Testing HyperThreading in TMPGenc ***updated Mulittasking test*** A must see **added test 1/10/2004**

[Duvie]

First this is one of the programs that has the switch in the program so it is easy to see with it on and off......


I was converting a VBR Divx 5.02pro codec file (2cd rip) to DVD standard....same settings in size, fps,....

Constant quality 85 (max 6500/min1500) with motion estimate....

HT on = +24.0%

Variable Bit Rate (2 pass) (max 6500/min1500/avg4000) with motin estimate....

HT on = +23.9%

Constant quality 85 (max 6500/min1500) with High Quality motion....

HT on = +8.8%

Variable Bit Rate (2 pass) (max 6500/min1500/avg4000) with Low Quality motion...

HT on = +20.8%

Automatic Variable Bit Rate quality 85 w/ motion estiamte....

HT on = +26.2%


Take what you can out of this but it appears the p4 with HT has strengths and weaknesses....Overall still faster in every phase versus same speed and settings with HT off.


I tried many of the mpeg tools of multiplex to demultiplex but in most instances the cpu fails to even need mre then 30% cpu utilization...Maybe this is a limitation of the harddrive system since it is reading, merging or splitting, then rewriting....I will be able to verify that with my new raptor coming at the end of the week.


IN a side note I tried a few things to see how multitasking affected the performance....percentage is increase in time to encode same file...DVD standard 720x480 with Constant quality 85 (max 6500/ min 1500) with motion estimate.

Multitasking*****

<<< Tmpgenc (alone) >>>

HT on = 2:18
HT off in Tmpgenc = 2:53
HT Bios off = 2:54

HT enabled is 26.0% faster versus Bios diabled HT

<<< Tmpgenc + WinDVD player playing Shrek DVD off of HDD >>>

HT on = 2:41
HT off in Tmpgenc = 3:26
HT Bios off = 3:34 ***jerkiness of some playback unless DVD player was in foreground (did not happen elsewhere)***

HT enabled is 32.9% faster then Bios disabled HT and playable in any situation...28% faster then HT off in Tmpgenc alone suggesting the winDVD playing was benefiting from HT on in the OS.

<<< Tmpgenc + TV Tuner + Capture 720x480 DVD 8mbits/sec >>>

HT on = 3:12
HT off in Tmpgenc = 3:45 (dropped frames < 1%)
HT Bios off = 4:45 (dropped frames <1%)

Ht enabled is 48.4% faster then with HT disabled in the Bios and no dropped frames....17.1% faster then with HT off in Tmpgenc again suggesting still having the Bios enabled HT benefited in the OS environment....

<<< Tmpgenc + Winrar archiving 737files in 42folders (294mb) >>>

HT on = 2:51 (winrar = 4:52)
HT off in Tmpgenc = 4:15 (winrar = 3:51)
HT Bios off = 5:34 (winrar = 5:53)

Ht enabled is 95.3% faster in getting the encoding done while still getting the archiving done 20.9% faster....Interesting turning off tmpgenc HT took 49.2% longer in the encoding but the archiving was the fastest by 20% showing some sort of OS prioritization I dont get...Must be an efficiency thing in terms of thread management.



***********updates 11/11/2003***********



<<< Tmpgenc + TV Tuner + Capture 720x480 DVD 8mbits/sec + DVD playing Mummy Returns off of HDD>>>

HT on = 3:46
HT off in Tmpgenc = 4:28 (dropped frames < 1%)
HT Bios off = 6:17 (dropped frames <1%) (some but limited jerkiness noted in DVD playback)

Ht enabled is 66.9% faster then with HT off in bios and system was very responsive and opening folders didn't cause hiccups in any playback or capturing of frames.

<<< Tmpgenc + TV Tuner + Capture 720x480 DVD 8mbits/sec + DVD playing Mummy Returns off of HDD + Besweet Wav to AC3 (same file as below)>>>

HT on = 9:16 for TMPGenc (8:16 for Besweet)
HT off in TMPGenc = 9:35 for TMPGenc (8:02 for Besweet) (dropped frames <1%)
HT Bios off = 15:52 for TMPGenc (9:58 for Besweet) (dropped frames <1%) DVD playback major jitter and jerky...system near non responsive until besweet was done then system was more responsive and DVD was slightly jittery...

HT enabled is 71.2% faster then HT Bios off and still got the ac3 file encoded 20.6% faster and the system activity was night and day difference. HT off in TMPGenc offered a uniquie thing it was about as fast in TMPGenc (the closest it has been) while actually getting the AC3 file encoded faster though the DVD playback and capturing suffered slightly. I think the hT being disabled in TMPGenc has little effect since the overall OS is still being HT enabled and so many other apps are being run.





