i just ordered a quad core...

psor · Post by **psor** » Sat Jan 26, 2008 2:20 pm

@dougal2

Even if I'm a bit late, but I was wondering what settings you'd used.

e.g. ... PT; biPT; MLT; biMLT

take care
psor

dougal2 · Post by **dougal2** » Mon Jan 28, 2008 5:17 am

I used the diffuse_transmitter_test scene included with indigo as a benchmark, and all default settings (apart from setting num_threads to 4).

Grimm · Post by **Grimm** » Mon Jan 28, 2008 10:32 am

dougal2 wrote:WOO! machine did indeed turn up today.
Got XP32 and Vista64 on it already.

Did a quick speed test with 1.0.4 diffuse_transmitter_test test scene:

~ 210,000 samples per second.

I'm just going to have a look for a solution to why 0.9t4.5 64bit doesn't run then I'll post figures for that when I can get it running.

edit: 64bit version doesn't seem to be any faster.
I'd be interested you see other people's speed for this test scene, and what machine you ran it on.

Hi Dougal2,

I ran the diffuse_transmitter_test on my microwulf system but I'm not sure how to read the samples per second. Here is the output:

------------------------------------------------
Display took 0.71327 s
Time elapsed: 1 m, 54 s
Done 14460000.00000 samples (53.55556 samples per pixel)
125899.28064 samples / second (7.94286 micro-seconds / sample)
saving tone-mapped image to 'renders/im1201472071.png'
done.
------------------------------------------------
Display took 0.83701 s
Time elapsed: 2 m, 5 s
Done 14460000.00000 samples (53.55556 samples per pixel)
115044.11885 samples / second (8.69232 micro-seconds / sample)
------------------------------------------------
Display took 0.71319 s
Time elapsed: 2 m, 16 s
Done 14460000.00000 samples (53.55556 samples per pixel)
106008.27084 samples / second (9.43323 micro-seconds / sample)
------receiving frame from 10.0.0.3:------
num_samples: 2720000.000000
width: 1200
height: 900
------frame received.------

My system consists of three nodes, each with a AMD X2 4000+ processor for a total of six cores. Do I add up the samples/second of all of them which would add up to 346951 samples/second or do you look at them separately? My samples seem to increase over time though, maybe it's the OS settling down?

edit: oops I forgot to add that I'm running the 64 bit linux build version 1.0.1 on Fedora 8.

dougal2 · Post by **dougal2** » Mon Jan 28, 2008 11:40 am

you should look at the output on the master node, that includes all samples from all slaves. I would let it run for more than a few minutes though to make sure that all slaves have contributed. Or at least run it for longer than the frame_upload_period, maybe 2 or 3 times longer so that all slaves and the master have had a chance to run at full speed.

Grimm · Post by **Grimm** » Mon Jan 28, 2008 12:31 pm

Ok here is the output from the master node:

------------------------------------------------
Display took 0.71033 s
Time elapsed: 33 m, 55 s
Done 381240000.00000 samples (1412.00000 samples per pixel)
187264.19077 samples / second (5.34005 micro-seconds / sample)
saving tone-mapped image to 'renders/im1201477283.png'
done.
------------------------------------------------
Display took 0.92118 s
Time elapsed: 34 m, 6 s
Done 381240000.00000 samples (1412.00000 samples per pixel)
186264.96229 samples / second (5.36870 micro-seconds / sample)
------------------------------------------------
Display took 0.71049 s
Time elapsed: 34 m, 17 s
Done 381240000.00000 samples (1412.00000 samples per pixel)
185295.31391 samples / second (5.39679 micro-seconds / sample)
------receiving frame from 10.0.0.2:------
num_samples: 2970000.000000
width: 1200
height: 900
------frame received.------

I let it run for about a half an hour, but I'm still not sure how to read it. here is what the nodes reported for their samples:

node 1:

------Uploading frame------
Frame size: 12.360 MB
Num Samples: 2980000.00000
------frame uploaded.------
Approx. samples per second: 99333.33333
------Uploading frame------
Frame size: 12.360 MB
Num Samples: 2970000.00000
------frame uploaded.------
Approx. samples per second: 99000.00000
------Uploading frame------
Frame size: 12.360 MB
Num Samples: 2970000.00000

node2:

