Iray Server Render tests

[DigiDotz]

Just a note to mention my first tests of iray server on amazon web services (cloud computing) are done
AWS Iray Server Rendering - Digi-Dotz 3D
http://www.digi-dotz.com/index.php/digidotz-blog/172-aws-iray-server-rendering
Happy to report Daz Studio's network rendering is enabled in Daz Studio 4.9.3 and working well
I still have the higher powered GPU tests to do ...although i wont do the highest as it costs a bomb ;-)

So far I think its a great option if you need a higher powered machine eg. my pc cant even render certain scenes

 

[phdubrov]

Good, and you can try Iray Server free.
But what I'd like to see is comparing Iray renders with MentalRay renders - DS can save mi scenes
And yearly price is the same.

 

[DigiDotz]

Thanks I'll check it out

 

[Pendraia]

Thanks for the info Digi...

 

[DigiDotz]

no more noisy fans an lock-up during rendering ;-)
Well, I'll probably use it myself when doing promo's as last time it took ages rendering on my old pc

 

[phdubrov]

Price is the question.
For example, 4k HQ render in Cycles can take up to 16h on Google and about 10$. 20$ or more for Poser/SF for the need of paid OS.
But I tend to think that Daz promos underrendered and use at least 4 /100% as final qualty setting in Iray.

 

[DigiDotz]

Price is the question, I will try to make sense of it later after a few more tests, I've been using spot price instances so way cheaper than "on demand"
Just tested the larger GPU instances available

I am nearing my budget limit for this month though

 

[phdubrov]

Google has preemptible (basicly can be killed without notice) instances, can be good for CPU rendernode. ~0.20/h for 24 cores/21Gb (custom type machine).
If you didn't work with google cloud before they give you 300$ testdrive credits for 2 month.

 

[DigiDotz]

I have just signed up with ....the google. The Console is nicer. Its limited to 8 vCpu during the trial, which is fine to install things and even test 3delight rendering maybe, although still not sure what exactly a vCPU equals -half a cpu core?

On the AWS pricing front -who is good at mathematics to work out the best deal?
The results.....

G2.2xlarge = $0.13 per hour (Spot Instance) Render Time - 21 minutes
G2.8xlarge = $0.49 per Hour (Spot Instance) Render Time - 7 minutes
c4.8xlarge = $1.58 per Hour (Spot Instance) Render Time - 10 minutes
P2.xlarge = $0.29 per Hour (Spot Instance) Render Time - 8 minutes
p2.8xlarge = $8.67 per Hour (On Demand) Render Time - 1 minutes

p2.xlarge maybe the question doesnt make sense

 

[seachnasaigh]

I have just signed up with ....the google. The Console is nicer. Its limited to 8 vCpu during the trial, which is fine to install things and even test 3delight rendering maybe, although still not sure what exactly a vCPU equals -half a cpu core?

Probably just refers to a single processor core; an AMD octal would be 8 CPU, a HyperThreaded Xeon octal would be 16 CPU. But I'm just guessing.
That's the way Vue, Lux, and Task Manager reckon CPU count. It's a misnomer, but a commonly encountered misnomer.

On the AWS pricing front -who is good at mathematics to work out the best deal?
The results.....

G2.2xlarge = $0.13 per hour (Spot Instance) Render Time - 21 minutes => 4.55 cents
G2.8xlarge = $0.49 per Hour (Spot Instance) Render Time - 7 minutes => 5.72 cents
c4.8xlarge = $1.58 per Hour (Spot Instance) Render Time - 10 minutes => 26.3 cents
P2.xlarge = $0.29 per Hour (Spot Instance) Render Time - 8 minutes => 3.87 cents
p2.8xlarge = $8.67 per Hour (On Demand) Render Time - 1 minutes => 14.4 cents

That would be the cost for the stated render time at the stated rate, if I understand you correctly.

 

[DigiDotz]

I think so

So although every render scene and size will be different, using this benchmark and as you pay per hour even if render takes 1 minute

G2.2xlarge - best for single renders if they fit on a 4GB GPU
P2.xlarge - Renders requiring 4Gb -12Gb GPU
P2.8xlarge - Rendering lots of jobs in a rush, requiring 4Gb -12Gb GPU
C4.8xlarge - Renders that wont fit 12GB GPU's can use vCpu's

....