tS7.11 Benchmarking - Quad Core

About Truespace Archives

These pages are a copy of the official truespace forums prior to their removal somewhere around 2011.

They are retained here for archive purposes only.

tS7.11 Benchmarking - Quad Core // Hardware

1  |  

Post by Shike // Apr 18, 2007, 10:54am

Shike
Total Posts: 511
pic
Ok, I got my new QuadCore computer a couple of weeks ago, and planned to compare it's render performance with my old one...but played games instead, hey, that's also a way to test performance ;)

(I can now run STALKER and OBLIVION on max settings...finally :banana: )


So for the test I made this scene(and rendered to file at 1000x1000):

(nothing fancy since I wanted it to be compatible with all engines, Phong and Mirror shaders and an array of infinite ligths casting raytraced shadows.)

5463


And if you want to compare your computer, here's a zip-file containing the trueSpace7.11 scene (3.1Mb):


5464



OLD computer: 3.2 GHz Pentium4 (early chip, before prescott I believe)

2Gb memory.


NEW computer: 2.6 GHz Core2 Quad Extreme (See signature for details)

4Gb memory (XP sees 3Gb, applications see 2Gb...will get Vista64 someday;))


Result for the different render-engines:

LIGHTWORKS with Multithread / without Multithread

Old: 2 min 53 sec / 3 min 5 sec

New: 33 sec / 1 min 55 sec


VRAY 1.0 with Multithread / without Multithread

Old: 1 min 8 sec / 1 min 10 sec

New: 14 sec / 45 sec


VIRTUALIGHT with Multithread / without Multithread

Old: 2 min 8 sec / 2 min 8 sec

New: 43 sec / 43 sec

(note, didn't include the 35 sec it took to transfer the tS data to the Virtualight process, started the clock when the render started.


Even though the old P4 listed 2 "virtual" cpu's in taskmanager/performance it sure didn't make much difference to the result !? :confused:

Old P4, Multithread on: shows CPU use 100%, both graphs maxed.

Old P4, Multithread off: CPU use 50%, each graph on 50%....and results are still that similar?


Also discovered a couple of confusing things with the QUAD core:

With lightworks and Vray it showed 100% (all 4cores maxed)with multithread and 25% without.

Virtualight however showed 50% with multithread both on/off, so it doesn't seem to care about that setting, and can only use 2 cores?


EDIT: Comparing result in a clearer way:

5502

Post by prodigy // Apr 18, 2007, 11:07am

prodigy
Total Posts: 3029
pic
COOL!!

Congratulations,

Maybe a XP bug, remember a quadcore its the lastest technology..

If i must to render a big big render, i will send to you to make me that favor :D

Cheers!

Post by Shike // Apr 18, 2007, 11:07am

Shike
Total Posts: 511
pic
Also, if someone wants to post their results, here's some details about how the test was set up.


1.

Before even loading the scene, make sure the Bridge is set like this:

(not necessary if you only plan to test Lightworks)

5465


2.

Use Camera1 in the scene.


3.

These were the render settings, and I only checked/unchecked Multithread to test the differences.

EDIT: NOTE! Antialiasing set to NONE (the icon doesn't follow the setting correctly!)

And not shown here but all VRay settings (GI, Caustics...) are turned off.

5466



4.

I rendered to file, BMP, and size 1000x1000

5467


5.

Start the clock when the actual render starts.

( at the OK-click for the BMP question about numbering the files for Lightworks and VRay, and for Virtualight when you actually see it starting to render on screen.)

Stop it when the CPU Usage goes down.

5468



Wonder if anyone will take me up on this, but have no doubt that someone soon will beat my times (any overclockers out there?):D

Post by Shike // Apr 18, 2007, 11:10am

Shike
Total Posts: 511
pic
COOL!!


Congratulations,


Maybe a XP bug, remember a quadcore its the lastest technology..


If i must to render a big big render, i will send to you to make me that favor :D


Cheers!


If you're thinking about the Virtualight scores...I wouldn't be surprised if it only supports a maximum of 2 cores? (this is actually the first time I tried that engine, so there might be a mistake on my side :rolleyes: )

Post by prodigy // Apr 18, 2007, 2:42pm

prodigy
Total Posts: 3029
pic
Tested on my:

P4 3.0ghz HyperThreading ~ Mobo INTEL Perl L ~ 1Giga Dual Channel 400mhz ~ Gforce 6600Vtd256 With DualView

Truespace 7.11 ~ 1000x1000 ~ Antialiasing = None ~ No Global Illumination


Rendered with VRay ~ 1Minute 10 Seconds.

Post by TomG // Apr 18, 2007, 3:05pm

TomG
Total Posts: 3397
Remember that some P4s are not actual dual core - the HT ones were just two virtual CPUs, but physically the same CPU, so you'd see some change from multithread on or off, but not as much as the difference between having two real cores and only using one of them.


Might not be the cause, but might be :)


HTH!

Tom

Post by Changa // Apr 18, 2007, 9:38pm

Changa
Total Posts: 187
pic
I have:

LightWorks - 1min 50 sec

Vray - 37 sec

with multithread ON


Dual Xeon 3.4, 4 gb memory.

