Video encoding benches with Hyperthreading Vs non HT

[oldfart]

I was doing some video encoding work last night with TMPGEnc (http://www.tmpgenc.net/e_main.html). For those that are not familiar with it, TMPGEnc is one of the best MPEG video encoders available. I downloaded the new 2.520 version to see whats new. It's always had multi thread support, and now supports hyperthreading as well.

I encoded a 4:25 avi file to SVCD using 2 pass variable bitrate encoding, high quality motion search precision. The template I use is my own to fit 1 Hr of video at best possible quality on a 700 Meg CD.

TEST SYSTEM
Abit IC7 | P4 2.6C @ 3.34 GHz | 257 FSB | 5:4 ratio | DC DDR 412 | cas 2-2-2-6 | Win XP Pro SP1a

Hyperthreading enabled
11:52

Hyperthreading disabled
13:03

This is a 10% improvement in encoding time.

It takes
2.687 x video source length to encode with HT
2.955 x video source lenght to encode without HT

If you do the math, a 1 Hr source encode:
HT = 2 Hr 41 min
non HT = 2 hr 57 min
Improvement in encoding time = 16 min.

Hyperthreading does provide a nice boost in this application. It would be great to see more programs optimized for HT.

I'll try some other templates such as a CBR to see if the same type type of numbers are there.

EDIT Added benches

Here are some benches using templates that are included with TMPGEnc:

DVD
HT = 4:36
non HT = 5:11
13% improvement

VCD
HT = 2:28
non HT = 2:52
16.5% improvement

SVCD
HT = 3:48
non HT = 4:37
22% improvement :eek2: :eek2:

[Bobbyeyes]

WHAT????:-)
no synthetic benchmarking....
just a real world bechmark??
:-)
nice work Oldfart

[oldfart]

LOL

I am VEHEMENTLY OPPOSED to synthetic benchmarks. You wont see me wasting my or anyone else's time with them. :rolleyes:

I care about programs that I ACTUALLY USE, not xyzMark, synthetic mem benches that dont represent any real world performance.

Here are some benches using templates that are included with TMPGEnc:

DVD
HT = 4:36
non HT = 5:11
13% improvement

VCD
HT = 2:28
non HT = 2:52
16.5% improvement

SVCD
HT = 3:48
non HT = 4:37
22% improvement :eek2: :eek2:

Wow! HT rocks with this program!!
:rock: :rotflmao:

[pointreyes]

Originally posted by Bobbyeyes
WHAT????:-)
no synthetic benchmarking....
just a real world bechmark??
:-)
nice work Oldfart

:lol:

Wait! :stop:
Real world benchmarking...is....is....is that possible? :scratch: :wack:
You mean that we can actually trust this? :eyepop:
I thought that only synthetic benches were valid-to use real world benchmarks is amazing! :p Wow! Thanks for these type of benchmarks, Oldfart. :)

BTW: I'm with you with doing real world benching. If Oracle does not work quickly (and accurately) enough then the proc/mb/ram is a complete waste of my money.

[Bobbyeyes]

OMG...
yea...
what is this forum comming to!:wave:

i could not agree more...
i remember about 3.5 yrs ago i had(and still have) a slot 1
on the famous asus p3bf
i had not upgraded anything for a year, but was upgrading my shareware sissandra prog.
each upgrade, my scores got better and better!
but for an enthusiast/hobbyist.. i guess it is a staple.

[toymaker]

Very interesting, Oldfart. And thanks.

Real world performance. What a concept, eh? :idea: ;) :D

[shimmishim]

where's thugs...

and his... "my cheapo setup gets better mem scores than yours at synthetic benchmarks!!!!"

hahaha...

nice post oldfart!

[oldfart]

This thread was intended to show that HT does have a benefit when the app is coded for it. TMPGEnc seems to benefit quite a bit. Another video app I use, Pinnacle Studio 8. does not support HT at all. I wish it did since it is pretty slow compared to TMPGEnc, but I like its editing and menu features.

This was not a real world Vs synthetic bench thread (my position is pretty well known on that already) or a HT CPU vs non HT CPU system statement of any kind, nor to compare this setup to anyone else's.

Its just a thread to show how HT helps in this one particular application.

[ThugsRook]

Originally posted by shimmishim
where's thugs...

and his... "my cheapo setup gets better mem scores than yours at synthetic benchmarks!!!!"

hahaha...

nice post oldfart!
im right here shim, and when you find a way to test memory performance in a real world application reguardless of cpu ~ you let me know ok?

[shimmishim]

Originally posted by ThugsRook
im right here shim, and when you find a way to test memory performance in a real world application reguardless of cpu ~ you let me know ok?

sure thing! :p

[toymaker]

Now Now. Thugs' reviews are always balanced, informative and often entertaining.

Some of the reviews that I've seen in other forums and websites are quite the opposite.

[shimmishim]

Originally posted by toymaker
Now Now. Thugs' reviews are always balanced, informative and often entertaining.

