DiPhyxDiPhyx
Workspace
Prepaid · by the hour · three clouds

GPU and HPC machinesa minute after you ask

Pick a machine and it is yours in about a minute.No cloud account of your own, no keys to hold — you pay for the hours it runs.

WorkspaceSee pricing
GPU trainerG4dn XLarge · Tesla T4 · 4 coresRunning
Mesh solverT4g 2XLarge · 8 cores · 32 GBStarting
Nightly batchD2s v5 · 2 cores · 8 GBStopped
~1 minReady infrom create to running
$0.014Fromper hour, the smallest machine
NVIDIAGPUsTesla T4 and L4 cards, by the hour
$0.00Stoppedper hour — only the disk counts
The panel

A machine you can work on

No cloud console, no second tab, no SSH client on your laptop.

diphyx.com/workspace/servers
GPU trainerRunning
G4dn XLarge · Tesla T4 · 4 cores · AWS
StatusRunningThe cloud says it is up.
Address54.71.20.118Public address
Machine4 cores16 GB memory
Cost$15.78per day at this size
Terminal
$nvidia-smi--query-gpu=name,memory.total --format=csv,noheaderTesla T4, 15360 MiB$docker run--gpus all -v /data:/data trainer:latestepoch 25/40 · loss 0.121 · 37sepoch 26/40 · loss 0.118 · 38s$
DiskReady
200 GB
Firewall2 open
22 · 8000
dxflowReady
dxflow 1.8.2

Pick the size of the problem

Three kinds of machine, with the rate you would pay for each.

CPU machines
from$0.014/ hour

Editing, serving, testing and anything that just needs to stay up.

T2 Micro1 core · 1 GB$0.014
E2 Medium2 cores · 4 GB$0.040
D2s v52 cores · 8 GB$0.115
GPU machinesMost asked for
from$0.631/ hour

Training, inference and anything that wants CUDA under it.

G4dn XLargeTesla T4 · 4 cores$0.631
G4dn 2XLargeTesla T4 · 8 cores$0.902
G6 XLargeNVIDIA L4 · 4 cores$0.966
Compute machines
from$0.081/ hour

Solvers and batch runs that want more cores and more memory than a desk.

T4g Large2 cores · 8 GB$0.081
T4g XLarge4 cores · 16 GB$0.161
T2 2XLarge8 cores · 32 GB$0.445
How the bill works
One tick on the create form

Machines that arriverunning dxflow

Containerized workflows, files, shells and an agent —installed, licensed and open in a console before you get there.

What is dxflow
agent
Run the trainer on the GPU, with the data in /dataBuilt the workflow and started it. Docker has the CUDA image already.How far along is it?Epoch 26 of 40, loss 0.118 and still falling. About 9 minutes left.Serve the checkpoint on port 8000 when it landsQueued. The port is open, so it will answer as soon as the run ends.
workflow
Training runRunning
pytorch:2.4-cudagpus: all/data
26 / 40
artifact
dataset.zip4.2 GB
config.yaml1.4 KB
ckpt-26.pt812 MB
The catalogue

Software that isalready built

Every workflow is a container, built and pinned.It opens in a tab — nothing to compile, no cluster queue.

GenomicsMolecularStructuralSimulationVisualizationAnalyticsAIDevelopmentInfrastructureDesktopBrowserGraphicsMessaging
Browse workflows
GROMACS4 cores · 32 GB · GPUGROMACS4 cores · 32 GB · GPU
Jupyter Lab4 cores · 8 GB · browserJupyter Lab4 cores · 8 GB · browser
OpenFOAM4 cores · 32 GB · shellOpenFOAM4 cores · 32 GB · shell
Ollama4 cores · 8 GB · browserOllama4 cores · 8 GB · browser
Scipion4 cores · 64 GB · GPUScipion4 cores · 64 GB · GPU
VS Code Server2 cores · 4 GB · browserVS Code Server2 cores · 4 GB · browser

What people run on them

Work that outgrew a laptop, without the cluster queue it usually comes with.

AI and machine learning

Train, fine-tune or serve a model on a GPU for the afternoon it takes.

Simulation and HPC

Solvers, molecular dynamics and CFD runs that outgrew a desk machine.

Data and batch work

Pipelines, renders and long jobs that need a big disk for a while.

No bill arrives later

You buy credit first, and every machine spends from it while it runs.

PrepaidA balance you top up. There are no invoices, and no debts to settle.
By the hourCounted while a machine runs. A stopped one only keeps paying for its disk.
Warned firstAn email while hours remain, and another before anything is deleted.
Start with a small machineNothing runs until you say so, and nothing is charged until it does.
WorkspaceSee pricing