------Uploading frame------
Frame size: 12.360 MB
Num Samples: 2710000.00000
------frame uploaded.------
Approx. samples per second: 90333.33333
------Uploading frame------
Frame size: 12.360 MB
Num Samples: 2710000.00000
------frame uploaded.------
Approx. samples per second: 90333.33333
------Uploading frame------
Frame size: 12.360 MB
Num Samples: 2720000.00000

dougal2 · Post by **dougal2** » Mon Jan 28, 2008 12:33 pm

is the master a "working" master, or just a non-rendering master ?

you should be able to get another 90k samples/sec if the master is working also.

start the master with the -n wm switch, and try again.

Grimm · Post by **Grimm** » Mon Jan 28, 2008 12:48 pm

I tried the -n wm switch but indigo complained:

./indigo_console -n wm testscenes/diffuse_transmitter_test.igs
Indigo Renderer v1.0.1, Linux 64-bit Release build.
usage:
indigo.exe [-n (s|m)] [-h hostname:port] scenepath.xml

Should I start another slave process on the master node?

dougal2 · Post by **dougal2** » Mon Jan 28, 2008 12:53 pm

ah, I think -n wm is only for a more recent version.

A simple alternative is indeed to start a slave on the master, as you said.

Grimm · Post by **Grimm** » Mon Jan 28, 2008 1:30 pm

Darn, you mean to tell me that I have been running at only 2/3 capacity?

Here is the output from the master process:

Display took 1.12389 s
Time elapsed: 34 m, 9 s
Done 558340000.00000 samples (2067.92593 samples per pixel)
272447.79421 samples / second (3.67043 micro-seconds / sample)
------receiving frame from 10.0.0.2:------
num_samples: 2980000.000000
width: 1200
height: 900
------frame received.------
------------------------------------------------
Display took 0.92981 s
Time elapsed: 34 m, 20 s
Done 561320000.00000 samples (2078.96296 samples per pixel)
272448.81699 samples / second (3.67041 micro-seconds / sample)
------receiving frame from 10.0.0.3:------
num_samples: 2710000.000000
width: 1200
height: 900
------frame received.------

Thanks Dougal!!!

dougal2 · Post by **dougal2** » Mon Jan 28, 2008 9:46 pm

that looks more like it

CoolColJ · Post by **CoolColJ** » Sun Feb 24, 2008 5:56 pm

still on a P4 here, might wait till 2009 for the new Intel 8 core CPU and Mothboards it will need before I upgrade

suvakas · Post by **suvakas** » Sun Feb 24, 2008 7:46 pm

Hey CoolColJ, life is too short to wait like that

2008 just began. Go get a quad now

psor · Post by **psor** » Sun Feb 24, 2008 8:25 pm

Yup, go for it! The cheapest node I could put together (in my mind) was like:

Code: Select all

Intel DG33BU (mATX LGA775, max 8GB)         ~100,- Euro
Intel Core 2 Quad Q6600 FSB 1066                 ~210,- Euro
Kingston Value RAM PC 533 4096 (KIT 2x2GB) ~  90,- Euro
...

Imagine ten of them in a Beowulf cluster! *Hello Grimm!* ;o)
So ya, it's time to take the risk! The head node is not included!

take care
psor

Grimm · Post by **Grimm** » Mon Feb 25, 2008 9:09 pm

CoolColJ, I agree with suvakas, go for it. It doesn't take many nodes to get much better performance then from a single box (unless it has that 80 core IBM cpu in it

).

psor, 10 of those nodes would scream.

400 euros isn't too bad, what is that, about $600 US? That is about what 2 of my dual core nodes cost and it would perform better too. Although my cost ($300 US) also includes the case, power supply, and hard drive.

psor · Post by **psor** » Mon Feb 25, 2008 10:13 pm

@Grimm

Hell ya, ... I probably would go without hdd and without a case.
I have to take a look at those really small PSU's, then I would
build the case on my own with all ten inside.

btw. this is just an idea in my head, but it gets more clear
every day. So lets see what happens at the end of the year.

I just have to remember you at the picture ... *whistle*

take care
psor

i just ordered a quad core...

Who is online