SUNK Anywhere extends CoreWeave’s unified training system to cross-cloud and on-prem environments, giving you a faster and more secure path from POC to production.
No more losing time switching tools, workflows, or operating models every time the infrastructure changes. SUNK Anywhere lets you operate demanding AI workloads with consistent workflows and operational discipline across environments and clouds. That consistency helps platform teams expand without fragmentation, and researchers use familiar scheduling and workflows as their infrastructure footprint grows.
1
00:00:03,520 --> 00:00:04,160
Hi.
2
00:00:04,400 --> 00:00:05,760
My name is Deok.
3
00:00:05,760 --> 00:00:08,080
And I'm the product manager for SUNK here at CoreWeave.
4
00:00:08,400 --> 00:00:10,920
SUNK is the unified training system
5
00:00:10,920 --> 00:00:13,480
that teams already run on CoreWeave.
6
00:00:13,480 --> 00:00:16,760
Now, SUNK Anywhere extends it to your infrastructure,
7
00:00:16,760 --> 00:00:18,160
wherever you have it.
8
00:00:18,160 --> 00:00:20,440
As workloads grow beyond one provider,
9
00:00:20,440 --> 00:00:24,120
most teams end up with a different stack in every environment.
10
00:00:24,120 --> 00:00:26,440
The different tooling and operating practices
11
00:00:26,440 --> 00:00:27,920
across these environments
12
00:00:27,920 --> 00:00:30,640
lead to friction that slows down your model development.
13
00:00:30,640 --> 00:00:35,040
SUNK Anywhere enables one operating model across GKE,
14
00:00:35,040 --> 00:00:37,600
at Google Cloud, EKS with AWS,
15
00:00:37,600 --> 00:00:40,560
and other Kubernetes deployments, wherever you have it.
16
00:00:40,560 --> 00:00:41,800
Existing features are still supported:
17
00:00:41,800 --> 00:00:42,960
Existing features are still supported:
18
00:00:42,960 --> 00:00:43,120
Existing features are still supported:
19
00:00:43,120 --> 00:00:45,040
the SUNK pod Scheduler, for example,
20
00:00:45,040 --> 00:00:48,120
allows running inference, sandboxes, and training workloads
21
00:00:48,120 --> 00:00:50,880
side-by-side on one cluster for higher cluster utilization.
22
00:00:50,880 --> 00:00:52,760
side-by-side on one cluster for higher cluster utilization.
23
00:00:52,760 --> 00:00:55,280
User provisioning supports SCIM and LDAP
24
00:00:55,280 --> 00:00:57,600
for safer and faster access management.
25
00:00:57,600 --> 00:01:00,280
And researchers get their own customizable
26
00:01:00,280 --> 00:01:02,120
environment to boost their productivity.
27
00:01:02,120 --> 00:01:04,320
Using CoreWeave provided agent skills,
28
00:01:04,320 --> 00:01:04,760
you save time in deploying SUNK
29
00:01:04,760 --> 00:01:05,600
you save time in deploying SUNK
30
00:01:05,600 --> 00:01:07,960
you save time in deploying SUNK
31
00:01:07,960 --> 00:01:09,600
by having the appropriate configs and test cases
32
00:01:09,600 --> 00:01:11,520
to validate you have a working cluster.
33
00:01:11,520 --> 00:01:13,200
You don't just get SUNK.
34
00:01:13,200 --> 00:01:16,440
The deployment also stands up the surrounding components
35
00:01:16,440 --> 00:01:18,040
a production cluster needs,
36
00:01:18,040 --> 00:01:21,280
like shared storage, GPU and Slurm metrics,
37
00:01:21,280 --> 00:01:23,240
wired into your cluster's monitoring,
38
00:01:23,240 --> 00:01:26,280
and custom dashboards ready to use.
39
00:01:26,280 --> 00:01:28,160
On GKE, for example,
40
00:01:28,160 --> 00:01:31,160
these dashboards drop into your cloud monitoring out of the box.
41
00:01:31,160 --> 00:01:34,000
And EKS follows the same pattern.
42
00:01:34,000 --> 00:01:36,880
Support is 24-7 over Slack
43
00:01:36,880 --> 00:01:39,120
with Core Weave’s Teams.
44
00:01:39,120 --> 00:01:41,560
The same support quality the customers already know.
45
00:01:41,560 --> 00:01:44,560
SUNK anywhere, one unified training system,
46
00:01:44,560 --> 00:01:46,440
now on your infrastructure.
47
00:01:46,520 --> 00:01:48,360
Reach out if you want to try it.
