CPU Cluster Renderings in a CORTEX workstation

Hello,dear friend

The forum is too COOL!!!

Now we used 8 ALXEAXT cameras in a 2D 2K film project,the DIT hopes that we could transcode more than 8 ARRIRAW files at the same time into a CORTEX workstation.

What we need the cortex feature

1.We hope that the CORTEX could support the CPU Cluster Renderings,I mean that the core architecture of the software could support more than 4 or 6,8Xeon processor.

2.When we input all ARRIRAW files into the CORTEX workstation,the dual CPU processor could decode the ARRIRAW files interactively
eg,one CPU decodes the odd shot(shot1,shot3,shot5,shot7…),another CPU decodes the even shot
(shot2,shot4,shot6,shot8…).

thank you so much, THE FORUM IS GREAT!

1 Like

We do support mutli-core systems, and will take advantage of most of the CPU processing power when transcoding to different deliverables.

Cortex is designed to maximize the speed of encoding each clip, so you’ll likely see that the CPU usage is quite high when doing a single transcode. Rather than trying to run multiple processes at a time, a single process is multi-threaded.

Doing multiple transcodes at a time would slow down each process since they would take resources away from each other, and the overall throughput would not increase.

One other thing to note is that with some formats, most of the work for decoding the frames is done on the GPU (and ARRIRAW is one of them). So to achieve the best performance for the case you mentioned, we would recommend running on a system with a new, fast GeForce card like the 780ti.

I would recommend running some tests to see whether the throughput of a single system meets your requirements, and if you really do need more throughput than a single system can offer, you should check out the Enterprise Edition:

Hello,dear dave

thank you so much for your suggestions

1.if possible,could you tell me that how many Xeon processors could the CORTEX software support?

2.Doing multiple transcodes at a time would slow down each process since they would take resources away from each other, and the overall throughput would not increase.

Sorry I didnot express it clearly

I think the cluster renderings is too important.Because the R3D and MXF use the CPU to decode。In this film project ,we use 8 ALEXA XT and 3 RED dragon.but the 6k footage is about 5%

I assume that the CORTEX could support the 10 Xeon processors,the first CPU decodes the first footage …the tenth CPU decodes the tenth footages,they are decoded at the same time。Again,the first CPU decodes the 11th footage…the tenth CPU decodes the 20th footage.

3.So to achieve the best performance for the case you mentioned, we would recommend running on a system with a new, fast GeForce card like the 780ti.

How many GeForce 780ti does the CORTEX software support?

thank you so much again

We have only tested it on up to 12 cores (2 x 6c Xeons). With many formats it will saturate all cores when doing a single transcode.

If you find otherwise, we can discuss ways to optimize further… If you are keen to experiment further, it is possible to run the actual rendering application via the command line so you could launch as many processes at a time as you would like.

Today, we support using up to 2 GPUs for a single process, but most configurations start with a single GPU, which matches pretty well with typical storage and CPU resources that people have.

Hello,dear Dave

thank you so much for your reply。

  1. the develop feature of the CPU Cluster Renderings into the CORTEX is a realistic and practical request or this is just a fantasy

2.Now I assume that we had 4 CORTEX workstations,all ARRIRAW files into one card are copied in the storge of the first CORTEX workstation。So could we achieve when the first CORTEX workstation transcodes the first ARRIRAW file,the second CORTEX workstation aslo transcodes the second ARRIRAW file at the same time?

thank you so much

sincerely

What is the ultimate goal here?

I assume that it is simply to process all the footage as quickly and efficiently as possible.

Currently, I believe the existing implementation meets this goal better than the implementation you are describing, but if you have a test case that demonstrates otherwise, we will certainly consider further improvements.

Yes, each workstation will start transcoding the next clip waiting in the queue as it finishes its current task.

But the files must be on shared storage in order for multiple Cortex systems to process them simultaneously with a clustered Enterprise system.