[GH-ISSUE #26] How to access GPU with agentEnv #15

New issue

Closed

opened 2026-10-29 14:18:21 +00:00 by bsakaguchi · 2 comments

bsakaguchi commented

2026-10-29 14:18:21 +00:00

Owner

Originally created by @fy2462 on GitHub (Jul 28, 2026).
Original GitHub issue: https://github.com/kvcache-ai/AgentENV/issues/26

Hi guys:

Thank you very much for the great work on open source.

I have a question: how can Firecracker make efficient use of GPU resources, and is this agent infra system restricted to running only on CPU clusters?

Originally created by @fy2462 on GitHub (Jul 28, 2026). Original GitHub issue: https://github.com/kvcache-ai/AgentENV/issues/26 Hi guys: Thank you very much for the great work on open source. I have a question: how can Firecracker make efficient use of GPU resources, and is this agent infra system restricted to running only on CPU clusters?

bsakaguchi closed this issue

2026-10-29 14:18:21 +00:00

bsakaguchi commented

2026-10-29 14:18:22 +00:00

Author

Owner

@yingdi-shan commented on GitHub (Jul 28, 2026):

Hi, thank you for your interest in AgentENV!

Currently, AgentENV does not support GPU, since it's built on Firecracker microVMs, which are CPU only by design. Firecracker doesn't expose the PCI or device passthrough needed for GPU access. So yes, at the moment the agent infra is restricted to CPU clusters.

Supporting GPU workloads would require substantial architectural changes, including adding a QEMU backend alongside Firecracker, redesigning our snapshot mechanism, and more. It's a nontrivial undertaking, so it would be a longer term effort rather than something we can do quickly.

@yingdi-shan commented on GitHub (Jul 28, 2026): Hi, thank you for your interest in AgentENV! Currently, AgentENV does not support GPU, since it's built on Firecracker microVMs, which are CPU only by design. Firecracker doesn't expose the PCI or device passthrough needed for GPU access. So yes, at the moment the agent infra is restricted to CPU clusters. Supporting GPU workloads would require substantial architectural changes, including adding a QEMU backend alongside Firecracker, redesigning our snapshot mechanism, and more. It's a nontrivial undertaking, so it would be a longer term effort rather than something we can do quickly.

bsakaguchi commented

2026-10-29 14:18:23 +00:00

Author

Owner

@fy2462 commented on GitHub (Jul 28, 2026):

Hi, thank you for your interest in AgentENV!

Currently, AgentENV does not support GPU, since it's built on Firecracker microVMs, which are CPU only by design. Firecracker doesn't expose the PCI or device passthrough needed for GPU access. So yes, at the moment the agent infra is restricted to CPU clusters.

Supporting GPU workloads would require substantial architectural changes, including adding a QEMU backend alongside Firecracker, redesigning our snapshot mechanism, and more. It's a nontrivial undertaking, so it would be a longer term effort rather than something we can do quickly.

Thank you for your patient explanation. It seems that using Kata in GPU scenarios might be more appropriate.

@fy2462 commented on GitHub (Jul 28, 2026): > Hi, thank you for your interest in AgentENV! > > Currently, AgentENV does not support GPU, since it's built on Firecracker microVMs, which are CPU only by design. Firecracker doesn't expose the PCI or device passthrough needed for GPU access. So yes, at the moment the agent infra is restricted to CPU clusters. > > Supporting GPU workloads would require substantial architectural changes, including adding a QEMU backend alongside Firecracker, redesigning our snapshot mechanism, and more. It's a nontrivial undertaking, so it would be a longer term effort rather than something we can do quickly. Thank you for your patient explanation. It seems that using Kata in GPU scenarios might be more appropriate.

bsakaguchi referenced this issue

2026-10-29 14:19:13 +00:00

[GH-ISSUE #262] Add offline workload replay for scheduler strategy evaluation #73

bsakaguchi referenced this issue

2026-10-29 14:19:58 +00:00

[PR #27] [CLOSED] feat(scheduler): add projected least-loaded placement strategy #126

bsakaguchi referenced this issue

2026-10-29 14:20:07 +00:00

[PR #51] [CLOSED] feat(scheduler): add snapshot artifact locality preference #141

bsakaguchi referenced this issue

2026-10-29 14:20:21 +00:00

[PR #68] feat(scheduler): add resource-bounded grouped round-robin scheduling #154