Mon, Dec 23, 7:32 PM CST

Renderosity Forums / Poser - OFFICIAL



Welcome to the Poser - OFFICIAL Forum

Forum Coordinators: RedPhantom

Poser - OFFICIAL F.A.Q (Last Updated: 2024 Dec 23 7:16 pm)



Subject: Poser and single core processor vs quad core speed?


drifterlee ( ) posted Sun, 01 July 2007 at 1:00 PM · edited Tue, 03 December 2024 at 9:45 PM

There is a Intel® Core™2 Quad Processor Q6600 Gateway on sale right now with a good graphics card. How much faster will Poser render on a quad core as opposed to my 3.4 GHz P4 singe core with 2 gigs of RAM, or will it make any difference? Thanks


Gareee ( ) posted Sun, 01 July 2007 at 2:08 PM

Well, if the cores are all faster then what you currently own, probably about 3-4 times your current rendering speed. What I found when I upgraded though, was that I started always using higher quality setting for renders, because they didn;t impact my rendering speed much any longer.

Way too many people take way too many things way too seriously.


drifterlee ( ) posted Sun, 01 July 2007 at 5:08 PM

It has a 512 Radeon crossfire Graphics card in it.


svdl ( ) posted Sun, 01 July 2007 at 8:09 PM

Significantly faster. I've got a Q6600 myself, with 8 GB of RAM running XP 64 bit. Not only is the rendering fast, it can also render scenes of much higher complexity. 10 Millenium 3 figures, all with different hires textures, hair, clothes, environment - it eats 2.5 Gb, but it renders! Default render settings, default lighting - a couple of minutes and it's finished.

That Radeon might be a problem. The older Radeon cards have lousy OpenGL support, and XP 64 bit isn't supported at all.

The pen is mightier than the sword. But if you literally want to have some impact, use a typewriter

My gallery   My freestuff


drifterlee ( ) posted Sun, 01 July 2007 at 8:30 PM

It's a new Radeon card - the Cross fire.


fiontar ( ) posted Sun, 01 July 2007 at 8:40 PM · edited Sun, 01 July 2007 at 8:41 PM

Quote - It's a new Radeon card - the Cross fire.

512 Radeon Crossfire isn't a model name. Madel numbers for processors and everything else seems to be more confusing than ever before, so don't worry about making the mistake, but if you can find out the actual model number we can help you evaluate it. :)

I've actually heard people say that the new Radeons have better OpenGL than NVidia, which is a turn around from a couple years ago.


svdl ( ) posted Sun, 01 July 2007 at 8:43 PM

Can you give a link to that machine? Then I will know what's in there, and if the graphics are any good.

The pen is mightier than the sword. But if you literally want to have some impact, use a typewriter

My gallery   My freestuff


fiontar ( ) posted Sun, 01 July 2007 at 8:46 PM

Quote - Significantly faster. I've got a Q6600 myself, with 8 GB of RAM running XP 64 bit. Not only is the rendering fast, it can also render scenes of much higher complexity. 10 Millenium 3 figures, all with different hires textures, hair, clothes, environment - it eats 2.5 Gb, but it renders! Default render settings, default lighting - a couple of minutes and it's finished.

That Radeon might be a problem. The older Radeon cards have lousy OpenGL support, and XP 64 bit isn't supported at all.

Extra cores tend to add diminishing returns in most apps, but the way the Firefly renderer breaks up the rendering process between cores, it might be a better gain than for other applications.

Would you mind timing the renders using 1 core, 2 core and 4 core? It would be helpful for a lot of people trying to decide if 4 cores is the way to go at this point in time. (You can select Poser.exe in the Windows Task Manager Processes tab, right click it and under "set affinity" turn off cores for that process. Make sure you aren't rendering with the "seperate process" option in Poser and time the renders).

It would really be appreciated by myself and I'm sure a lot of other people as well. :)


svdl ( ) posted Sun, 01 July 2007 at 9:03 PM

I don't have any Poser benchmark ready, but I do have the Vue 6 Infinite benchmarks for three systems:

Athlon64 3500+ : VueMark = 127
Athlon64x2 4400+ : VueMark = 250
Q6600 2.4 Ghs: VueMark = 437

The Athlon64x2 is effectively the same as two Athlon64 3500+ chips in a single package, and Vue (both 5 and 6) almost perfectly double their rendering speed. 
A single AMD 64 bit core is somewhat more efficient than a single Intel core, that's why the Q6600 doesn't double the score of the 4400 x2.