Post by Bobbins // Apr 19, 2007, 12:05am

Bobbins
Total Posts: 506
VirtuaLight is single threaded only, so no amount of CPUs, cores, hyperthreading, enabling/disabling multi-threading or similar will make much of a difference.


Remember also that scenes may be reliant on textures which can be skewed by differences in memory requirements as much as processing power, lights that might require shadow maps to be calculated (for LightWorks shadow map calculations are single threaded) and shaders - virtually all third party shaders for tS are single threaded too.


I've run tS comparisons on systems from true dual CPU up to 8-way processors and been able to tailor scenes to run faster on one machine than another by carefully selecting lights, textures and shaders, so as always treat any benchmarks with care.

Post by parva // Apr 19, 2007, 3:13am

parva
Total Posts: 822
pic
4Gb memory (XP sees 3Gb, applications see 2Gb...will get Vista64 someday)

yepp the memory prices are very low, so 4gb are not very expensive (200€ here for 4gb mdt ddr2 mem).

From vista64 I would stay away as long as the drivers are updated. Too many companys have no actual 64bit driver and I heard a lot of trouble with vista64.

Finally gratulation for the new system shike! ;) looks fantastic and it will be definitvly a timesaver during your model work & renderings.

Post by Shike // Apr 19, 2007, 4:04am

Shike
Total Posts: 511
pic
Finally gratulation for the new system shike! ;) looks fantastic and it will be definitvly a timesaver during your model work & renderings.


Thanks ! Yep, Vista64 will have to wait until I can do more with it than browse the internet ;)

And yes, timesaver it'll be... I might even try SubsurfaceScattering again, since on my old computer it took too long for my extremely limited patience :D


@Bobbins:

Agree about the scene-influence, that's was one of the reasons I only used a rather simple scene and shaders. Textures and shadowmaps might be a good benchmark to test the Overall performance on the computer, my test was only to check the CPU-cores and verify that my new computer wasn't a big waste of money ;)


@Prodigy & Changa:

Nice to see some more submissions ! :banana:

Post by Shike // Apr 19, 2007, 11:51pm

Shike
Total Posts: 511
pic
Ok, I hope I didn't calculate this completely wrong....

5503

Seems to comply with things that I've read in other threads, that 4 cores doesn't give a 4x improvement (looking at blue column)


And if I run singlethreaded applications the gain with the new QUAD won't be that big....unless counting VirtuaLight that seem to show the biggest improvement between old and new computer in single thread....?? :confused:

(either it's highly optimized for singlethread since Bobbins mentioned it doesn't support multi, or my lack of Virtualight knowledge affected the results..?)

Post by Bobbins // Apr 20, 2007, 1:27am

Bobbins
Total Posts: 506
It's been a long held myth that going from 1 to 2 CPUs (or cores as the current trend is) will increase performance by a factor of 2. In fact there is an inherent overhead in trying to schedule work between the CPUs/cores that takes away some of the extra power.


Going back a while, the basic rule of thumb to get a crude idea of power was that doubling the number of CPUs would create a 24% loss of the available power, so:

1 CPU = 1.0x

2 CPUs = 1.52x (2 x 1.0 - 24%)

4 CPUs = 2.31x (2 x 1.52 - 24%)

8 CPUs = 3.51x (2 x 2.31 - 24%)


I don't know what the approximate factoring is for the current crop of dual and quad core CPUs, but there will be a loss in expected processing power due to the scheduling overheads. Some processor technologies and architectures scale better than others too - e.g. Itanium (IA-64) CPUs did better than the IA-32 based processors due to their design being based around multiple CPU use and the execution of explicitly parallel code.

Post by bigL // Apr 20, 2007, 10:50am

bigL
Total Posts: 16
Okay, here is the truth:


vray1: 10sec.

lightworks: 23sec.


(always multithread on)


2x Xeon Clovertown 2.33GHz, 4GB Ram, XP64


regards,

bigL

Post by Shike // Apr 21, 2007, 2:52am

Shike
Total Posts: 511
pic
@Bobbins:

Interesting, first time I've actually seen speedcalculations like that. :)


@BigL:

WOW ! That's impressive ! :D It didn't take long before I got beaten :rolleyes:


Not familiar enough with the Xeon family though (only know that our customers choose it for workstations and servers)

How come BigL's 2xXenon2.33 beats Changas 2xXenon3.4 ?

Is it the latest generation of Xenon, making it just as useless to compare GHz as between my old P4 and new Core2?

Or is the speed-increase bexause of XP64...didn't think the renderengines took advantage of that?

(If it's XP64, then it would be very interesting to see a Vista64 submission :) )

Post by bigL // Apr 21, 2007, 4:21am

bigL
Total Posts: 16
Hi Shike,


well, this comparison was not really fair. The Clovertown is a Quadro-Xeon with the newest Cores....so i have 8 Cores -)))


and its a shame, but the new intelcores are much better then the old p4 based ones.....i know this because of

my old machine with 2x Xeon 3.6GHz and i was lucky with it for 2 years, until i saw the new Renderspeeds of the newest Intel-Cores.....so i knew, i need something new.....i also was going to spend my money on a Quadro Kentsfield, but then i saw the new Xeons 5300 Series......


