Building a renderfarm - Follow up - Benchmarks

[Lucetius]

In the previous thread below, the question arose as to what the performance/experience has been with different computers used in a renderfarm. We have a somewhat diverse mix of machines here and I ran a quick test for a comparison.

Take this with a grain of salt as this was set up for how we work here. Hopefull this helps.


Here is the scene I rendered.



Details of the scene are as follows:

Rendered at 2000x1000
Approx 2.5 million polys
Cars are all Vray mesh
Grass is displaced.
1 standard Max sun with Vray shadows (1.0)
Vray enviroment (1.0)
I.S. = QMC 2,4
GI = Irradiance (1.0) QMC (.5)
Imap = min -3, max -2, clr .2, nrm .2, dist .1, hsph 40, samp 40
QMC samp = .85, .001, 15

I submitted 13 jobs to Backburner and set them up to render 1 at a time. The results and computer specs are as follows:

#1 DIY. - Workstation
2 x Opteron 248, Tyan S2885
4gig, dual channel, Crucial, DDR400, ECC
Render time: 1:07:23

#2 Boxx tech - Workstation
2 x AMD MP 2800, Tyan S2468
3.5gig PC2100 ECC
Render time: 1:37:13

#3 DIY - Workstation
2 x AMD MP 2800, Tyan S2469
3.5gig Crucial PC2100, ECC
Render time: 1:37:57

#4 DIY - Workstation
2 x AMD MP 1900, Tyan S2466
2.5gig Crucial PC2100, ECC
render time: 2:05:11

#5 DIY - 2u render slave
2 x AMD MP 2800, Tyan S2469
1gig Crucial PC2100, ECC
render time: 1:36:24

#6 Boxx tech - 1u render slave
2 x Intel Xeon 2.4ghz, Super Micro ??? NO HT
1gig ram, ????, ECC
render time: 2:05:23

#7 Boxx tech - 1u render slave
2 x AMD MP 2000, Tyan S2468
1gig ram, ????, ECC
Render time: 2:02:04

#8 DIY 2u render slave
2 x AMD MP 2800, Tyan S2469
2gig ram, Crucial PC2100
Render time: 1:37:10

#9 Shuttle kit - DIY
1 x Intel P4 3.4ghz "Extreeme edition"
1gig ram, Dual Channel, Crucial, DDR 400
Render time: 2:19:09 HT Disabled
could not get this to render with HT enabled

#10 Shuttle kit - DIY
1 x Intel P4 2ghz
512 meg ram, Crucial,
Render time: 5:20:04

#11 Shuttle kit - DIY
1 x Intel P4 3.0ghz
1 gig ram, Dual Channel, Crucial, DDR400
Render time: 2:04:33 HT Enabled
Render time: 2:35:30 HT Disabled

All computers are running Win XP SP1

[Dynedain]

Great breakdown, thanks - gives some good food for thought

[photonC]

Awesome opteron rocks, I considering to make my next workstation and it seem that AMD is what i'll get, thank you for the tests and the time put into it.

[nagboy]

What about the price / performance ratio, which one is best?

[Robert]

One thing is really strange, the Xeon. It's to slow, maybe there is something wrong with it.

We just expanded our renderfarm and we choose Xeons only, simple because we did test before and the Xeons where faster then the AMDs. (About 25%)

Well, guess it's HT that makes the differenz...

regards,

robert

[tammo]

the opteron has 4gigs of ram, thats why its faster in this test.
xeons with HT should be the best solution for vray rendering.

[ihavenofish]

one or 2 things poke out to me.

for the most part the results are consistent. faster chips, faster render.

the xeon is slow ish, but it is only a 2.4, so it may be that slow. the 1gb ram is likely not an issue with speed, cause the other machines have 1gb as well, and produce predictable result (the dual athlons particularly).

the one glaring oddity is the shuttle boxes. the p4 3.4ghz being slower than the 3.0 with the same ram. coul dyou post what operating system and service packs these are running? (also for the other machines if you got the time)

(edit: eep. missed the HT note. i guess its the 3.4 is a tiny bit faster)

later

[Giovanni]

lucetius,
thank you for the testing. it's great.

OT question:

why are your cars "Vray mesh"?
You mean, proxy meshes?

Is it faster to render?

Also, I assume you had uv mapped cars models, which you converted to vray mesh. Is the mapping automatically preserved?

thank you

regards,

gio

[Lucetius]

For us, the AMD's have provided a better value/performance ratio. The MP2800's have been great and they were cheap in comparison.

Ihavenofish: All the machines are running WinXP SP1. As you caught, if the 3.4 EE would have rendered with HT, probably would have seen a jump in speed. We run WinXP on all computers mainly for the ability to be able to Remote Desktop. It's nice to not have to leave my desk and go to the server room to switch over the slaves from Backburner to DR or vice-versa. (I'm lazy)

Giovanni: We exported most of our cars as VRmesh files mainly because of the memory usage. It's a hassle to work in a file that's bloated with 2.5million polys worth of cars. It was a bit of a pain to get them all set up but we organized it so that that each car has a multi sub material and that all the material slots are in the same order. i.e. windows are slot 1, chrome tint slot 2, etc. we just insert the car, apply the appropriate material and then change the paint color to suit our needs.

We're currently working on a project that has 32 tile roofs, all geometry. Our scene file scene file was over 200 megs. (just geometry) we Vray meshed the tile roofs and the file size dropped downn to 20megs.

[theo]

xeons with HT should be the best solution for vray rendering.

my experience has also proven this to be not true.
dual opteron 248 are faster than dual xeons 2.8 here. (average 25%)
and I would recommand to turn HT off for all rendering Xeons systems ... from my experience rendertimes are slower with HT enabled.

For us, the AMD's have provided a better value/performance ratio.The MP2800's have been great and they were cheap in comparison.

agree, we have 3 self made dual MP2800 these systems are cheap and fast renderboxes.

on a side note, while expensive, the dual opteron system is a dream to work with.. I have been astonished by the speed and responsiveness of this system.
one exemple, application start time:
max7 starts in about 5 seconds on the dual opteron
while on all others systems it takes between 30sec to 1 minute.

I highly recommend iwill motherboard ( http://www.iwill.net ) to build a workstation.
and theses http://www.iwill.net/product_2.asp?p_id=36 looks like cute renderboxes

[RErender]

Quote:
xeons with HT should be the best solution for vray rendering.

my experience has also proven this to be not true.
dual opteron 248 are faster than dual xeons 2.8 here. (average 25%)
and I would recommand to turn HT off for all rendering Xeons systems ... from my experience rendertimes are slower with HT enabled.

There is something wrong with that experience, there is almost always a marked difference between HT and not in my experirnce, usually a 20%-25% speed difference. It does use more memory, maybe in your test it was doing too much swapping?

[TRGraphics]

I agree with rerender. There is about a 20% - 25% difference with HT enabled versus not enabled with vray on my Zeon 2.8 system. It seems that the setup of each individual makes a big difference in rendertimes on different machines.

I'm very happy with my zeon's and won't be changing any time soon.

Tony

[theo]

I agree with rerender. There is about a 20% - 25% difference with HT enabled versus not enabled with vray on my Zeon 2.8 system. It seems that the setup of each individual makes a big difference in rendertimes on different machines.

is that with single or dual xeons ?

[RErender]

Should be about the same for both, but my experience is with a dual 3.0

[TRGraphics]

Mine is a dual with 2 gig of ram on an Asus pc-dl motherboard.

Tony