Think I'll wirte a script to time the renders.

The pen is mightier than the sword. But if you literally want to have some impact, use a typewriter

My gallery   My freestuff


svdl ( ) posted Sun, 01 July 2007 at 9:33 PM

By the way, I checked the Gateway FX8020. And frankly, you can do a LOT better for the same money.

Here's the system I configured:

Intel Q6600 CPU (same as the Gateway)
2x2 GB DDR2-667 (twice as much as the Gateway)
MSI P35 Platinum mainboard (a more modern chipset than the 975X of the Gateway, with IEEE1394, 8.1 sound, SPDIF, Gbit LAN, the whole kaboodle)
ATI Radeon X1950 Pro /512 MB (same as the Gateway)
DVD Dual Layer +/- RW unit (same as the Gateway)
500 GB SATA II hard disk (same as the Gateway)
Antec Performance P180B case (silent, elegant and lots of room. You can mount 6 harddisks, 4 5.25" units, plus a floppy drive)
Zalman ZM-600 HP PSU: silent, more than enough power (600W).
Floppy drive
Card reader
Keyboard + mouse.
Windows XP Pro 64 bit

In the Netherlands, it would cost less than € 1500 including assembly, VAT and shipping. The VAT rate in the Netherlands is 19%, so this system should cost significantly less in the US.

What you get instead of the Gateway:

  • a more powerful machine (4 GB RAM expandable to 8 GB, instead of 2 GB expandable to 4 GB)
  • a much better system case, more silent and much more room for addons
  • a better PSU.
  • and for less money.

You can also opt to leave out a card reader, a modem, depending on what you need.

The pen is mightier than the sword. But if you literally want to have some impact, use a typewriter

My gallery   My freestuff


drifterlee ( ) posted Sun, 01 July 2007 at 10:14 PM

The Gateway is at bestbuy.com. They also have the HP 6200 + AMD dual core for around $900.


fiontar ( ) posted Mon, 02 July 2007 at 6:44 PM

Still hoping for some quick benchmarks comparing 1,2 and 4 cores. No need for a custome script, just a scene that takes a few minutes to render and a watch or timer. :)


svdl ( ) posted Mon, 02 July 2007 at 6:48 PM

Sorry, working on a scene right now, the quad core system is busy with Vue 6 Inf.

By the way, the .pz3 I'm importing is more than 1 GB - and P7 is still going strong!

The pen is mightier than the sword. But if you literally want to have some impact, use a typewriter

My gallery   My freestuff


drifterlee ( ) posted Mon, 02 July 2007 at 7:28 PM

Sounds great!


operaguy ( ) posted Tue, 03 July 2007 at 11:03 AM

this script just times the current render, but it works great.

#-------------------------------------------------------------------------------

TimeCacheRender.py

(c) 2005-04-12 Geoff Hicks (GeoffIX/gwhicks)

Initiate a Render to the render cache and display

the render duration.

#-------------------------------------------------------------------------------

import os
import time
import types
import poser

scene = poser.Scene()
toNewWindow = scene.RenderToNewWindow()
scene.SetRenderToNewWindow(0)
starttime = time.clock()
scene.Render()
stoptime = time.clock()
print "Render time = " + '%.2f' % (stoptime - starttime) + " seconds."
scene.SetRenderToNewWindow(toNewWindow)
#-END---------------------------------------------------------------------------


fiontar ( ) posted Tue, 03 July 2007 at 4:02 PM · edited Tue, 03 July 2007 at 4:04 PM

Operaguy,

     Thank you very, very much for the script! :)

Here are some numbers I pulled from my AMD 64 4400+ X2 Dual Core Processor:

Simple Low Quality Render
Render time = 151.07 seconds. 1 Core/2 Threads
Render time = 104.29 seconds. 2 Cores/2 Threads

Moderate High Quality Render
Render time = 1546.12 seconds. 1 Core/2 Threads
Render time = 842.21 seconds. 2 Cores/2 Threads
Render time = 789.52 seconds. 2 Cores/4 Threads*

You'll notice that the benefits of the additional core are much greater the more complex or higher quality the render. This is because the pre-render portion of the process (loading textures and meshes, calculating shadow maps, etc...) only uses a single core. Once that is out of the way and the actual render starts, you get full advantage of the cores/threads.

