I know there have been some problems in the past but I’ve been having great success running TD in the cloud on AWS, using their G4dn instances, which have a NVIDIA Tesla T4 with 16gb of GPU RAM. Using a newer remote desktop tool like Parsec (which is designed for gamers to share screens at 60fps with low latency) makes working on the client side as fluid as working on my desktop, but with about 20ms latency. All the NVIDIA-specific features that I tested also work.
you need to get a quota increase in order to use the g4 instances. From my experience they will almost automatically approve a 16-core increase in any region.
As for the rates, the base machine with 4 CPU cores and 16gb of RAM is $0.82/hr (this includes Windows licenses, it’s cheaper if you can bring your own). Plus you get 25gbps network bandwidth.
I currently have 4 of the 4xlarge machines running vMix and routing a ton of NDI streams between machines with no frame loss. I imagine TouchIn/Out would also work fine but I haven’t tested it.
And fwiw, both Microsoft Azure and IBM Cloud have similar GPU machines (Tesla V100/P100 GPUs) also with Windows images that have the drivers loaded. It looked like IBM also did not require you to request a quota increase to use them. You could also look at the Quadro VDI offerings which are made to be a “second desktop” but those have some monthly fees as well as the hourly.
Update: looks like you can also use a G4 AMI that has Teradici support which is meant for high end gfx apps, which gives you 4k60p in 4:4:4 and probably solves the monitor issue. It’s $.50 per hour or you can buy a Teradici license for around $100 a month.
The link you posted seems to be GPU services designed for ML and not running interactively. It seems you deploy your workload as a docker container via ssh and I didn’t see any support for Windows. Also their location in Sweden means guaranteed latency from the US west coast for anything with realtime capabilities.
Lately I’ve been running all on AWS using the Teradici AMI (which is an extra $.50 per hour or $250/yr) but it has the Nvidia drivers installed and gives me great 60fps for working locally even on complex TD projects.
Now if Derivative would support by-the-minute cloud licensing (even for just TouchPlayer) I could create massive clusters of machines to tackle literally any GPU intensive workload (with one-way latency out of the cloud around 60ms).
Hey @flytrap - thanks for posting your experience with this. I’m curious about the Teradici AMI, are you using their PCoIP client to access this or still using Parsec as described in your original post? I assume it is not mandatory to use their client but just wanted to confirm.
Teradici is great if you need multimonitor but Parsec has come a long way (and can also do multi monitor but I haven’t used it) and has some really useful tools to get setup on different cloud providers. I use Parsec mostly now.
Hello, thank you for your post ! This was great to read.
I do too using AWS cloud GPU instances to run TD.
How about Value ladder, working correctly? Because I had some problem using Value Ladder in TD running in EC2 instances. it skipps some value from 1 - 101 or more bigger number.
I use NICE DCV, RDP, Persec. but each Protocol had different attitude but either one didn’t works correctly.
well, other than that, experiience was wonderful.
You don’t even feel you using cloud instances, You feel like as if you are using your PC under the table.
NICE DCV was good in terms of image quality, but, it had bit of latency and tiering in
quick motion or changing image compare to Persec.
Persec respons so quick, say, if you wanted to flash color back and forse with your keyboard input, you hardly notice for the latency and had no tiering happing.(at least I check).
So, I do recomend Persec if you need quick motion, and go with NICE DCV if you need image quality or color accuracy.
