Instance Tiers

GPU instance types, specifications, and pricing

Instance Tiers

Choose the right GPU instance for your model size and performance needs.

Tier Overview

Tier	GPU	VRAM	Price	Best For
XS	1x L4	24GB	~$0.20/h	7B-13B models
S	1x L40S	48GB	~$0.60/h	13B-34B models
M	4x A10G	96GB	~$1.80/h	30B-70B INT4
L	4x L40S	192GB	~$3.50/h	70B FP16
XL	8x A100	320-640GB	~$12/h	70B-180B
XXL	8x H100/H200	640-1128GB	~$20/h	405B

---

Tier XS - Entry Level

Best for small models (7B-13B parameters).

GPU XS

Spec	Value
GPU	1x NVIDIA L4
VRAM	24GB
vCPUs	4
RAM	16GB
Storage	250GB NVMe
Network	10 Gbps
Price	~$0.20/h

Recommended models: LLaMA 3.2 1B, Gemma 3 4B, Mistral 7B, LLaMA 3.1 8B

GPU XS (2x)

Spec	Value
GPU	1x NVIDIA L4
VRAM	24GB
vCPUs	8
RAM	32GB
Storage	450GB NVMe
Network	15 Gbps
Price	~$0.35/h

---

Tier S - Small Production

Best for medium models (13B-34B parameters).

GPU S

Spec	Value
GPU	1x NVIDIA L40S
VRAM	48GB
vCPUs	4
RAM	16GB
Storage	250GB NVMe
Network	10 Gbps
Price	~$0.60/h

Recommended models: LLaMA 3.1 13B, CodeLlama 13B, Llama 4 Scout 17B

GPU S (2x)

Spec	Value
GPU	1x NVIDIA L40S
VRAM	48GB
vCPUs	8
RAM	32GB
Storage	450GB NVMe
Network	15 Gbps
Price	~$1.00/h

Recommended models: CodeLlama 34B, Mixtral 8x7B, Qwen3 30B

---

Tier M - Medium Production

Best for large quantized models (30B-70B INT4).

GPU M

Spec	Value
GPU	4x NVIDIA A10G
VRAM	96GB
vCPUs	48
RAM	192GB
Storage	3.8TB NVMe
Network	40 Gbps
Price	~$1.80/h

Recommended models: LLaMA 3.1 70B (AWQ), Qwen2 72B (AWQ), DeepSeek 67B (AWQ)

GPU M (2x)

Spec	Value
GPU	4x NVIDIA A10G
VRAM	96GB
vCPUs	96
RAM	384GB
Storage	3.8TB NVMe
Network	50 Gbps
Price	~$3.00/h

GPU M (4x)

Spec	Value
GPU	8x NVIDIA A10G
VRAM	192GB
vCPUs	192
RAM	768GB
Storage	7.6TB NVMe
Network	100 Gbps
Price	~$5.00/h

---

Tier L - Large Production

Best for full-precision large models (70B FP16).

GPU L

Spec	Value
GPU	4x NVIDIA L40S
VRAM	192GB
vCPUs	48
RAM	384GB
Storage	3.8TB NVMe
Network	40 Gbps
Price	~$3.50/h

Recommended models: LLaMA 3.1 70B, Qwen2 72B, DeepSeek V3 70B

GPU L (2x)

Spec	Value
GPU	4x NVIDIA L40S
VRAM	192GB
vCPUs	96
RAM	768GB
Storage	3.8TB NVMe
Network	50 Gbps
Price	~$6.00/h

Recommended models: Mixtral 8x22B

GPU L (4x)

Spec	Value
GPU	8x NVIDIA L40S
VRAM	384GB
vCPUs	192
RAM	1536GB
Storage	7.6TB NVMe
Network	100 Gbps
Price	~$10.00/h

---

Tier XL - Enterprise

Best for very large models (70B-180B).

GPU XL

Spec	Value
GPU	8x NVIDIA A100 (40GB)
VRAM	320GB
vCPUs	96
RAM	1152GB
Storage	8TB NVMe
Network	400 Gbps EFA
Price	~$12.00/h

Recommended models: Falcon 180B, DeepSeek R1 180B

GPU XL (80GB)

Spec	Value
GPU	8x NVIDIA A100 (80GB)
VRAM	640GB
vCPUs	96
RAM	1152GB
Storage	8TB NVMe
Network	400 Gbps EFA
Price	~$18.00/h

---

Tier XXL - Colossal

Best for the largest models (405B).

GPU XXL

Spec	Value
GPU	8x NVIDIA H100 (80GB)
VRAM	640GB
vCPUs	192
RAM	2048GB
Storage	8TB NVMe
Network	3200 Gbps EFA v2
Price	~$20.00/h

Recommended models: LLaMA 3.1 405B (FP8), DBRX 132B

GPU XXL (H200)

Spec	Value
GPU	8x NVIDIA H200 (141GB)
VRAM	1128GB
vCPUs	192
RAM	2048GB
Storage	8TB NVMe
Network	3200 Gbps EFA v2
Price	~$30.00/h

Recommended models: LLaMA 3.1 405B (FP16)

---

Choosing the Right Tier

By Model Size

Model Parameters	Precision	Recommended Tier
1B - 8B	FP16	XS
7B - 13B	INT4	XS
13B - 30B	FP16	S
30B - 34B	INT4/FP16	S / S-2x
70B	INT4/AWQ	M
70B	FP16	L
70B - 180B	FP16	XL
405B	FP8	XXL
405B	FP16	XXL-H200

By Use Case

Use Case	Recommended Tier
Development/Testing	XS
Small production	S
Cost-optimized production	M (quantized)
High-quality production	L
Enterprise/Maximum quality	XL / XXL