The way Poser Firefly utilizes additional cores is very efficient, but the boost as a percent will vary from image to image. With two cores, one core renders the top half of the image and one core renders the bottom half. If there is more computationally intensive content in one half vs. the other (human characters have more to render in the top half and raytracing transmapped hair seems to be a big resource hog), one core can finish before the other, which leaves that core idle for the rest of the render.

With 4 threads, it splits the image into quarters.

If anyone is going to test this out on their system, make sure to quit Poser after each render and restart the application. Poser will reuse some pre-render information from the same session, to make sure that doesn't skew the results you need to quit and restart poser. :)

*note the third line of test two is only two cores, but with 4 threads. The added performance was likely due to the way the scene was quartered up between the threads. Even though two threads finished before the other two, both cores were active for the entire render.


operaguy ( ) posted Tue, 03 July 2007 at 5:21 PM · edited Tue, 03 July 2007 at 5:26 PM

Thanks fiontarWould it be too much to ask you to render a baseline one core, one thread run?

Second, I had heard about this 'when I'm thru with my sector I just stand idle' approach. This is unfortunate! In my renders, it is almost always one sector that holds up the entire thing, namely the head of a/the character with reflect in the eyes, dynamic hair and often hair casting shadows. Sometimes the head/hair only consumes 10% of the frame, but requires 60% or more of the render time. Hell, sometimes 80% of the render time. Even with four sectors, you know that fate will put that head in only one sector just to torment me.

Also, your advice is smart about quitting each time you and coming back in for a fair comparison. Additionally, there is another consideration, the opposite. Do one render and throw away, then do the timed render. That way, resources that are common to the scene stay loaded (texture maps!) and the tests do not count the loading into the comparison.

This last approach would be from the mindset of the animator. If we are rendering 500 frames, the load-up on the first frame is 'an investment' that does not have to be made subsequently. 

Sad to hear that rendering the shadow map is single thread/core. For animation, it needs to be calculated for every frame. I'd have loved to have heard all four cores attack this. (note: there are strategies with animation for not rerendering shadows, multi pass, background in post, etc.)

Fiontar, i'm not asking you to change your testing method...we can still see the comparison your way. Thanks for doing the testing.

::::: Opera :::::


operaguy ( ) posted Tue, 03 July 2007 at 5:54 PM

I'd like to re-float my "render condo" idea again, as part of the mix of judging these powerful multi-thread systems.

This idea is strictly for animation, not stills.

Instead of investing in an expensive quad, what about investing in four nodes. You'd go for low-cost 'boxes' that do the best possible render job on 1 singleton CPU. I bet you could create a box for $500-$700. (Irony, the cost of purchasing the OS becomes more and more significant a percentage of the whole when thinking this way. Damn you, Bill! You too, Steve!)

You have to license Poser 4 times; you can probably get a quantity discount or site license for this. I paid $129 for my second and third licenses. Each node sould be fed an identical runtime folder.

Then, instead of figuring out how to "set up a poser render farm" you just send the pz2 to all four nodes and do "make movie", assigning a different frame range to each. You can have them deposit their frames in a common folder, or each in their own folder for combining later.

This approach works. I have three licenses for Poser 5 (only one kicked up to 7 so far) with one being a Mac license. I've done this and it works.

One of the advantages, of course, is if there is a flaw or failure in one, the other three continue on their merry way and the run is not stopped, as it would in a single machine with four threads, were that to fail. It also provides great flexibility....you 'get to see' the results of the animation in four separate parts of the sequence; you could keep three nodes going and resend a modified pz2 to the fourth if necessary; not all nodes would have to attack the same pz2, they could all be working on different scenes, rendering a folder of images with After Effects, etc.

