Skip to content

Commit 25f4317

Browse files
peterschmidt85peterschmidt85
committed
[Docs] Improve Kubernetes documentation
Updated `FAQ`
1 parent c1c0f3b commit 25f4317

1 file changed

Lines changed: 82 additions & 42 deletions

File tree

‎docs/overrides/home.html‎

Lines changed: 82 additions & 42 deletions
Original file line numberDiff line numberDiff line change
@@ -438,81 +438,121 @@ <h4>Co-Founder @CUDO Compute</h4>
438438
<div class="tx-landing__highlights_text">
439439
<h2>FAQ</h2>
440440
</div>
441-
441+
442442
<div class="tx-faq__list">
443+
<div class="tx-faq__item">
444+
<div class="tx-faq__item-title">
445+
How does dstack differ from Slurm?
446+
<div class="tx-faq__item-title-icon">
447+
</div>
448+
</div>
449+
450+
<div class="tx-faq__item-body">
451+
<p>
452+
<span class="highlight">dstack</span> fully replaces Slurm. Its
453+
<a href="/docs/concepts/tasks" target="_blank">tasks</a> cover job submission, queuing, retries, GPU
454+
health checks, and scheduling for single-node and distributed runs.
455+
</p>
456+
457+
<p>
458+
Beyond job scheduling, <span class="highlight">dstack</span> adds
459+
<a href="/docs/dev-environments" target="_blank">dev environments</a> for interactive work,
460+
<a href="/docs/concepts/services" target="_blank">services</a> for production endpoints, and
461+
<a href="/docs/concepts/fleets" target="_blank">fleets</a> that give fine-grained control over
462+
cluster provisioning and placement.
463+
</p>
464+
465+
<p>
466+
You get one platform for development, training, and deployment across cloud, Kubernetes, and
467+
on-prem.
468+
</p>
469+
</div>
470+
</div>
471+
443472
<div class="tx-faq__item">
444473
<div class="tx-faq__item-title">
445474
How does dstack compare to Kubernetes?
446475
<div class="tx-faq__item-title-icon">
447476
</div>
448477
</div>
449-
478+
450479
<div class="tx-faq__item-body">
451-
<p>Kubernetes is a widely used container orchestrator designed for general-purpose deployments.
452-
To efficiently support GPU workloads, Kubernetes typically requires custom operators, and it
453-
may not offer the most intuitive interface for ML engineers.</p>
454-
455-
<p><span class="highlight">dstack</span> takes a different approach, focusing on container
456-
orchestration specifically for AI
457-
workloads, with the goal of making life easier for ML engineers.</p>
458-
459-
<p>Designed to be lightweight, <span class="highlight">dstack</span> provides a simpler, more
460-
intuitive interface for
461-
development,
462-
training, and inference. It also enables more flexible and cost-effective provisioning
463-
and management of clusters.</p>
464-
465-
<p>For optimal flexibility, <span class="highlight">dstack</span> and Kubernetes can complement
466-
each other: dstack can handle
467-
development, while Kubernetes manages production deployments.</p>
480+
<p>
481+
Kubernetes is a general-purpose container orchestrator. <span class="highlight">dstack</span> also
482+
orchestrates containers, but it provides a lightweight and streamlined interface that is purpose
483+
built for ML.
484+
</p>
485+
486+
<p>
487+
You declare
488+
<a href="/docs/concepts/dev-environments" target="_blank">dev environments</a>,
489+
<a href="/docs/concepts/tasks" target="_blank">tasks</a>,
490+
<a href="/docs/concepts/services" target="_blank">services</a>, and
491+
<a href="/docs/concepts/fleets" target="_blank">fleets</a>
492+
with simple configuration. dstack provisions GPUs, manages clusters via fleets with fine-grained
493+
controls, and optimizes cost and utilization, while keeping a simple UI and CLI.
494+
</p>
495+
496+
<p>
497+
If you already use Kubernetes, you can run <span class="highlight">dstack</span> on it via the <a
498+