Fans of DVDshrink 3.0 beta 5 will be glad to know it benefits well in the HT world. Beta 5 upgrade states support for and it is clear as the DVDshrink website states that HT benefits by anaylsing and processing at same time.

DVDshrink of a 6.4gb file to 4.3gb with a 39% compression.

HT on= 17:26
HT Bios off = 21:44
HT enabled is 24.7% faster


Here is some fans of Besweet whcich uses Lame encoder as well as AC3enc....I already established that current version does not benefit from HT on or off in encoding 1 instance, but 2 instances is different....I encoded the Mummy Returns wav file into AC3 using Besweet, also coverted same 131mb wav file to MP3 224bit/s....

<<< Besweet WAV to AC3 >>>

HT on 1 instance = 4:08
HT on 2 instance = 6:57
HT Bios off 1 instance = 4:24
HT Bios off 2 instances = 8:24

Corrected....2 instances were done in 68% more time...Ht off was a bit more time as Ht on 1 instance was 6.4% faster

<<< Besweet WAV to MP3 >>>

HT on 1 instance = 0:27
HT on 2 instances = 0:41
HT Bios off 1 instance = 0:27
HT Bios off 2 instances = 0:54

2 instances completed 2 of the same file in 0:14sec more or ~50% more time. Versus Bios off the work was done in 31.7% faster...


Update....Cadd rendering test

<<< CINEBENCH 2003 (CINEMA 4D BENCH) >>>

HT on single cpu = 323 (1:22)
HT Bios off single cpu = 323 (1:22)
HT on multiple cpu = 381 (1:09)

Ht on with mutliple cpus finsihed the rendering of the scene 18% faster. It was neat to watch the screen start rendering from the top as well as 1/2 way down to the bottom. Basically split the job in half from top to bottom. Cinebench 2003 is more optimised as the 2000 version only showed an 11% gain with HT on and multiple processor test.


Updated: 1/10/2004

Another app that is HT optimised is MP3 Maker PLatinum

I incuded a 128mb wav file to mp3 high quality, 224kbits, joint stereo,48khz....

HT on = 2:32
HT off = 2:58

Represents a 17% increase over no HT....Only runs a max od 77-80% cpu utilization on my cpu so it is not using all it can...


SETI@HOME (benchmark work unit, with vers 3.08 i386-winnt-cmdline.exe)

HT on (1 instance) = 2 hours 7 minutes
HT on (2 instance) = 2 hours 59 minutes ( 1 per 1 hour 29.5 minutes )

HT off (1 instance) = 2 hours 8 minutes

Not HT opitmised but 2 instances can bring a heathy gain for SETI maniacs....

 

[Duvie]

Bump....

let me know if you want me to test something else why I have bios disabled HT.....Otherwise I am going back cause as you can see it makes a difference...tangible and real in single apps and multitasking....HT is truly an added bonus....

 

[jiffylube1024]

Nice results. Looks like HT has a significant difference after all even in today's apps!

I wonder how HT affects some popular games (like UT2k3, etc) and even 3dmark 2k1/2k3 .. care to try any of those?

 

[Duvie]

I dont have UT demo to test, and I pretty much stay away from synthetic benches like 3dmark, PCmark, sissoft and all of that..... would do the UT demo if I had it or access to a fast download of it, but I don't....


Anyone know of any mp3 encoding programs that are geared towards HT...Besweet in ac3 encoding didn't seem to change at all with any of the factors and using the latest version of the program....

 

[Shimmishim]

hey!

good stuff....

if you do a google search for ut2k3 demo you should find a good site then i've uploaded the test utility to my webpage and i can give you the link.


\just PM me.

 

[Duvie]

Originally posted by: Shimmishim
hey!

good stuff....

if you do a google search for ut2k3 demo you should find a good site then i've uploaded the test utility to my webpage and i can give you the link.


\just PM me.

Yeah but the 142mb download of the demo file minus the patches is a bit much for good ole' 56k modem guy like myself!!!!

 

[InlineFive]

Originally posted by: Duvie

Originally posted by: Shimmishim
hey!

good stuff....

if you do a google search for ut2k3 demo you should find a good site then i've uploaded the test utility to my webpage and i can give you the link.


\just PM me.

Yeah but the 142mb download of the demo file minus the patches is a bit much for good ole' 56k modem guy like myself!!!!