::::: Opera :::::

Question: With multicore/threaded machines and Poser's "render to outside process" can you 'instance'? In other words, is it possible to have more than one actual render running on one computer?


svdl ( ) posted Tue, 03 July 2007 at 6:23 PM

A quad core system is less expensive than 4 single core systems: drifterlee mentions a quad system for $1500. And you'll only need a single OS license (included) plus a single Poser license.

But a quad core system also performs less than 4 single core systems combined - provided that the single core systems have enough RAM and disk to work with. 

AMD dual core systems aren't that expensive anymore ($900 for a decent base system), so maybe two AMD dual core systems might be the way to go...

As far as running more than one render on a single computer - not right away, but you can do it using VMware. The advantage of VMware over two physical machines is that your Windows license is valid for both the host machine and all virtual machines on this particular host. Plus, VMWare is free. You'll probably need a Poser license for each VMware instahce however.

VMware becomes a great option when you have a machine with more than 4 cores, for example dual Xeon quad core system (8 cores). Make sure you have lots of RAM on such a machine. VMWare recognizes all installed RAM and will divide it over the hosted virtual machines.

The pen is mightier than the sword. But if you literally want to have some impact, use a typewriter

My gallery   My freestuff


fiontar ( ) posted Tue, 03 July 2007 at 6:37 PM

Operaguy,

     Here is a quick run of the simple render numbers run as you requested, after an initial "foundation" render to load things into memory. (The dialogs still show textures being loaded and shadowmaps rendered on subsequent runs, but it is faster than the initial run, so maybe the second time is quicker because these things are now in ram rather than being loaded from the HD?)

Simple Low Quality Render (buffered)
Render time = 123.58 seconds. 1 Core/1 Thread
Render time = 125.03 seconds. 1 Core/2 Threads
Render time = 79.23 seconds. 2 Cores/2 Threads
Render time = 86.87 seconds. 2 Cores/4 Threads

Notice that the render times went up in this case if the number of threads was set higher than the number of cores. Even though there was a slight benefit in the high render quality test in the previous thread from going to 4 threads, it may be a hit or miss thing depending on how the workload is split based on the composition of the scene. I'll try to test that out further later tonight. (Previous informal tests had led me to chose 2 cores/ 2 threads for my renders).


fiontar ( ) posted Tue, 03 July 2007 at 6:49 PM

svdl,

     For me, the second core gave a speed boost of ~85% over a single core with high render settings. That's why I'm very curious for your tests when you have time. First, to see how the Intel scales 1 core vs. 2 and of course then to see how it scales 2 cores vs. 4. :)

     4 cores have other benefits beyond raw rendering time, I'm not at all trying to discount quad core as an option (I'm looking into it for my next upgrade personally). It will just be very helpful for all of us looking at quad core systems to see some actual numbers.

     I think the pivot point between cost/benefit of buying two dual core systems vs. a quad core system isn't necassarily settled, plus a quad core system get's bonus points for being more managable from a practical use standpoint for many people.


fiontar ( ) posted Tue, 03 July 2007 at 8:20 PM

Operaguy,

     As to the Seperate Process Renderer, it isn't quite ready for prime time.

Moderate High Quality Render
Render time = 842.21 seconds. 2 Cores/2 Threads
Render time = 919.93 seconds. Seperate Process Render 2/2

Not only did it take longer, but there was noticable color banding in the background of the image that was not there when rendered at the same settings rendering directly in Poser. Also, it still ties up Poser while using the seperate process render, so there is no advantage there. Seems like a feature they didn't really finish development on and just stuck it in there. I'm guessing we'll see an improved version or offshoot in Poser 8?


svdl ( ) posted Thu, 05 July 2007 at 4:43 PM

System:
CPU: Intel Core 2 Duo Q6600 at 2.4 Ghz
RAM: 8 GB DDR2-667
Graphics: nVidia 7800 GTX
Disks: WD Raptor 10,000 RPM 74 Gb (OS)
       WD Raptor 10,000 RPM 74 Gb (Apps)
       Seagate Barracuda 7200.10 320 Gb (Data)

Windows XP 64 bit SP2
Poser 7 SR 2

Test scene: Jim Burton's P5 test render scene (with raytracing and dynamic hair)
Set to reuse shadow maps and keep textures loaded.