href="/docs/concepts/backends#kubernetes" target="_blank">Kubernetes</a> backend.
499+
</p>
468500
</div>
469501
</div>
470-
502+
471503
<div class="tx-faq__item">
472504
<div class="tx-faq__item-title">
473-
How does dstack differ from Slurm?
505+
Can I use dstack with Kubernetes?
474506
<div class="tx-faq__item-title-icon">
475507
</div>
476508
</div>
477-
509+
478510
<div class="tx-faq__item-body">
479511
<p>
480-
Slurm excels at job scheduling across pre-configured clusters.
512+
Yes. You can connect existing Kubernetes clusters using the <a
513+
href="/docs/concepts/backends#kubernetes" target="_blank">Kubernetes</a> backend and run
514+
<a href="/docs/concepts/dev-environments" target="_blank">dev environments</a>,
515+
<a href="/docs/concepts/tasks" target="_blank">tasks</a>, and
516+
<a href="/docs/concepts/services" target="_blank">services</a> on it.
517+
Choose the Kubernetes backend if your GPUs already run on Kubernetes and your team depends on its
518+
ecosystem and tooling.
519+
See the
520+
<a href="/docs/guides/kubernetes" target="_blank">Kubernetes</a> guide for setup and best practices.
481521
</p>
482-
483-
<p><span class="highlight">dstack</span> goes beyond scheduling, providing a full suite of
484-
features tailored to ML teams,
485-
including cluster management, dynamic compute provisioning, development environments, and
486-
advanced monitoring. This makes <span class="highlight">dstack</span> a more comprehensive
487-
solution for AI workloads,
488-
whether in the cloud or on-prem.
522+
<p>
523+
If your priority is orchestrating cloud GPUs and Kubernetes isn’t a must, <a
524+
href="/docs/concepts/backends#vm-based" target="_blank">VM-based</a> backends are a better fit
525+
thanks to their native cloud integration.
526+
For on-prem GPUs where Kubernetes is optional, <a href="/docs/concepts/fleets#ssh"
527+
target="_blank">SSH fleets</a> provide a simpler and more lightweight alternative.
489528
</p>
490529
</div>
491530
</div>
492-
531+
493532
<div class="tx-faq__item">
494533
<div class="tx-faq__item-title">
495534
When should I use dstack?
496535
<div class="tx-faq__item-title-icon">
497536
</div>
498537
</div>
499-
538+
500539
<div class="tx-faq__item-body">
501540
<p>
502-
dstack is designed for ML teams aiming to speed up development while reducing GPU costs
503-
across top cloud providers or on-prem clusters.
541+
<span class="highlight">dstack</span> accelerates ML development with a simple, ML‑native interface.
542+
Spin up <a href="/docs/concepts/dev-environments" target="_blank">dev environments</a>, run
543+
single‑node or distributed <a href="/docs/concepts/tasks" target="_blank">tasks</a>, and deploy <a
544+
href="/docs/concepts/services" target="_blank">services</a> without infrastructure overhead.
504545
</p>
505-
546+
506547
<p>
507-
Seamlessly integrated with Git, dstack works with any open-source or proprietary frameworks,
508-
making it developer-friendly and vendor-agnostic for training and deploying AI models.
548+
It radically reduces GPU costs via smart orchestration and fine‑grained <a
549+
href="/docs/concepts/fleets" target="_blank">fleet</a> controls, including efficient reuse,
550+
right‑sizing, and support for spot, on‑demand, and reserved capacity.
509551
</p>
510-
552+
511553
<p>
512-
For ML teams seeking a more streamlined, AI-native development platform, <span
513-
class="highlight">dstack</span>
514-
provides an alternative to Kubernetes and Slurm, removing the need for
515-
MLOps or custom solutions.
554+
It is 100% interoperable with your stack and works with any open‑source frameworks and tools, as
555+
well as your own Docker images and code, across cloud, Kubernetes, and on‑prem GPUs.
516556
</p>
517557
</div>
518558
</div>

0 commit comments

Comments
 (0)