GPU cloud in Jakarta: NVIDIA RTX PRO 6000 Blackwell and L40S

IONext rents NVIDIA RTX PRO 6000 Blackwell Server Edition 96GB and NVIDIA L40S 48GB GPUs located in Jakarta, Indonesia. GPUs are available as dedicated GPUs, GPU virtual machines, or through managed inference with an OpenAI-compatible API.

Which GPUs does IONext offer?

Spec sources: NVIDIA's official product pages for the RTX PRO 6000 Blackwell Server Edition and L40S.

	NVIDIA RTX PRO 6000 Blackwell Server Edition	NVIDIA L40S
Architecture	NVIDIA Blackwell	NVIDIA Ada Lovelace
GPU memory	96 GB GDDR7	48 GB GDDR6 with ECC
Memory bandwidth	1,597 GB/s	864 GB/s
Interface	PCIe Gen5	PCIe Gen4 x16
Multi-Instance GPU (MIG)	Yes, up to 4 isolated instances	No
Max power	Up to 600 W	350 W
Video encode and decode	4 NVENC, 4 NVDEC	3 NVENC, 3 NVDEC (including AV1)

How large a model fits on each GPU?

GPU memory for model weights is roughly the parameter count times the size per parameter: 2 bytes at FP16/BF16, 1 byte at FP8, and about 0.5 bytes with 4-bit quantization. Leave extra room for the KV cache, which grows with context length and concurrent requests.

Model size	FP16 weights	FP8 weights	1x RTX PRO 6000 (96 GB)	1x L40S (48 GB)
8B	about 16 GB	about 8 GB	Fits	Fits
31B (for example Gemma4-31B)	about 62 GB	about 31 GB	Fits at FP16	Fits at FP8
27B (for example Qwen3.8-27B)	about 54 GB	about 27 GB	Fits at FP16	Fits at FP8
70B	about 140 GB	about 70 GB	Fits at FP8	Needs 2 GPUs at FP8
Over 200B (for example DeepSeek-V4-Flash, 304B)	over 400 GB	over 200 GB	Needs several GPUs	Needs several GPUs

This table estimates weight memory only; it is not a speed measure. For sizing against a real workload, IONA inside ORBIT recommends the VM size and model for the use case.

When should you choose RTX PRO 6000 Blackwell, and when L40S?

Choose the RTX PRO 6000 Blackwell for models above 30B parameters at full precision, long context, fine-tuning, or when one GPU needs to be split into isolated workloads with MIG. Its 96 GB of memory reduces the need to split a model across GPUs.

Choose the L40S for models up to about 30B parameters with quantization, computer vision, video processing (AV1 encode and decode), and inference whose memory needs fit in 48 GB.

How do you rent a GPU from IONext?

1. Dedicated GPU. One or more GPUs allocated to a single customer, suited to steady production workloads and fine-tuning.
2. GPU virtual machine. A VM with an attached GPU, created and monitored in ORBIT alongside other VMs, storage and pipelines.
3. Managed inference. No GPU to manage at all: call open-weight models through one OpenAI-compatible endpoint.

Pricing is not published and is scoped to each requirement. All billing is in Indonesian Rupiah. Request a quote or a sizing session. System integrators sizing GPUs for a client project can work with IONext through the IONext Partner Program.

What infrastructure supports the GPUs?

• Storage: a Ceph cluster with about 200 TB usable capacity (3x replication), plus NVMe storage nodes. Available as S3-compatible object storage and block storage, all in Indonesia.
• Network: 25 GbE and 100 GbE data center fabric, Virtual Private Cloud, managed load balancer.
• Security: Web Application Firewall, anti-DDoS, audit trail, backup and disaster recovery. Certification details on the Security and Compliance page.
• Orchestration: virtual machines and managed Kubernetes.

Frequently asked questions

Where are IONext GPUs located? In Jakarta, Indonesia. Customer data and workloads are processed on this infrastructure and are not sent abroad.

Can I rent a dedicated GPU? Yes. IONext provides dedicated RTX PRO 6000 Blackwell and L40S GPUs to a single customer.

Is IONext GPU pricing published? No. Quotes are scoped by GPU type, quantity and duration, billed in Indonesian Rupiah.

Does IONext support MIG? The RTX PRO 6000 Blackwell Server Edition supports Multi-Instance GPU with up to 4 isolated instances. The L40S does not support MIG.

Company profile and leadership are on the About IONext page.