Renders to 400x400 window:

1 core, 4 threads, Poser process: 35.5 seconds
2 cores, 4 threads, Poser process: 27.5 seconds
3 cores, 4 threads, Poser process: 24.5 seconds
4 cores, 4 threads, Poser process: 22.5 seconds
4 cores, 4 threads, FFRender: 35 seconds

Renders to 800x800 window:

1 core, 4 threads, Poser process: 77.5 seconds
2 cores, 4 threads, Poser process: 50 seconds
3 cores, 4 threads, Poser process: 41.5 seconds
4 cores, 4 threads, Poser process: 40 seconds
4 cores, 4 threads, FFRender: 44 seconds

Seems like the inter process communication between Poser and FFRender takes quite a bit of time.

The pen is mightier than the sword. But if you literally want to have some impact, use a typewriter

My gallery   My freestuff


operaguy ( ) posted Thu, 05 July 2007 at 5:05 PM

svdl,

first, nice system....

  1. which of the three drives do you have the OS swap file located?
  2. have you come to believe that running separate HD for the three elements is smarter than RAID-0?

on the test runs:

  1. are you saying that FFRender is a 'regular' firefly render and that "Poser process" is 'render to a separate process?  (and thus your comment that 'render in separate process' is more efficient in communication overhead than 'separate process?')

  2. since a 800x800 render contains 4x the pixels of 400x400, wonder why it only takes roughly 2x the time to render to the larger window?


svdl ( ) posted Thu, 05 July 2007 at 6:14 PM

At 1): The swap file is evenly distributed over the two WD Raptors, 6 GB of swap file each.

At 2): RAID-0 will almost double the transfer speed, but it will NOT improve access time. Since Windows (and all other modern operating systems) consist of many processes, many of which can access drives, it's smarter to have Process A access disk 1, while at the same time Process B accesses disk 2. If disk 1 and 2 were combined in a RAID 0 array, both processes would compete for the same disk.
Since most of the tasks my computers do involve many small disk transactions, it's smarter to make those transactions independent by having them access independent disks. Only when you do very large file transfers (think harddisk recording), the increased transfer rate of RAID 0 offsets the cost of shared access. And even in the case of harddisk recording I'd recommend placing the OS on its own separate drive, so that its drive access does not interfere with the recording.

At 3): FFRender is the separate process that Poser 7 can start. Poser process means an in-process render. So I meant it exactly the other way around.

At 4) The funny part was that the actual rendering didn't take that much time. Loading objects was the most time consuming phase in the renders, and loading objects has to do with drive access. Anyway, my task manager told me that while Poser was loading objects it used only 1 core.

Now if I would render again at 1600x1600 resolution, I'd expect the actual render time to quadruple, and I'd expect the object loading time would be less influential.

The pen is mightier than the sword. But if you literally want to have some impact, use a typewriter

My gallery   My freestuff


operaguy ( ) posted Thu, 05 July 2007 at 7:16 PM

So on a modest-sized frame, "load objects" takes up proportionally larger amount of total time. I wonder why, in animation, we have to "load objects" for each frame. Shouldn't there be a 'keep objects' like there is for "keep textures." At any rate it is too bad it only uses 1 core to 'load objects.'

Since there is an overhead hit of some kind (12 seconds on the smaller, 4 seconds on the larger) to 'render in a separate process, it would be contra-indicated to use. What advantages would an animator gain with 'render to outside process' if willing to pay for that hit?

::::: Opera :::::


operaguy ( ) posted Thu, 05 July 2007 at 7:31 PM

Attached Link: Best Buy system

I wonder if this is what Drifterlee is looking at.

Does not say the wattage of the power supply.

Still, pretty good at $1050 with 2Gig RAM and one 500G HD, HDTV and good sound on board. GeForce 8500 GT is the modest end of the 8-series.


svdl ( ) posted Thu, 05 July 2007 at 7:57 PM

"Render in separate process" doesn't have much of an advantage when rendering animations. When rendering complex stills, however, the fact that the separate process has its own 2 GB (or 4 GB on XP 64 bit) address space means that more complicated scenes can be rendered. The render process doesn't carry the memory overhead of a Python interpreter, a graphical user interface, and who knows how many tools, so more address space is available for the actual render.

