen serious antivirus protection until you've seen fully-armed McAfee VirusScan Enterprise chewing on a ~500MB self-extracting .EXE file that's full of .CAB files. :evil:

 

[Jeff7181]

My main interest would be running SETI@Home while playing a game. Looks as though with HT enabled performance drops about 25%... which I consider to be unacceptable. I'm assuming by default your S@H client priority is set to normal, where mine is always set to low by SETI Driver. Maybe try setting it to low and do the test again with HT enabled?

 

[THUGSROOK]

Originally posted by: Duvie

Originally posted by: THUGSROOK
what does UT bench w/o any multitasking for you?

It is posted there...It is the first set of numbers....

ah ok missed that.

in that case ~ HT doesnt look that good for gamers.
like i said before, gamers will be more interested in the game running well, not the background task. most gamers will be happy to shutdown all background tasks to get better gameplay.

and again, id be worried about errors in the background task.

 

[Wolfsraider]

nice

thanks duvie

I run seti and game constantly.I refuse to turn off seti when gaming and have yet to run a game that requires it.

the tests with the avi (dropped frame rates) are very interesting,one question though...

can you run 2 seti instances and ut2003 with ht enabled/disabled?....thats what i would be doing if i can justify the upgrade.

I game about 5 or more hours a day
work on my website a lot lately-having as many as 30 windows open(512 mb ram tops out here)
run seti 24/7
burn cd's
record tv shows -have to run reduced quality or risk dropped rates encoding.
etc...

Wife says i can upgrade but looking around i may downgrade MHz for an overclocking p-4c and add a new mobo and ram.

thanks
mike

 

[FPSguy]

I see a few issues with your tests:

1. As you note, your video card is your bottleneck when it comes to high-FPS games. You have a processor that is running faster than the vast majority of people and an FSB that is faster than the vast majority of people, and then you have this Radeon 8500 that is a couple generations back from many hard core gamers' video cards. I think the test would be more telling if a hardcore gamer's video card were used and if the CPU and FSB were both a little slower (say, CPU at 3.2GHz). Not that anyone would expect you to buy a new video card for this test. I'm just saying that the results from your system don't necessarily translate well to a gamer's system.
2. Turning off hyperthreading may negatively affect the performance of the game itself. Few if any games take true advantage of multiple processors today, but more will be doing so in the future and even if the game is not set up to take advantage of multiple processors, I seem to recall reading that hyperthreading still can speed up a game. Your test shows this to a very limited extent, but because your video card is bottlenecking the FPS rate the difference is very small. For people with faster video cards, the difference could be more pronounced.
3. I think most FPS gamers are not very interested in what is going on in the background, except to the extent of having as little as possible going on in the background so they can get every possible advantage in terms response time.

 

[Jeff7181]

can you run 2 seti instances and ut2003 with ht enabled/disabled?....thats what i would be doing if i can justify the upgrade.

Ummm... what makes you think three threads, 2 of them being CPU dependant, would be any more "playable" on a HT setup than two threads, 1 of them being CPU dependant on a setup without HT???? If anything it would be even worse because as Duvie's tests CLEARLY show, having HT enabled is no where near the same as having two processors.

 

[Wolfsraider]

yep i know

but if it runs it thats great for me

as stated it would be about 5 hours a day maybe more,if it could handle decently.

if not i am looking at a dual proc system-more money etc

just curious since hee hee he could test the waters a bit maybe ....please


mike
edit>
seti only takes 50% on p-4c,so many are running two threads, but if i could still game that was my question

i run 1 instance @100% now and still game-p4b

so i was curious as to its possibility

 

[Jeff7181]

Originally posted by: Wolfsraider
yep i know

but if it runs it thats great for me

as stated it would be about 5 hours a day maybe more,if it could handle decently.

if not i am looking at a dual proc system-more money etc

just curious since hee hee he could test the waters a bit maybe ....please


mike
edit> seti only takes 50% so many are running two threads but if i could still game that was my question

i run 1 instance @100% now and still game

so i was curious as to its possibility

I'd have to say no... look at the results with just 1 instance of S@H running... a 25% loss in performance is not good.

 

[Wolfsraider]

ok
i see what your saying

i am a little slow i guess

say you were running a 15000 rpm drive a 9700 pro and a 2.4/2.6 oc to 3.2

you still think it wouldn't be playable running two instances?

if no then i guess i'll wait to see if the prescott makes a diff.if not looks like dual proc will be the only way to do this,right?

mike

 

[THUGSROOK]

you should probably build a 2nd system

 

[Jeff7181]

Originally posted by: Wolfsraider
ok
i see what your saying

i am a little slow i guess

say you were running a 15000 rpm drive a 9700 pro and a 2.4/2.6 oc to 3.2

you still think it wouldn't be playable running two instances?

if no then i guess i'll wait to see if the prescott makes a diff.if not looks like dual proc will be the only way to do this,right?

mike

Absolutely not... it has nothing to do with the hard drive... you can put 500 15,000 RPM SCSI drives in RAID 0 and it won't make a bit of difference how the game runs.

It has nothing to do with the video card either because SETI@Home doesn't use the video card.

From Duvie's previous results it looks like you can hope for a maximum of about a 50% increase with Hyper Threading enabled... and that's being VERY generous for a game since games require integer and floating point calculations.

I would guess for the best efficiency, you should run one thread per virtual CPU... so... if you had Hyper Threading on your current setup, your games would probably be faster, or, you'd finish S@H WU's in the same amount of time regardless of whether you were playing a game or not, possibly a few more minutes than if it was just sitting there doing nothing but S@H.

I can run S@H and play a game at the same time too... but I can play for about 3 hours and the progress on the SETI WU doesn't increase by more than a few percentage points because most of the CPU is being used for the game, and the SETI client is set to low priority, so it lets everything else have first crack at the CPU and takes whatever's left over, if there is any.

So yeah, don't expect to run 3 threads, ESPECIALLY when two of those are so heavily dependant on CPU power... cause even with a dual CPU rig it won't perform that great if you try and run 3 threads that require a lot of CPU power.

 

[Wolfsraider]

Originally posted by: THUGSROOK
you should probably build a 2nd system

heh heh if i build a second system this one will go to the wife and i still get to crunch on it

but at 11 to 15 wu's a day on seti using 2 networked computers adding a dual proc ht enabled would allow an additional 20+ wu's

decisions decisions

wife is hooked on getting this system so she wants me to upgrade woot and it has been a year and a half (by the time i upgrade....


oops back on topic-still curious though
sorry duvie
mike

 

[Wolfsraider]

Originally posted by: THUGSROOK
you should probably build a 2nd system

heh heh if i build a second system this one will go to the wife and i still get to crunch on it

but at 11 to 15 wu's a day on seti using 2 networked computers adding a dual proc ht enabled would allow an additional 20+ wu's

decisions decisions

wife is hooked on getting this system so she wants me to upgrade woot and it has been a year and a half (by the time i upgrade....


oops back on topic-still curious though
sorry duvie
mike