Real world apps - clock for clock: Agena vs. Deneb

[PetNorth]

Hi all!

I’ve done a lot of real world app benches with this new baby compared clock for clock with Agena (and Kentsfield too, see at the end of the thread). I hope you guys enjoy it!

These are the basic hardware parts:

Phenom X4 9850 @2400mhz
Phenom II X4 940 @2400mhz
M3A79-T Deluxe
4GB DDR2-800 4-4-4-12
HD 2600 Pro
Samsung 500Gb

Basic software:

Windows Vista SP1 x64 (Superfetch service, disabled). CoreAVC 1.7.0 decoder for H264 stuff; MPC - MPEG-2 Video Decoder (Gabest) for MPEG2 stuff; AviSynth 2.5.7.

The benchmark suite is as follows:

1) Audio Encoding:

- dBpoweramp Music Converter R13.1, wav file (Pink Floyd classic Shine on you Crazy Diamond parts 1-5 previously extracted from CD), to:
aac (Nero encoder) 256 Kbps CBR
mp3 (Lame 3.98) 256 Kbps CBR High Quality
wma (WMA 10 Pro) 256 Kbps CBR
- iTunes v8.0.2, wav file (Pink Floyd classic Shine on you Crazy Diamond parts 1-5 previously extracted from CD), to:
aac 256 Kbps CBR
mp3 256 Kbps CBR
Apple Lossless, Automatic

2) Archiving:

- 7-Zip v4.60 x64, built in bench.
- WinRAR v3.80, built in bench.

3) Image processing:

- Paint.NET v3.36 x64, PdnBench v3.20, it's the bench from the Paint.NET developers. http://paintdotnet.forumer.com/viewt...p?f=16&t=21669
- Photoshop CS4 x64, Retouch Artists Speed Test http://retouchartists.com/pages/speedtest.html

4) Scientific:

- Mathematica v7.0 x64, MMA7.0-Test_parallel, the test used for ages to measure Mathematica perfomance. http://www2.staff.fh-vorarlberg.ac.at/~ku/karl/mma.html

5) Rendering:

- 3ds Max 2009 SP1 x64 & V-Ray v1.50.SP2 x64, cornelltestnew.max scene, the bench scene from ChaosGroup Forum (the V-Ray developers). http://www.chaosgroup.com/forums/vbu...ad.php?t=30937
- Blender 2.48a x64.
The benchmarking scene from Blender Community (grass_tutorial.blend) http://www.blendernation.com/2006/06...-render-times/ and Blender Forum link http://blenderartists.org/forum/showthread.php?t=69557
And Blender Render Benchmark v0.2 (test.blend) http://www.blendernation.com/2007/03...benchmark-v02/ and author link http://www.eofw.org/bench/
- Cinebench R10 x64, to measuse Cinema 4D perfomance.
- Maxwell Render v1.7.1 x64, sculpture.mxs scene, the benchwell scene to measure Maxwell perfomance. http://www.benchwell.com/
- Maya 2009 x64 & MentalRay, Benchmark_Maya.mb scene from Zoo Render, another classic scene to measure Maya perfomance. http://www.zoorender.com/

6) Video encoding:

- Auto Gordian Knot v2.50a. Unforgiven DVD to DivX 6.8.5 and to XviD 1.2.1 (2CD, all settings Auto). http://www.autogk.me.uk/index.php
- TMPGEnc 4.0 Xpress v4.6.2.266.
1080p HD trailer (Behind the Enemy Lines MPEG2, extracted from The Transporter Blu-Ray) to SD mp4, MPEG4 AVC, 848x480, 3.000kb/s, Audio AAC 160kbps
1080p HD trailer (Die Hard 4 H264 extracted from Master and Commander Blu-Ray) to SD WMV, WM9 Advanced Profle, 848x480, 3.000kb/s, Audio 160kbps.
- Windows Media Encoder 9 x64, RWLabs_WME_test, a conversion from a 1080p HD Windows Media file to a 720p HD Windows Media file. http://www.rwlabs.com/category.php?page=wme
- x264_Benchmark_HD_v2.0, updated to x264 r1069. I've updated the x264 codec because the two builds included are really old (eigh months old the newer). So, I've updated it to x264 r1069, one of the latest builds. http://www.techarp.com/showarticle.aspx?artno=520 and the updated files here http://www.xtremesystems.org/forums/...&postcount=293