It may be that "render to separate process" is the first step towards a Poser render farm, which would definitely be an asset to animators. We'll see when Poser 8 comes out.

The pen is mightier than the sword. But if you literally want to have some impact, use a typewriter

My gallery   My freestuff


fiontar ( ) posted Fri, 06 July 2007 at 9:48 AM

svdl,

     Thanks for what you ran. I understand the reason for wanting to run some standardized benchmark, but I had already covered the point that the pre-rendering overhead, which uses one core, would skew results on short renders. :) If one where to look at your numbers alone, the conclusion would be that four cores is not worth while. However, with most renders being of higher quality, the pre-render overhead becomes insignificant and the benefit of additional cores really kicks in. We don't get to see that with these numbers.

     This still leaves my questions unanswered.

     Assuming the render is complex and large enough to overcome the pre-render overhead, I've seen that with my AMD X2 dual core, the second core has provided a performance boost of 85%. If the scene where balanced such that both cores ran for the entire render (with the image split up into halves or quarters based on the number of threads, some threads end before others) the benefit may be even higher.

     True AMD quad core is still a few months away I believe, and I can not extrapolate X2 performance to AMDs current X2+X2, two socket "quad core" systems. Theoretically, Intel's Quad Core should offer a nearly linear benefit for each additional core, once overhead and thread distribution are factored in, for large, more complex renders. Unfortunately, the benchmark tests are not sufficient to show how the Intel quad Core scales in pure rendering performance.

     [Also, (on a side note) maybe someone else might want to test this out for themselves, but I found the seperate process renderer to be undesirable for more than the fact that it produces a performance hit, rather than a performance benefit. When layering renders of the same test scene at the same settings, one from the regular FF render and one from the seperate process FF renderer, I could clearly see that the results were not consistant. There was severe color banding in the background from the seperate process render not present in the standard FF render.]


fiontar ( ) posted Fri, 06 July 2007 at 1:41 PM · edited Fri, 06 July 2007 at 1:54 PM

OK, I ran the Jim Burton P5 benchmark,
http://www.renderosity.com/mod/freestuff/index.php?user_id=2541
under the same conditions, plus ran some math on the results:

AMD64 X2 4400+ (2.4 Ghz, 2GB DDR Ram (480 Mhz))

400x400 (Re-use Shadow Map, keep textures in memory).
1 Core(s), 1 Thread(s): Render time = 42.35 seconds.
1 Core(s), 2 Thread(s): Render time = 45.24 seconds.
2 Core(s), 2 Thread(s): Render time = 31.33 seconds.
2 Core(s), 4 Thread(s): Render time = 32.52 seconds.

800x800 (Re-use Shadow Map, keep textures in memory).
1 Core(s), 1 Thread(s): Render time = 105.56 seconds.
1 Core(s), 2 Thread(s): Render time = 108.00 seconds.
2 Core(s), 2 Thread(s): Render time = 63.14 seconds.
2 Core(s), 4 Thread(s): Render time = 63.43 seconds.

1600x1600 (Re-use Shadow Map, keep textures in memory).
1 Core(s), 1 Thread(s): Render time = 360.52 seconds.
1 Core(s), 2 Thread(s): Render time = 358.26 seconds.
2 Core(s), 2 Thread(s): Render time = 192.74 seconds.
2 Core(s), 4 Thread(s): Render time = 182.18 seconds.

Best results, performance boost from 1 core to 2 cores:

400 x 400: 35.2%
800 x 800: 67.2%
1600x1600: 96.7%

***svdl's Intel Core 2 Duo Q6600 at 2.4 Ghz


Best results, performance boost from 1 core to 2 cores:
400 x 400: 29.1%
800 x 800: 55.0%
1600x1600: ???

Best results, performance boost from 2 cores to 4 cores:

400 x 400: 22.2%
800 x 800: 25.0%

Best results, performance boost from 1 core to 4 cores:

400 x 400: 57.8%
800 x 800: 93.75%
1600x1600: ???

*AMD Athlon64 4400+ X2 (2.4Ghz) 2 Cores vs. Intel Core 2 Duo Q6600 (2.4Ghz) 4 Cores

Intel quad-core advantage vs. AMD dual-core Both@2.4Ghz:

400 x 400: 39.3%
800 x 800: 57.9%
1600x1600: ???