Somebody with broadband send him a CD of all the files and patches.

 

[Duvie]

Please do!!!!


I updated the above scores with a TMPenc + WinRAR archiving multitasking...Very interesting take a look....

 

[Shimmishim]

get with my on PM, i'll send you a copy of the demo and the benching utility (by hardocp)

 

[rogue1979]

Originally posted by: Duvie
Bump....

let me know if you want me to test something else why I have bios disabled HT.....Otherwise I am going back cause as you can see it makes a difference...tangible and real in single apps and multitasking....HT is truly an added bonus....

That's some very impressive benchmarking, looks like you put a good deal of work into it. Don't take this the wrong way, but I haven't seen any benchmarks like these, much less actually use them. I am very interested in hyperthreading benchmarks for stuff that is common for most people. Gaming, CD-burning, Seti and multi-tasking simple functions like web-surfing, MP3 playing at the same time and the more "normal" stuff.

I am curious to see what the true benefits are for my own personal use from someone on this forum, not a commercialized review that could be biased. Hardcore benchmarks are fine, but I am looking for an honest opinion on whether or not hyperthreading can actually be "felt" or "noticed".

 

[Accord99]

Originally posted by: Duvie

Anyone know of any mp3 encoding programs that are geared towards HT...Besweet in ac3 encoding didn't seem to change at all with any of the factors and using the latest version of the program....

LAME is single-threaded only, but there is a way to take advantage of HT with it. You can manually use two LAME processes running concurrently and have each encode 50% of the WAV files. The following shows the results of a test I did with a relatively conservative Dell P4 3GHz/i865PE chipset.

Original Length: 77:15 (4635sec)
Single LAME Process encode time: 218sec
Two LAME Processes encode time: 169sec
27.4% improvement in throughput.

I did the same test on a P3 933, which showed a slight decrease in performance with two LAME processes, so the benefits shown are definitely due to HT.

I used the same Lame encoder settings used in Anandtech's A64/AFX review for comparison purposes. There, the 64 bit binary was much faster than the 32 bit binary, but only resulted in an encoding speed of 20X realtime, while two LAME processes on my HT P4 show a combined encoding speed of over 27X realtime. This improvement usually also occurs in other single-threaded applications, such as WinRAR or QuickPAR, where you can gain over 20% improvements in throughput, as long as you don't get bottlenecked in the hard drive.

 

[Duvie]

Originally posted by: Accord99

Originally posted by: Duvie

Anyone know of any mp3 encoding programs that are geared towards HT...Besweet in ac3 encoding didn't seem to change at all with any of the factors and using the latest version of the program....

LAME is single-threaded only, but there is a way to take advantage of HT with it. You can manually use two LAME processes running concurrently and have each encode 50% of the WAV files. The following shows the results of a test I did with a relatively conservative Dell P4 3GHz/i865PE chipset.

Original Length: 77:15 (4635sec)
Single LAME Process encode time: 218sec
Two LAME Processes encode time: 169sec
27.4% improvement in throughput.

I did the same test on a P3 933, which showed a slight decrease in performance with two LAME processes, so the benefits shown are definitely due to HT.

I used the same Lame encoder settings used in Anandtech's A64/AFX review for comparison purposes. There, the 64 bit binary was much faster than the 32 bit binary, but only resulted in an encoding speed of 20X realtime, while two LAME processes on my HT P4 show a combined encoding speed of over 27X realtime. This improvement usually also occurs in other single-threaded applications, such as WinRAR or QuickPAR, where you can gain over 20% improvements in throughput, as long as you don't get bottlenecked in the hard drive.

NIce!!! I think I will try that.....I need to download the latest lame version.....

 

[Duvie]

Originally posted by: rogue1979

Originally posted by: Duvie
Bump....

let me know if you want me to test something else why I have bios disabled HT.....Otherwise I am going back cause as you can see it makes a difference...tangible and real in single apps and multitasking....HT is truly an added bonus....

That's some very impressive benchmarking, looks like you put a good deal of work into it. Don't take this the wrong way, but I haven't seen any benchmarks like these, much less actually use them. I am very interested in hyperthreading benchmarks for stuff that is common for most people. Gaming, CD-burning, Seti and multi-tasking simple functions like web-surfing, MP3 playing at the same time and the more "normal" stuff.

I am curious to see what the true benefits are for my own personal use from someone on this forum, not a commercialized review that could be biased. Hardcore benchmarks are fine, but I am looking for an honest opinion on whether or not hyperthreading can actually be "felt" or "noticed".