7) Video editing:

- Premiere Pro CS4. A project with two 1080p HD trailers (Behind the Enemy Lines MPEG2, extracted from The Transporter Blu-Ray; and Die Hard 4 trailer H264 from Master and Commander Blu-Ray), exported to:
mp4 720p, MainConcept H264 Codec, HDTV 720p 23.976 High Quality preset.
avi 720p, XviD v1.2.1 Codec, HDTV profile, 5000Kbps. No audio.

- Vegas Pro v8.0c. A project with a 1080p HD trailer (Behind the Enemy Lines MPEG2, extracted from The Transporter Blu-Ray, because Die Hard trailer isn't recognized in Vegas), exported to:
mp4 720p, MainConcept AVC Codec, Main Profile, 5000Kbps. No audio.
avi 720p, XviD v1.2.1 Codec, HDTV profile, 5000Kbps. No audio.



All tests are easily reproducible by anyone.


Test results (screenshots):


dBpoweramp Music Converter AAC



dBpoweramp Music Converter MP3



dBpoweramp Music Converter WMA



iTunes (Screenshots aren’t posible. Bench is measured manually)

7-Zip



WinRAR



Paint.NET



Photoshop (Screenshot isn’t posible. Bench is measured manually)

Mathematica



3ds Max



Blender





Cinebench



Maxwell



Maya



Auto Gordian Knot & DivX



Auto Gordian Knot & XviD



TMPGEnc H264



TMPGEnc WM9 Advanced Profile



Windows Media Encoder HD



x264 HD



Premiere Pro H264 HD



Premiere Pro XviD HD



Vegas Pro H264 HD



Vegas Pro XviD HD




And finally, the summary



Well, clock for clock with all these real world apps, Deneb is ~5% average faster than Agena; For reference, here is included Kentsfield too http://www.xtremesystems.org/forums/...64#post3557964

[villa1n]

is this assuming linear scaling, IE, would these percentages be the same if both Processors were clocked at 3 ghz?

[LowRun]

Thanks for the long work there

[Oliverda]

Why did you do the benches at 2.4GHz?

[charged3800z24]

wow, nice compare. I really appreciate the time you put in. I need to get some of those benchmarks to run on mine. I only have a couple apps.

[Lightman]

Massive work again!!
Thank you very much!

[PetNorth]

thanks guys!

villa1n, I think linear scaling is equal for both processors. I'll do some test at 3.0 anyways

Oliverda, because the initial test was Kent vs Agena, and my purpose was to test them at the same clock with all the rest settings at default, FSB, HT, NB etc (my Kent is Q6600).

[demonkevy666]

thanks guys!

villa1n, I think linear scaling is equal for both processors. I'll do some test at 3.0 anyways

Oliverda, because the initial test was Kent vs Agena, and my purpose was to test them at the same clock with all the rest settings at default, FSB, HT, NB etc (my Kent is Q6600).

you should do a 200mhz increase on the NB of Agena and deneb show the % increase vs each.

[villa1n]

you should do a 200mhz increase on the NB of Agena and deneb show the % increase vs each.

Yeah, if you have the time :P It would be interesting to see the gains on the NB and how performance scales with that. Thanx for all the hardwork

[Oc-Ghost]

Great work looking forward to tests at 3GHz

[mAJORD]

Awesome work PEtnorth, as usual

the resuls vs kentsfield will probably create a stir. I'd love to see yorkfield.. someone lend the man a CPU! i say it creates a stir because some reviews show a pretty hefty, and somewhat odd advantage for yorkie's at times.. only a few percent, but could be attributed to DDR3 the're using in reviews


phII Vs York (6M & 3M) with matching DDR2 would be intersting.



I'm working on a performance comparison for NB speeds. I should make sure you're not nearly finished one of your own, but I guess a double up of results is OK at times


I'll post it in my thread later.