Intel dual-core advantage vs. AMD dual-core Both@2.4Ghz:*

400 x 400: 13.9%
800 x 800: 26.3%
1600x1600: ???

*****Extrapolated based on 2 cores running on Intel Core Duo 4
core processor.

Note: My AMD processor is 1 1/2 years old, newer models at higher Mhz would fare better.

**Summary:
**Based just on raw rendering power, not taking into effect the additional multi-tasking benefits of a quad core system, the performance advantage rendering in Poser 7 of an Intel Core Duo Quad Core vs. Dual core is approx 25%. The cost differential between similarly specce'd Dual Core and Quad Core systems would need to be weighed vs. a 25% performance difference for someone deciding primarily on Poser Rendering performance.

Also note that standardized benchmarks for other 3D rendering programs show a higher advantage of quad core vs. dual core. Poser 7's multi-core rendering process could definitely be improved. With the current approach of halving or quartering the image based on the number of processes, it's just to easy to run into situations where some threads finish well before others, leaving some cores idle for the final portion of the render process.


svdl ( ) posted Fri, 06 July 2007 at 2:00 PM

AMD Athlon64 4400+ X2 (2.4 Ghz)  
Are you sure about 2.4 Ghz? I also have an Athlon64 4400X2 Socket 939, it runs at 2.25 Ghz (not overclocked in any way)

I'll do the render tests on the Quad on 1600x1600.

The pen is mightier than the sword. But if you literally want to have some impact, use a typewriter

My gallery   My freestuff


AnAardvark ( ) posted Fri, 06 July 2007 at 2:27 PM

Attached Link: Product Information Sheet

> Quote - There is a Intel® Core™2 Quad Processor Q6600 Gateway on sale right now with a good graphics card. How much faster will Poser render on a quad core as opposed to my 3.4 GHz P4 singe core with 2 gigs of RAM, or will it make any difference? Thanks

 

Oh boy will it make a difference! I don't think, however, that the Quad Processor will make a performance difference over a single Intel Core 2  duo (dual processor) for rendering, but you will be able to devote two of the Quad's processors to rendering, and use the other two for web-browsing etc. (In retrospect, I almost wish I'd waited before getting my new machine, but my old one was crashing constantly.) The Quad processor is really just a pair of the Intel Core2's on a single chip.

The Intel Core2 duo is the sweetest processor to come along in a long time. For the following reasons:

  1. 4 MB L2 cache shared between the two processors. Cache sharing is more efficient than the older architecture where each processor had its own cache. 4 MB is a heck of a lot of cache. (For a contrast, the Celeron has about 1/2 MB.) For those of you who might not remember, cache memory is the ultra fast memory which allows the processor to keep data which is likely to be used soon "near to hand".
    2. The chip architecture is significantly improved. They've essentially reduced the time taken to execute each machine instruction by allowing for fetching four instructions at once per processor, as well as improved the pipelining, and allowing pre-fetch of data to allow out-of-order execution of instructions (essentially overlapping executing of instructions which don't use the same data.) The net result is that a single processor does more at a 2.6 GHz clock speed than a traditional P4 can do at a much higher clock speed. (I've seen benchmarks as high as 4 GHz) It is also supposedly pretty easy to overclock if you are into that sort of thing.
  2. The chip is more suited to the more advanced motherboards. It can handle front-side bus speeds of up to 1033 MHz. (This is the transfer speed between internal components other than I/O devices.) So you can use really fast memory as well.

BTW, mine has been remarkably stable -- I've had it about four months and it has only had a system crash twice (both within the past week, and on the same program.)


svdl ( ) posted Fri, 06 July 2007 at 2:45 PM

Here's the stats for the P5 test scene at 1600x1600:

1 core 4 threads:277.68 seconds
2 cores 4 threads: 173.72 seconds
4 cores 4 threads: 101.62 seconds.

So a quad core WILL render faster than a dual core at the same frequency.

The pen is mightier than the sword. But if you literally want to have some impact, use a typewriter

My gallery   My freestuff


svdl ( ) posted Fri, 06 July 2007 at 3:17 PM

Some more test results from my Athlon4400x2 (2x1 Mb Cache) Socket 939, 2.25 Ghz on WinXP 64 bit!

