Cryo-EM software installed and working on your machine, remotely
RELION, cryoSPARC and the rest of the stack, built against your driver, CUDA and scheduler and left in a state your next student can maintain. Most single-workstation installs are done within a day of access, from $500.
Remote only. You or your IT handle the hardware and the base operating system; I take it from the GPU driver and CUDA up. Need a disk or GPU swapped, or the OS reinstalled? Have that done locally with SSH enabled, then send me access.
To quote, I need:
- Remote access for the job: a temporary SSH account or a screen share
- The operating system and its version
- The GPU model (and the driver version, if you know it)
I do every job myself: Ian Cooney, Ph.D. (Stanford; first-author Science and Nature Communications).
Already have an error message? Paste it in and I can usually name the cause before quoting.
- Workstation installfrom $500
RELION, cryoSPARC and supporting tools built against your CUDA and driver versions, documented so your next student can rebuild it. Usually done within a day of access.
- Cluster or multi-node install$1,500–3,000
Scheduler integration, environment modules, shared storage and scratch layout, plus submission templates that request the right resources.
Fixed price agreed before any work starts. If the job turns out to be smaller than quoted, the quote goes down.
What I install and configure
RELION
Source builds against your CUDA and MPI versions, GPU selection, MotionCor2 and CTFFIND wiring, and the scheduler integration that makes multi-node runs actually distribute instead of silently falling back to one node.
cryoSPARC
Master and worker installation, worker registration and lane configuration, CUDA toolkit matching, SSD caching, licence setup, and the reverse-proxy and firewall work needed to reach it from outside the machine it runs on.
GPU, CUDA and drivers
The layer where most installs actually fail. Driver and toolkit version compatibility, multi-GPU visibility, memory limits, and diagnosing whether a crash is your build, your card or your data.
Clusters and schedulers
SLURM and PBS submission templates that request the right resources, environment modules, shared storage and scratch layout, and queue configuration that does not leave GPUs idle.
Supporting tools
CTFFIND, Topaz, crYOLO, IMOD, EMAN2, ChimeraX, Phenix and Coot, installed in environments that do not fight each other over Python and CUDA versions.
Cloud and remote workstations
AWS or on-prem GPU instances sized for your particle counts, with data transfer that finishes and costs that do not surprise you.
Why this is worth paying for
The install is not the hard part. The hard part is that a broken environment fails in ways that look like a data problem: jobs that die at random particle counts, refinements that run but never improve, GPUs that sit idle while a job takes three days. People lose weeks deciding whether their sample is bad when the answer is that MPI was never configured.
A postdoc week costs more than this service does. If the machine has been fighting you for more than a couple of days, that is the point at which handing it over is the cheaper option.
Tell me what is broken
OS, GPU model and the error, if you have one. You get a fixed price back within two business days. Remote only: the machine needs to boot with its OS and network first.
Questions
How does remote installation work on our machines?
Usually a screen share or a temporary SSH account you create and revoke, working alongside whoever administers the machine. I do not need permanent access, and I do not need credentials for anything beyond the machine I am installing on. If your institution requires internal IT to do the work, I can direct them instead and review the result.
Do you do hardware repair or on-site work?
No. The work is remote only. The machine has to boot with its operating system installed and a network connection; you or your IT handle that, along with any disk, GPU or other hardware change. I take it from the GPU driver and CUDA up. If the OS needs reinstalling after a hardware failure, have it done locally with SSH enabled and I pick it up from there.
What do you need from us to quote and start?
Three things: remote access for the job (a temporary SSH account or a screen share), the operating system and its version, and the GPU model. The driver version, the software you want and any error text help too. With those I can quote a fixed price; everything else I can read off the machine once I have access.
How long does it take?
Most single-workstation installs are done within a day once I have access. Cluster work depends more on your environment than on the software, because the time goes into scheduler and storage configuration rather than the install itself. I tell you which of the two you are before quoting.
What does it cost?
Workstation install, from $500. Cluster or multi-node install, $1,500–3,000. Fixed price agreed before any work starts. If the job turns out to be smaller than quoted, the quote goes down. Send the machine spec and what you are trying to run and you will get a number, not an hourly estimate that grows.
Our install nearly works but jobs crash. Is that this service?
Yes, and it is the most common request. A build that compiles but fails at run time is almost always a version mismatch between the driver, the CUDA toolkit and the software, or a scheduler requesting resources that do not exist. Send the error text and I can usually tell you the cause before quoting.
Will you document it so we can maintain it ourselves?
Yes. You get the build steps, the environment configuration and the submission templates in writing. The point is that your next student can rebuild it without calling anyone, which is also why this is worth doing properly once.