I agree...However the above test are what in fact I do.....Believe it or not I have about 30 hours of PVR I need to backup of cartoons and shows for my kid and when I am doing that I am in a constant state of converting my 50+ titles of DIVX to DVD to burn on dvd-rs.....This test is not a random thought but something I have used, tried to use, and definitely intend to use.

More so some of my basic task will be running cadd (doing work) and running tmpgenc in the background...Wen it still takes you 2-3 hours to do the type of quality conversion I want you need to maximize the PC's use...I have a couple hours a night I can spend by the PC during the week and it is nice to maximize that.

MP3 playing, burning and websurfing is so minor that often times this didn't effect my multitasking back in my 1-1.4ghz tbird days....I am a power user. Basic users HT wont change your life....

 

[Accord99]

Originally posted by: rogue1979
That's some very impressive benchmarking, looks like you put a good deal of work into it. Don't take this the wrong way, but I haven't seen any benchmarks like these, much less actually use them. I am very interested in hyperthreading benchmarks for stuff that is common for most people. Gaming, CD-burning, Seti and multi-tasking simple functions like web-surfing, MP3 playing at the same time and the more "normal" stuff.

Well, HT turns the P4 from an average processor for SETI into the King of SETI, easily outpacing any other processor in terms of daily output of work units.

 

[Jeff7181]

Does that surprised you? I'm not real surprised... well... maybe a little surprised that the temp only increase by 3C at load, and that you're only getting about a 20% increase in performance when you're able to run 2 threads at the same time. I would expect something more along the lines of 50-75% ... but maybe my expectations are too high.

BTW... if no one else delivers on the CD full of broadband goodies, let me know, I'll fill a couple CD's with a bunch of crap you can play with =)

 

[Thor86]

Nice work Duvie.

 

[Duvie]

Originally posted by: Jeff7181
Does that surprised you? I'm not real surprised... well... maybe a little surprised that the temp only increase by 3C at load, and that you're only getting about a 20% increase in performance when you're able to run 2 threads at the same time. I would expect something more along the lines of 50-75% ... but maybe my expectations are too high.

BTW... if no one else delivers on the CD full of broadband goodies, let me know, I'll fill a couple CD's with a bunch of crap you can play with =)

remember windows may see it as 2 processors but it is not 2 physical processors...Also remeber this 20-30% number is what Intel stated awhileback....So it shows he was correct and the amd clan that still to this day likes tostate this is over-exaggerated or that it doesn't happen in real world apps, or that in single apps t hurts you more is a bunch of crap....


Look at multitasking!!! I have a couple more to put up but the more you do the bigger and better the percentage gets.....



I was expecting maybe 10% but never this high since Ifigured the intel marketing would have been exaggerating a bit. To think it should hav been more would have been hard to believe since is still just one physical cpu...

 

[sonoran]

Originally posted by: Duvie
Anyone know of any mp3 encoding programs that are geared towards HT...Besweet in ac3 encoding didn't seem to change at all with any of the factors and using the latest version of the program....

Magix MP3 Maker has been HT-enabled for some time now. This shows the 2.4C about 18% faster than a 2.4B with that software: Intel benchies of Magix MP3 Maker

 

[Jeff7181]

I was expecting maybe 10% but never this high since Ifigured the intel marketing would have been exaggerating a bit. To think it should hav been more would have been hard to believe since is still just one physical cpu...

Considering the SETI@Home results I'd expect more... IIRC some people were running 2 instances of it and averaging about 2:45 per WU, and about 2:30 with one instance.

 

[DAPUNISHER]

Thanks for putting all that work into researching HT Duvie :beer: I'm very interested in how that new Raptor impacts some of the tests when combined with HT too. BTW, when can we expect that new guide for converting DIVX to DVD to be posted over@bleedinedge?

 

[Duvie]

Originally posted by: DAPUNISHER
Thanks for putting all that work into researching HT Duvie :beer: I'm very interested in how that new Raptor impacts some of the tests when combined with HT too. BTW, when can we expect that new guide for converting DIVX to DVD to be posted over@bleedinedge?

Soon..I promise!!! This upgrading ove the past 2 weeks has kept be busy.....ONce I get my raptor in my setup will be done for a bit....

 

[Duvie]

Originally posted by: Jeff7181

I was expecting maybe 10% but never this high since Ifigured the intel marketing would have been exaggerating a bit. To think it should hav been more would have been hard to believe since is still just one physical cpu...