800x800 render:
1 core 4 threads: 106.99 seconds
2 cores 4 threads: 65.31 seconds

1600x1600 render:
1 core 4 threads: 345.77 seconds
2 core 4 threads: 191.52 seconds

The pen is mightier than the sword. But if you literally want to have some impact, use a typewriter

My gallery   My freestuff


fiontar ( ) posted Sat, 07 July 2007 at 5:30 AM

file_382103.jpg

> Quote - AMD Athlon64 4400+ X2 (2.4 Ghz)   > Are you sure about 2.4 Ghz? I also have an Athlon64 4400X2 Socket 939, it runs at 2.25 Ghz (not overclocked in any way) > > I'll do the render tests on the Quad on 1600x1600.

Thanks for running the 1600x1600 numbers, definitely gives us all a better baseline and makes the quad core a lot more attractive. The performance boost between 400 and 800 was so small, glad it scales much better at higher rez. :)

Very interesting as well how the P7 Firefly renderer scales at higher resolutions. With 1600x1600 being 4x the area in pixels than 800x800, I would have expected the render time to scale closer to 4X as well.

Looks like a quad core will be a good investment for Poser users. I'll be looking at it more seriously myself. I usually only upgrade when I can get at least a 40% performance boost over my current system and I just couldn't get that by upgrading to any of the current dual core processors.


Gazukull ( ) posted Sat, 07 July 2007 at 9:58 AM

No one has mentioned that the price on a Q6600 drops to 266 dollars on July 22nd.  Keep an eye out for that my fellow home builders :)


operaguy ( ) posted Sat, 07 July 2007 at 2:18 PM

".....266....."

Whoa. That is big news. How do you know that?


AnAardvark ( ) posted Sat, 07 July 2007 at 2:30 PM

Quote - Here's the stats for the P5 test scene at 1600x1600:

1 core 4 threads:277.68 seconds
2 cores 4 threads: 173.72 seconds
4 cores 4 threads: 101.62 seconds.

So a quad core WILL render faster than a dual core at the same frequency.

 

What is the difference between your dual and quad core setups? Supposedly the poser rendering engine can't use more than two physical cores. (I suppose that the other load on the system would be shifted to the unused processors, but I wouldn't imagine that it would make that big of a distance.)


svdl ( ) posted Sat, 07 July 2007 at 3:31 PM · edited Sat, 07 July 2007 at 3:32 PM

Poser 7 can use up to 4 threads, each thread can run on a separate core.
Other processes: the bare minimum, just the system itself. No virus scanner or whatever.

I monitored CPU usage during the render - all four cores were used by Poser.exe

My dual core setups are AMD Athlon64x2 machines, one 4400 and one 3800, both with 4 GB of DDR400 RAM. The Quad has 8 GB of DDR2-667

The pen is mightier than the sword. But if you literally want to have some impact, use a typewriter

My gallery   My freestuff


swordman10 ( ) posted Sat, 07 July 2007 at 5:19 PM

May I just add to this.

I have a two machine setup one system Intel quad core QX6700 running @2.8, 4 gigs ram.

second system AMD FX60, 64, running @2.7, with 4 gigs ram.

Without being technical it HAMMERS the dual core., In comparison renders it is on average 70+% faster. period. On Poser 7, or any other package that sees more than one processor.

On one test the quad had finished rendering before the dual had even started.!!!

Be under no illusions quad core processors give unrivelled and definate performance gains.


pjz99 ( ) posted Sat, 07 July 2007 at 5:32 PM

I have a Q6700 (Extreme Edition), and it is very very nice, but the quad core Core 2 base line chips Gazukull is talking are going to be very nearly as solid.

My Freebies


AnAardvark ( ) posted Sun, 08 July 2007 at 1:26 PM

Cool. I'll look into possibly replacing my dual with a core eventually, if my motherboard will handle it.


Privacy Notice

This site uses cookies to deliver the best experience. Our own cookies make user accounts and other features possible. Third-party cookies are used to display relevant ads and to analyze how Renderosity is used. By using our site, you acknowledge that you have read and understood our Terms of Service, including our Cookie Policy and our Privacy Policy.