TS 6.6 Render speed on Athlon Dual Core vs Pentium?

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.

TS 6.6 Render speed on Athlon Dual Core vs Pentium? // Hardware

1  |  

Post by ssutherland // Aug 7, 2006, 2:50pm

ssutherland
Total Posts: 16
I am in the process of upgrading my PC to get faster rendering. My current PC has:



Athlon XP 2200+, 1.8 GHz, 512 MB RAM, NVIDIA 6600 Graphics card.



I bought it on Ebay, so I don't know what the CACHE is.



I wanted to get a dual core system. Our IT folks at work said I should get a Dual Core Athlon, as it is superior. Best Buy said I should get a Pentium D. Searching the internet shows me that, for 3D Studio, some show Athlon faster and some show Pentium faster. The Best Buy argument seemed valid (Pentium is faster at crunching numbers...). Because I could not test different PC's at Best Buy, I purchased 2 for side by side comparison. Here is what I got:



PC 1: AMD Athlon 64x2 Dual Core 4200+, 2.2 GHz, 2000 Mb FSB, 512 Kb L2 Cache per core, 2048 Mb DDR Dual Channel RAM, Nvidia GeForce 6100.



PC 2: Pentium D 920, 2 GB PC2-4200 DDR2 SRAM, 2.8 GHz, 2x2MB L2 cache, 800 MHz FSB, Nvidia GeForce 6200SE



MY TESTS: I rendered several of the scenes that came with TS 6.6, including the one with the wooden bowl full of glass balls, the tin can that has been 'aged', and others. MULTITHREADING was enabled for both dual processor systems, and I can see each taking different pieces of the render.



RESULTS:



Computer 1 was 2-3 X faster than my original Athlon XP 2200+. The more ray tracing involved, the faster relative speed.



Here is the part I don't understand. PC 2, the 2800 GHz Pentium D, was consistently 1/2 the speed of computer 1 (Athlon dual core).



QUESTIONS:



1) What parameters of the PC most greatly affect rendering speed? I have heard lots of things, including the RAM, L2 CACHE, Graphics Card, etc.



2) Any idea why the Pentium D, with 5600 GHz of total speed, is only 50% faster than my Athlon 2200+ 1.8 GHz with ONLY 512 Mb RAM?



3) Could there be settings I am not using on the hardware to speed things up.


BTW, I kept the Dual Core Athlon system and returned the Pentium D.


Scott

Post by Alien // Aug 7, 2006, 10:12pm

Alien
Total Posts: 1231
pic
I am in the process of upgrading my PC to get faster rendering. My current PC has:



Athlon XP 2200+, 1.8 GHz, 512 MB RAM, NVIDIA 6600 Graphics card.



I bought it on Ebay, so I don't know what the CACHE is.

If you want to find out, grab yourself a copy of WCPUID (http://www.majorgeeks.com/download.php?det=435).



I wanted to get a dual core system. Our IT folks at work said I should get a Dual Core Athlon, as it is superior.

True, as you've already discovered. :)


Best Buy said I should get a Pentium D.

Wouldn't surprise me if it turned out they have a bigger profit margin on Intel-based stock.


Searching the internet shows me that, for 3D Studio, some show Athlon faster and some show Pentium faster.

There are some situations where a Pentium will come out on top, namely apps that are specifically coded to be able to take advantage of... whatever Intel's latest gimmick is [eg, MMX, SSE, SSE2, SSE3, etc], & against an AMD chip that doesn't yet have that feature the Pentium would have an advantage.


The Best Buy argument seemed valid (Pentium is faster at crunching numbers...). Because I could not test different PC's at Best Buy, I purchased 2 for side by side comparison. Here is what I got:



PC 1: AMD Athlon 64x2 Dual Core 4200+, 2.2 GHz, 2000 Mb FSB, 512 Kb L2 Cache per core, 2048 Mb DDR Dual Channel RAM, Nvidia GeForce 6100.



PC 2: Pentium D 920, 2 GB PC2-4200 DDR2 SRAM, 2.8 GHz, 2x2MB L2 cache, 800 MHz FSB, Nvidia GeForce 6200SE

http://homepage.ntlworld.com/alien42/smilies/moneybags.gif Daaaamn! I know you took 1 back, & were no-doubt intending to do so all along, but even so, I wouldn't mind having that kind of spending money.


MY TESTS: I rendered several of the scenes that came with TS 6.6, including the one with the wooden bowl full of glass balls, the tin can that has been 'aged', and others. MULTITHREADING was enabled for both dual processor systems, and I can see each taking different pieces of the render.

If you really wanna push it, try the glass toy scene at... 800x600, I think it was - took my XP3200+ 4.25-4.5 hours [it was ages ago, I don't recall the exact time].


RESULTS:



Computer 1 was 2-3 X faster than my original Athlon XP 2200+. The more ray tracing involved, the faster relative speed.



Here is the part I don't understand. PC 2, the 2800 GHz Pentium D, was consistently 1/2 the speed of computer 1 (Athlon dual core).

That amuses me, but it doesn't surprise me. ;)


QUESTIONS:



1) What parameters of the PC most greatly affect rendering speed? I have heard lots of things, including the RAM, L2 CACHE, Graphics Card, etc.

When rendering, graphics cards don't come into it, it's pretty much down to CPU & RAM, & also how they're connected on the motherboard.


2) Any idea why the Pentium D, with 5600 GHz of total speed, is only 50% faster than my Athlon 2200+ 1.8 GHz with ONLY 512 Mb RAM?

