Multi-cloud for demanding workloads
End to end integration with data centers. 5000 computation hours for trial.
Ready to compute
Installed operation system with lates software and remote desktop access. Mounted storage for output data
ASAP
Prepared solutions for a wide range of requirements. i.e 512 Cores shape or 10TB RAM machine.
Cost-cutting circle
Robot will match stats with cheaper configuration. Step by step, price drops 2-5 times
Cloud Providers
Use case #1: Oracle Cloud for Weather forecast [ WRF ]
Heavy MPI task
Total MPI time - 40.91% (2 nodes, 36 mpi process per node, 2 OpenMP thread per mpi process);

MPI_Bcast - 22.6 %, msg. size - 4 byte (456480 colls in 140 sec.)

MPI processes exchanges between neighbors;

MPI_Wait is highly unbalanced (pic. Communication balance by rank ).

Total ram usage: 86.9 Gb.


Use case #2: Amazon AWS Cloud for High Freaquency Electromagnetic modeling [ ANSYS HFSS ]
How control panel works
  • 1
    Modeling on personal desktop
    Engineer creating model and upload input data to rocketcompute web panel.
  • 2
    Launch high-performance machine or cluster
    Choose matching shape and click launch.
  • 3
    Start computation
    Log in to high-performance machine open preinstalled software and start
  • 4
    Monitoring
    There is ANSYS log right from rockectcompute web panel.
  • 5
    Work with the result
    After computation ends we shut down high performance machine and launch cheaper graphic optimized machine. And you can work with data without warring to spend lot's of money on clouds.

    There is also possible to download output data to your local machine
Performance Agent Code
How analytics agent works
  • 1
    Copy
    Engineer creating model and upload input data to rocketcompute web panel.
  • 2
    Launch high-performance machine or cluster
    Choose matching shape and click launch.
  • 3
    Start computation
    Log in to high-performance machine open preinstalled software and start
  • 4
    Monitoring
    There is ANSYS log right from rockectcompute web panel.
  • 5
    Work with the result
    After computation ends we shut down high performance machine and launch cheaper graphic optimized machine. And you can work with data without warring to spend lot's of money on clouds.

    There is also possible to download output data to your local machine
j
Benchmark Algorithm

Blah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blahBlah blah blah
Pricing
About
Failing to pick and integrate the right cloud configuration for computational heavy workloads will lead to overspending (80% in some cases) and a huge waste of time.

Not only distributed computations themselves require thought-through architecture to perform well (do you really need RDMA? what storage to pick to match your workload profile? etc.), but Cloud IaaS perform in unpredictable way depending on many hidden factors, so relying on price per core*hour to estimate your budget does not cut it, not even close.

We are a team of engineers with experience in applied math, HPC, virtualization and Cloud architecture. We did many simulations ourselves and for our customers, and we feel like we know quite a bit about how to pick the right configuration, which best fits the workload, and automate its set up.

Now we'd like to share our expertise in a form of SaaS tools available to everyone - we are at a semi-automatic stage now, so we would love to talk about your unique case as it helps us to see the whole picture.

We systematically benchmark all clouds, launching hundreds of simulations simultaneously and provide you access to conclusions and tools to do the same yourself.

Please do not hesitate, shoot us a message with any question or problem you might have. We'll be happy to help.

Eugene
E-mail:
e@rocketcompute.com