Considering the SETI@Home results I'd expect more... IIRC some people were running 2 instances of it and averaging about 2:45 per WU, and about 2:30 with one instance.

I have heard of the neative effect of HT enabledjust haven't seen it....I am planning on doing a FH test soon.....Obviously not optimised or needs to be optimised to really take effect...

I would bet however that running the seti with other activities running full board will show HT in a favorable light. Unless you have a dedicated rig for SETI running 24/7 the HT can help by allowing you to use the computer and still get the WU done as quickly as possible with minimal slowdown.

Single app tests appear to take advantage if they are optimised for it more so then not. In multitasking it does not appear that is the case....

 

[Accord99]

Originally posted by: Duvie

Originally posted by: Jeff7181

I was expecting maybe 10% but never this high since Ifigured the intel marketing would have been exaggerating a bit. To think it should hav been more would have been hard to believe since is still just one physical cpu...

Considering the SETI@Home results I'd expect more... IIRC some people were running 2 instances of it and averaging about 2:45 per WU, and about 2:30 with one instance.

I have heard of the neative effect of HT enabledjust haven't seen it....I am planning on doing a FH test soon.....Obviously not optimised or needs to be optimised to really take effect...

Actually, I think Jeff is referring to the large benefit of HT for SETI. Using a single instance does 1 WU every 150min, with 2 instances, each instance processes 1 WU every 165min, or effectively 82min per WU, which is a massive gain. Most likely, the cause of this gain is due to the relatively poor coding of the SETI client, resulting in stalls that slows down the performance with 1 instance, but ironically enables opportunities to better utilize the P4's execution resources with 2 instances.

 

[Jeff7181]

Originally posted by: Accord99

Originally posted by: Duvie

Originally posted by: Jeff7181

I was expecting maybe 10% but never this high since Ifigured the intel marketing would have been exaggerating a bit. To think it should hav been more would have been hard to believe since is still just one physical cpu...

Considering the SETI@Home results I'd expect more... IIRC some people were running 2 instances of it and averaging about 2:45 per WU, and about 2:30 with one instance.

I have heard of the neative effect of HT enabledjust haven't seen it....I am planning on doing a FH test soon.....Obviously not optimised or needs to be optimised to really take effect...

Actually, I think Jeff is referring to the large benefit of HT for SETI. Using a single instance does 1 WU every 150min, with 2 instances, each instance processes 1 WU every 165min, or effectively 82min per WU, which is a massive gain. Most likely, the cause of this gain is due to the relatively poor coding of the SETI client, resulting in stalls that slows down the performance with 1 instance, but ironically enables opportunities to better utilize the P4's execution resources with 2 instances.

Right... It takes 15 minutes longer to do two at once as opposed to a single one.

 

[Duvie]

Originally posted by: Jeff7181

Originally posted by: Accord99

Originally posted by: Duvie

Originally posted by: Jeff7181

I was expecting maybe 10% but never this high since Ifigured the intel marketing would have been exaggerating a bit. To think it should hav been more would have been hard to believe since is still just one physical cpu...

Considering the SETI@Home results I'd expect more... IIRC some people were running 2 instances of it and averaging about 2:45 per WU, and about 2:30 with one instance.

I have heard of the neative effect of HT enabledjust haven't seen it....I am planning on doing a FH test soon.....Obviously not optimised or needs to be optimised to really take effect...

Actually, I think Jeff is referring to the large benefit of HT for SETI. Using a single instance does 1 WU every 150min, with 2 instances, each instance processes 1 WU every 165min, or effectively 82min per WU, which is a massive gain. Most likely, the cause of this gain is due to the relatively poor coding of the SETI client, resulting in stalls that slows down the performance with 1 instance, but ironically enables opportunities to better utilize the P4's execution resources with 2 instances.

Right... It takes 15 minutes longer to do two at once as opposed to a single one.

well that is just flat amazing!!! I am going compared to what Intel shows at its own site and said in the marketing brief way back when.....I think that is almost approaching 2 physical cpu speeds and that does how the SETI program is coded poorly IMO....There is no way it should alomost do 2 as fast as 1 without 2 physical chips....Consider that the far left you have to toss out to go against the far right of negative benefit you have to toss out to come up with the average...


Also 15-30% in single apps and 40-90% in multitasking is nothing to balk at and still better then anything the amd chips are offering. How many of those test in encoding did the reviews run against the AMD64 in a single apps mode had those apps in hT enabled mode in the program... I would guess not many. As pointed out above many instances I can get a benefit in programs form HT is to run 2 instances and thus you know the reviews did not do that...