[PetNorth]

All comparisons are absolutely welcomed, mAJORD! waiting for yours!

Some of you guys asked for numbers with equal NB. OK, some days ago I run some of these tests with 9850 @3.0 & @2.4NB-HT. So, I've done I quick run with 940 @3.0 & 2.4NB-HT. No screenshots this time :P but table summary of course

[crazydiamond]

great stuff man

[villa1n]

All comparisons are absolutely welcomed, mAJORD! waiting for yours!

Some of you guys asked for numbers with equal NB. OK, some days ago I run some of these tests with 9850 @3.0 & @2.4NB-HT. So, I've done I quick run with 940 @3.0 & 2.4NB-HT. No screenshots this time :P but table summary of course

Very nice, so it seems the phII scales a bit better with clocks and NB than agena. Given the differences, should be even or a hair above Kents then IPC wise with a 2.4 NB?

What I find interesting, is that when it comes to games, it seems the PHII scales quite well, matching yorks and nehalem's and in some cases beating it... I guess that would come down to platform handling the instructions.. can't quite theorize what would cause that.

[informal]

Very nice, so it seems the phII scales a bit better with clocks and NB than agena. Given the differences, should be even or a hair above Kents then IPC wise with a 2.4 NB?

What I find interesting, is that when it comes to games, it seems the PHII scales quite well, matching yorks and nehalem's and in some cases beating it... I guess that would come down to platform handling the instructions.. can't quite theorize what would cause that.

Look here:
http://www.xtremesystems.org/forums/...64#post3557964

PetNorth already did the Deneb vs Kentsfield comparison and Deneb with 1.8Ghz NB is already 2% faster than Kentsfield .I'd guess with 2.4Ghz NB and looking at Agena results(Vs Deneb),Deneb would gain additional 1-2% over Kentsfield .

Good work PetNorth

[PetNorth]

Very nice, so it seems the phII scales a bit better with clocks and NB than agena. Given the differences, should be even or a hair above Kents then IPC wise with a 2.4 NB?

What I find interesting, is that when it comes to games, it seems the PHII scales quite well, matching yorks and nehalem's and in some cases beating it... I guess that would come down to platform handling the instructions.. can't quite theorize what would cause that.

Well, in my experience till now, Deneb at stock settings, is slighty but consistently faster than Kent, except some area such as audio encoding (there is some odd or irregular situation with iTunes though, IMO), or image editing.

[Stuen4y]

Really really good work!!! If you say you have Yorkfield and will do the tests with it you'll have no price.

[amdcian]

Thanks PetNorth Can you make some game benchmarking, I'm interested in it most

[Lightman]

Fantastic work PetNorth

With equal NB clocks Agena/Deneb looks more in line with my measurements of IPC scaling.

Games are amongs biggest IPC gainers comparing to Agena
Just look at my table:
http://www.xtremesystems.org/forums/...7&d=1231703017

BTW applications using SSSE3 and in case of Yorkie SSE4 will show large IPC gains compared to AMD K10. Everything else not using them is more or less on par.

[PetNorth]

WOW! great work there Lightman! It deserves its own thread and not included in a mobo thread

I think that adding games, Deneb is about 7-8% on average faster than Agena.

[marten_larsson]

Really great work!

Nice to see someone take the time to do this, really interesting results.

[ubuntu83]

That means the AM3 denebs will be equal to Yorks in clock for clock performance?

[amdcian]

That means the AM3 denebs will be equal to Yorks in clock for clock performance?

And what happen if Yorkies also use DDR3

[ubuntu83]

And what happen if Yorkies also use DDR3

In a lot of Deneb reviews Yorkies already have DDR3 memory so you can compare it there.

[informal]

That means the AM3 denebs will be equal to Yorks in clock for clock performance?

If Phenom II gains ~3% across the board from DDR3 1333 memory,it should be a close match to yorks with DDR3 ,clock for clock.
Plus there is an early word that AM3 Phenom IIs OC DDR3 extremely well ,so 1600+ Mhz should be easy to achieve with solid sticks.