Enabling GPU Support

This guide explains how to enable GPU support on your Compute Resource Node (CRN), allowing users to deploy instances with GPU capabilities. The process involves two main aspects:

1. Understanding the available GPU options and pricing
2. Configuring your CRN for GPU support

Prerequisites

Before enabling GPU support:

• Enable PAYG support and IPv6 on your CRN (this is required for GPU billing)
• Follow the steps at Enable PAYG

Available GPUs and Pricing

GPU instances are billed in Credits, where 1 credit corresponds to 10⁻⁶ USD/hour of resource cost. The figures below are shown in USD/hour for readability. Pricing is based on Compute Units (CU), where 1 GPU CU = 1 vCPU + 6 GB RAM + 60 GB disk. Each GPU model requires a fixed number of CUs, which determines both the resources allocated to the instance and its hourly cost.

GPU group	Rate per CU
Standard	$0.043125/hour
Premium	$0.08625/hour

The following GPUs are currently compatible with Aleph Cloud. More options will be added as they are tested and validated.

Standard GPUs

Consumer and workstation-grade GPUs suitable for most workloads:

GPU Model	vRAM	CUs	vCPU	RAM	Price (USD/hour)
RTX 4000 ADA	20 GB	3	3	18 GB	$0.1294
RTX A5000	24 GB	3	3	18 GB	$0.1294
RTX 3090	24 GB	4	4	24 GB	$0.1725
RTX A6000	48 GB	4	4	24 GB	$0.1725
RTX 4090	24 GB	6	6	36 GB	$0.2588
RTX 5090	32 GB	8	8	48 GB	$0.3450
RTX 6000 ADA	48 GB	11	11	66 GB	$0.4744
L40S	48 GB	12	12	72 GB	$0.5175

Premium GPUs

Datacenter-grade GPUs optimized for high-performance computing:

GPU Model	vRAM	CUs	vCPU	RAM	Price (USD/hour)
RTX PRO 6000	96 GB	14	14	84 GB	$1.2075
A100	80 GB	16	16	96 GB	$1.3800
H100	80 GB	24	24	144 GB	$2.0700
H200	144 GB	32	32	192 GB	$2.7600

Hardware Requirements

All GPUs must be connected via PCIe 4.0 16x each for optimal performance.

GPU Configuration Guide

Enabling Kernel Modules

TIP

It is possible to enable multiple GPUs on one CRN. The following steps will help you configure the required kernel modules.

Configuration Steps

WARNING

Execute these commands as root. On Ubuntu systems, use sudo.

1. Edit initramfs to attach GPU to vfio: vi /etc/initramfs-tools/modules and set the content to

attach : vfio vfio_iommu_type1 vfio_virqfd vfio_pci ids=10de:27b0,10de:22bc

replacing the id's 10de:27b0,10de:22bc with id's of the VGA card, there should be 2, the Video device and the GPU audio device. You can get them running lspci -nvv as root user.

2. Edit /etc/modules to ensure that lods GPU to vfio: vi /etc/modules and add

attach: vfio vfio_iommu_type1 vfio_pci ids=10de:27b0,10de:22bc

3. Modify nvidia drivers to load after the vfio: vi /etc/modprobe.d/nvidia.conf

set

softdep nouveau pre: vfio-pci
softdep nvidia pre: vfio-pci
softdep nvidia* pre: vfio-pci

4. Get the know alias from the PCI device: cat /sys/bus/pci/devices/0000:01:00.0/modalias replace pci address 0000:01:00.0 with the one for your GPU, you can get it using lspci

5. Configure vfio module to use that devices: vi /etc/modprobe.d/vfio.conf

blacklist nouveau
blacklist snd_hda_intel
alias pci:v000010DEd000027B0sv000010DEsd000016FAbc03sc00i00 vfio-pci
options vfio-pci ids=10de:27b0,10de:22bc

6. Enable modprobe vfio-pci

modprobe vfio-pci

7 . Update initramfs image with your changes:

update-initramfs -u -k all

8. Finally Reboot the server

reboot

9. Confirm that the vfio-pci kernel module is used for the card

lspci -k

Should display something similar to

[...]
01:00.0 VGA compatible controller: NVIDIA Corporation AD104GL [RTX 4000 SFF Ada Generation] (rev a1)
	Subsystem: NVIDIA Corporation AD104GL [RTX 4000 SFF Ada Generation]
	Kernel driver in use: vfio-pci
	Kernel modules: nvidiafb, nouveau, nvidia_drm, nvidia
01:00.1 Audio device: NVIDIA Corporation Device 22bc (rev a1)
	Subsystem: NVIDIA Corporation Device 16fa
	Kernel driver in use: vfio-pci
	Kernel modules: snd_hda_intel
[...]

10. Enable GPU support in Aleph-VM In the CRN configuration /etc/aleph-vm/supervisor.env, enable the GPU support

ALEPH_VM_ENABLE_GPU_SUPPORT=True

Don't forget to also set ALEPH_VM_PAYMENT_RECEIVER_ADDRESS and enable PAYG.

11. Confirm that the GPU are listed and supported on the CRN index page.

Start or restart your aleph-vm supervisor, open the index page, and it should list your GPU

GPUs

   • NVIDIA | AD104GL [RTX 4000 SFF Ada Generation] is compatible ✅