But as you can see, my Dual Quadro is just "a little bit" faster than yours, but costs as twice as yours......


so, have a lot of fun (and i know you will) -)))


btw1.: XP64 makes NO advantage for rendering with Truespace/Vray, yet.....


btw2.: the Truespace/Vray multithreading capabilities are not as good as i expected or as i wish. At the moment this is a little bit disappointing (see below btw3.). It seems that there are some calculation-algorithms (exspecially in Lightworks), that are just waiting on the other threads to be finished.


btw3.: try to use povray for benchmarking, then you will have fun -))))))) (and you see what good/perfect multithreading/scaling can do with your machine !!!)


regards,

bigL

Post by Shike // Apr 21, 2007, 4:49am

Shike
Total Posts: 511
pic
DUAL QUADS !? Wow, that's extreme!!!

I thought my computer was powerful, but yours...I bow before your might :D


Hopefully both Lightworks and VRay will be even more optimized in regards to multicore/cpu in the future.

They might be able to use Povray as a programming-reference ;)

Post by Jack Edwards // Apr 25, 2007, 12:29pm

Jack Edwards
Total Posts: 4062
pic
I must be doing something wrong because, got 7 sec on the Vray render at draft setting. Upping the AA to Normal, I get a more realistic 24 seconds.

Maybe there's a better way to distribute the scene file so that the user doesn't have to change render settings.

Machine:
Xeon (dual core) @ 3.3Ghz
2 GB Ram

-Jack.

Post by Shike // Apr 26, 2007, 12:52am

Shike
Total Posts: 511
pic
Hm, AA should be set to NONE (see point three in my description post above)

And all other Vray options should be off (forgot to screenshot that)


Agree that benchmarking could have been better done, like an official one with scripts that set it up correctly and actually gives the times in the end.

Unfortunately that's beyond my abilities :o

Post by Jack Edwards // Apr 26, 2007, 6:30am

Jack Edwards
Total Posts: 4062
pic
Hi Shike,


you're picture shows AA setting at draft (1/2) on Model side, but none on Player side.


-Jack.

Post by bigL // Apr 26, 2007, 11:12am

bigL
Total Posts: 16
here is another truth:


rendered with AA set to adaptive:


vray1: 31sec.

lightworks: 23sec. the same as AA=None !!


Dual Xeon Quadcore 2.33GHz, 4GB, XP64


bigL

Post by Shike // Apr 26, 2007, 11:36pm

Shike
Total Posts: 511
pic
@Jack:

Ops, I didn't notice that the menu and the icon for AA wasn't really

following eachother... though I remember reading about that in an old thread.

Edited my description field to say AA = NONE.


The menu setting seems to work for VRay and Lightworks, but doesn't update the icon, the icon however updates the menu.

So as long as it says AA None in the menu, it should be correct?


Also noted something strange... AA=None is clearly visible in VRay(staircase), but not for Lightworks? (lightworks always produce smooth edges for me)

Doesn't matter if I use Scanline or RayCast (Raycast usually gives a similar result to adaptive...at least it did in ts6.6)

Is Lightworks using Adaptive for both AA=NONE and =ADAPTIVE... would explain the results you get BigL !?


Guess I should have made more extensive testings before doing this Benchmark :(

Hope some script expert takes the challenge to make a more "official" and defenetly more reliable and comparable between the Render Engines.:o

Post by Matski007 // May 3, 2007, 2:07pm

Matski007
Total Posts: 539
pic
Ok I join in too heh


got 36seconds in VRay


Specs:

AMD Athlon 64 XC Dual Core Processor 4600+ 2.41GHz and 2GB RAM

WinXP64 SP1

Post by EyeInStein // May 5, 2007, 4:52am

EyeInStein
Total Posts: 38
27 seconds with default rendering (lightworks?) engine. Don't have vRay.

Post by prodigy // May 12, 2007, 8:47pm

prodigy
Total Posts: 3029
pic
Hi guys... i notice the vray 1.5 is slower than the 1.0... So Users with 1.0 and 1.5 can repeat the test with 1.5 to compare??

TIA! :)

Post by bigL // May 13, 2007, 2:04pm

bigL
Total Posts: 16
vray1.5 testresult (and this is not a kill joke):


19 seconds (without antialiasing)

69 seconds (with adaptive antialiasing)


vray1.0:

10 seconds (without antialiasing)

31 seconds (with adaptive antialiasing)


regards,

bigL

Post by Shike // May 13, 2007, 10:12pm

Shike
Total Posts: 511
pic
Hm, only been playing around a little bit vith VRay1.5 now....and since there are more functions now, maybe they slow it down abit?

Just a guess, and I have put "Read the manual" on my todo list to get comfy with everything new :D

( I feel lost...but in a good way ;) )
Awportals.com is a privately held community resource website dedicated to Active Worlds.
Copyright (c) Mark Randall 2006 - 2021. All Rights Reserved.
Awportals.com   ·   ProLibraries Live   ·   Twitter   ·   LinkedIn