KVARK Infrastructure
Hardware and operations, under your control.
Run KVARK on LM TEK servers or your own compatible Kubernetes infrastructure, choose its deployment model, and manage its hardware, environments and releases.
Run KVARK on LM TEK servers or your own compatible Kubernetes infrastructure, in the deployment model you choose.
Business value
Dense compute that asks less of your facility
In its specified configuration, the LM TEK server is liquid-cooled inside its chassis. You need no facility water or external cooling distribution unit, and fewer data-room changes. It aims to use less energy and give off less noise and heat, which can lower your costs.
Room to grow on infrastructure you choose
Run KVARK on compatible Kubernetes infrastructure and grow it by adding servers. KVARK places its services and models wherever there is capacity, following placement policies, so your hardware is better used and less needs managing by hand.
The LM TEK server
Put up to eight NVIDIA H200 GPUs to work in one LM TEK 4U server, KVARK’s primary hardware option. Its liquid cooling helps keep chip temperatures stable.
Up to eight NVIDIA H200 GPUs in one liquid-cooled 4U server, KVARK’s primary hardware option.
- In the specified configuration, the whole server draws about 6 kW.
- That configuration also supports operation at 30 °C ambient.
- The facility provides ventilation and heat removal, sized to the installed load.
Fig. 2 · The LM TEK server
- 1Operation at 30 °C ambientsupported for the specified configuration
- 2Up to eight NVIDIA H200 GPUsin one 4U server
- 3Closed-loop liquid coolingkept entirely within the chassis
- 4No facility water connectionand no external cooling distribution unit
- 5Approximately 6 kWwhole server, specified configuration
- 6Ventilation and heat removalby the facility, sized to the installed load
Deployment options
Choose where KVARK runs, on LM TEK servers or your own compatible Kubernetes infrastructure, and the deployment model that suits your operational and data requirements.
Choose your deployment model: on premises, private or hybrid cloud, or air-gapped with offline-ready components and integrations.
- On premises, or in a private or hybrid cloud.
- Air-gapped: fully isolated, with components and integrations fit for offline use.
Sketch · where KVARK can run
KVARKkvark.bank.example
- AI Services
- AI Hub
- AI Development
- Platform
Where it runs
- LM TEK serversThe primary hardware option
- Your Kubernetes environmentCompatible infrastructure you select
Deployment model
- On-premises
- Private cloud
- Hybrid cloud
- Air-gapped
- Air-gapped: fully isolated, runs offline
- The bank’s choice: LM TEK servers, on premises
Infrastructure management
Monitor the hardware and systems KVARK runs on and allocate their resources. Get alerts and manage incidents when something goes wrong.
Monitor hardware and systems, allocate resources, get alerts and manage incidents.
- Administer the infrastructure, with reports and analytics.
Example · hardware monitoring
Monitoring
Hardware
Monitor real-time hardware metrics including GPU, CPU, memory, and storage.
OverviewGraphs
GPUs
GPU 1
NVIDIA H200
54.5°Power Usage571.8W
GPU 2
NVIDIA H200
55.5°Power Usage588.3W
GPU 3
NVIDIA H200
52.5°Power Usage561.2W
GPU 4
NVIDIA H200
57.0°Power Usage602.6W
GPU 5
NVIDIA H200
51.5°Power Usage544.1W
GPU 6
NVIDIA H200
55.0°Power Usage579.0W
GPU 7
NVIDIA H200
57.5°Power Usage609.7W
GPU 8
NVIDIA H200
53.0°Power Usage556.4W
Environments and releases
Keep development and testing apart from production: run DEV, TEST and PROD, and move changes between them with CI/CD and controlled migration.
Run DEV, TEST and PROD, and move changes between them with CI/CD and controlled migration.
- Agree system updates with platform administration, to suit operational needs.
Sketch · from DEV to PROD
DEVkvark-dev.bank.exampleActive
- credit-memo-draftingSkill
TESTkvark-test.bank.exampleActive
- credit-memo-draftingSkill
PRODkvark.bank.exampleActive
- credit-memo-draftingSkill
- System updateAgreed with platform administration
Business value
Dense compute that asks less of your facility
The LM TEK server’s specified configuration needs no facility water and fewer data-room changes. It aims to use less energy, which can lower costs.
Room to grow on infrastructure you choose
Grow compatible Kubernetes infrastructure by adding servers. KVARK places services and models by capacity and placement policies, so your hardware is better used.
Start with one use case. Run it under your rules.
We show KVARK on your kind of documents, your access rules and your deployment constraints. Most teams start with a four to six week analysis and a six week pilot.
We show KVARK on your kind of documents, access rules and deployment constraints.