Some of the reviews that I've seen in other forums and websites are quite the opposite.

i know... i was just being a punk :)

someone's gotta give him a hard time!!! or else what fun would it be around here? :wave:

[LastRide]

Thats it shimm.Lay into that Thugs character real good.Make sure you bitch slap him real good..LOL..LOL.:na: J/K

[shimmishim]

Originally posted by LastRide
Thats it shimm.Lay into that Thugs character real good.Make sure you bitch slap him real good..LOL..LOL.:na: J/K



:wack: :rotflmao: :scratch: :beer: :eek2: :blush: :yummy:

[Bobbyeyes]

AW... cummon men.....lets get back to the ranch......
i am just getting into video editimng and oldfart has some info to share

[pointreyes]

Originally posted by Bobbyeyes
AW... cummon men.....lets get back to the ranch......

Oh give me a home
where the buffalo *burp* ro...ro....roammmmm.... :beer: :p

Sorry I could not resist.

<snip>test memory performance in a real world application reguardless of cpu ~ you let me know ok?<snip>

That's easy-with the Wife program-memory will always fail or perform miserably. Without the Wife program-memory will excel at relative things like the ns ratings on all chips on all the components within a person's computer. :p :)

Whoops, it's not Friday yet. :D

[ThugsRook]

yes please, I am not the topic here ~ HT testing is :-r


oldfart ~ how much cpu mhz does it take to make up for HT?
in other words ...whats it really worth?

[LastRide]

There is no queation HT gives you a little speed boost.I noticed a 200-300 point boost on 3DM2K1 with HT enabled.Whether its enough to persuade some people is another story (Thugs :))eva2000 had some encoding numbers earlier only they were a little more drastic.Can't remember what the test platform was and CPU used but I remember he had the C chip overclocked higher.

[pointreyes]

Originally posted by ThugsRook
yes please, I am not the topic here ~ HT testing is :-r


You mean HT does not mean Harassing ThugsRook? :look: :scratch:

Ok, seriously here's something else I'm curious about with HT. Does HT allocate more RAM? e.g. A single physical proc requires 100 Meg for a task but a HT proc requires say 175 Meg for the same task. If so, would this not cause an increase in performance? Yes, the extra threading helps but surely the possible increase in RAM helps as well. I'm kinda of thinking in lines of a dually box-it is has been suggested in the past that you have two sticks of RAM for a dually so that one stick is sort of allocated to each proc. Even though HT uses resources differently from a dually system, could the RAM be treated the same way between these two different technologies? Any ideas on what I'm trying to figure out here?

[SupermanCK]

NICE...

just had a suggestion

how fast do you have to have ur set up in order to catch your HT enabled test? or visa versa...

like X mhz(w/ HT) to Y mhz(w/o HT) = same test time.

[oldfart]

Not ignoring this thread. I've been tied up (darn job gets in the way) and haven't had a chance to reply. I'll get to the questions maybe tonight.

[oldfart]

Originally posted by ThugsRook
yes please, I am not the topic here ~ HT testing is :-r


oldfart ~ how much cpu mhz does it take to make up for HT?
in other words ...whats it really worth?
Ok....to summarize the testing I did earlier, here are the performance gains HT provides with 4 different encoding settings:

My custom SVCD settings
HT = 11:52
non HT = 13:03
10% improvement

DVD
HT = 4:36
non HT = 5:11
13% improvement

VCD
HT = 2:28
non HT = 2:52
16.5% improvement

SVCD
HT = 3:48
non HT = 4:37
22% improvement

I benched the SVCD setting to see how much CPU HT is worth.

2.78 GHz 213 FSB with HT = 4:37 encode time
3.34 GHz 257 FSB no HT = 4:37 encode time

It is "worth" 560 MHz. The scaling looks pretty close. The SVCD encoding showed a 22% increase with HT. If that is translated into CPU speed, 2.78 GHz + 22% = 3.39 GHz. Almost exactly the 3.34 GHz non HT speed needed to equal the HT benchmark.

[oldfart]

Originally posted by ThugsRook
im right here shim, and when you find a way to test memory performance in a real world application reguardless of cpu ~ you let me know ok?
I dont understand this one. I've been testing this way all along. Tweak your memory settings, run some real world benches, post the results. Same thing you are doing now. Just use different benches.

The problem with the memory benches is they show unrealistically large gains. I have yet to a single real world bench that shows anything even close to the same numbers. The synthetic benches show 10 - 30% gain which typically is <1% real world. This is why I (and many others) consider these 'benches" completely worthless and an utter waste of time.

[Bobbyeyes]

Oldfart is actually one of the few that actually presents real world tests results
just a thought, but, if this has not been done before,
would you, at your leisure, be willing to retard your memory, and run the same tests?(w/ hyper threading enabled)-

if tested, this may qualify for a 'real world' memory bench.
video editing will be one of the next big thing to do w/ puters
just like music WAS (and still is)

[Duvie]

Oldfart I love the review and this was a great test to show some of the HT gains (albeit not to many in single app'd programs unless optimized for it)....Can you run some test say like doing the encoding and some other test (maybe playing half-life2) and see how the performance does???