Several actually. :)

For starters on an Intel system data goes CPU -> Northbridge (http://en.wikipedia.org/wiki/Northbridge_%28computing%29) [chip on the motherboard that includes the memory controller] -> RAM -> Northbridge -> CPU.

Since the advent of the AMD64 chips, AMD has done things a bit differently [read: better]. Instead of having a seperate memory controller on the motherboard, AMD 64bit CPUs have their memory controller built in, & are linked directly to the RAM via a HyperTransport link (http://en.wikipedia.org/wiki/Hypertransport#Front-Side_Bus_Replacement). This is 1 of the main reasons why AMD64 chips outperform their Intel equivalent.


There are other advantages AMD64 chips have over Intel 1s, relating to cache & the way it's used... but I forget the exact details. :o :p


Another thing, not sure if it applies on single-chip, dual core AMD systems, or only on multi-chip systems, but back when they introduced the socket A Athlon MP for dual CPU motherboards, they came up with some kind of new algorithm for handling data on a 2-chip system - something to do with which CPU gets priority over a given bit of data if they both want it or something, I forget. The point is, AMD's had that technology since then [it's in Opterons as well], but Intel still hasn't started using it! That's partly why a 2CPU Opteron System can often [depending on the app & how it's coded] outperform a 4CPU Xeon system.


3) Could there be settings I am not using on the hardware to speed things up.

Depending on the motherboard & BIOS, there might be some settings in the BIOS that could improve performance, BUT unless you're willing to take the time to learn about them, what they do & any potential risks, I'd leave well alone IIWY.


BTW, I kept the Dual Core Athlon system and returned the Pentium D.

Don't blame you. :D

Post by chrisj // Aug 7, 2006, 11:58pm

chrisj
Total Posts: 239
pic
I've recently purchased a dual core athon 4600+, and it's pretty fast at rendering. One thing I've noticed is that it only seems to use one core in normal operation such as modelling, and the switch between modeller and player is slow. Also the undo function is ridiculaously tedious. This may be something to do with the bridge thingy. Can anyone enlighten me on whether I can make use of both cores. (Both are used during rendering).

Chris

Post by Bobbins // Aug 8, 2006, 12:29am

Bobbins
Total Posts: 506
Can anyone enlighten me on whether I can make use of both cores. (Both are used during rendering).

Chris


Multi-threading is only used during rendering and, with tS7, in the Player side physics.


When rendering there are some caveats, though:

VirtuaLight is not multi-threading.

Most third party shaders for LightWorks are not multi-threading.

Pre-processing for LightWorks is not multi-threading (shadow map generation, for example).

Post by Alien // Aug 8, 2006, 8:29am

Alien
Total Posts: 1231
pic
Most third party shaders for LightWorks are not multi-threading.

Hmmm... if there's more than 1 third party shader in use, couldn't 1 core handle 1, while the other handles another?

Post by Bobbins // Aug 8, 2006, 10:24am

Bobbins
Total Posts: 506
Not at all - that isn't how it works. If a pixel is being rendered that uses a non thread aware shader, then a second thread cannot be spawned. If a second thread is already in progress it will be held in a queue. You will see activity on both CPUs as the operating system scheduler assigns a thread to a free processor, but the total CPU useage will be close to 50% as there is only ever one thread shared between the two processors at any given time. (Note that this is a gross simplification!).


In this case it will actually render slower than on a single CPU due to the inherent overhead in scheduling multiple threads (often taken as around 24% of the available processing power in a two processor system).

Post by Alien // Aug 8, 2006, 10:49am

Alien
Total Posts: 1231
pic
I thought that on a multi-CPU/core system tS actually rendered 1 line/row of pixels per core? If so, wouldn't that mean that at any given time tS would be operating on 2 pixels at the same time? :confused:

Post by Bobbins // Aug 8, 2006, 11:29am

Bobbins
Total Posts: 506
That is true if both threads are thread-aware, which means that each thread is sure that the other thread will not adversely affect it's own execution.


In a non-thread aware situation you cannot guarantee this, so the scheduler will only allow one of the threads to execute on one CPU core at any given time. The other CPU core could be used by another application or by the operating system for it's own housekeeping duties, but not for another thread in the same application.

Post by Alien // Aug 8, 2006, 11:52am

Alien
Total Posts: 1231
pic
Wouldn't the fact that tS itself is multi-threaded make it thread-aware?


I'm beginning to think that the ideal hardware solution for tS would be a dual processor motherboard with 2 single core CPUs. That way, if you got the fastest CPU available [Opteron 256, 3.0Ghz], you have the benefit of the fastest single core CPU for situations where multiple cores are not an advantage, & still have 2 cores for situations where they are. Unfortunately, the cheapest I've seen them is almost £600 [£581], & that's just for 1 CPU. :(

Post by Bobbins // Aug 8, 2006, 12:30pm

Bobbins
Total Posts: 506
tS isn't completely thread aware, just a few little parts of it are:


LightWorks is multi-threading - but that's written by LightWork Design.

VRay is multi-threading - but that's written by Chaos Group.

tS Player Physics are multi-threading - and that is written by Caligari.


Unless LightWorks is rendering, VRay is rendering or Player physics in tS7 are running then you are not multi-threading trueSpace.

Post by Alien // Aug 8, 2006, 12:34pm

Alien
Total Posts: 1231
pic
Well, it was rendering with Lightworks that I had in mind.
Awportals.com is a privately held community resource website dedicated to Active Worlds.
Copyright (c) Mark Randall 2006 - 2024. All Rights Reserved.
Awportals.com   ·   ProLibraries Live   ·   Twitter   ·   LinkedIn