# Instance Specification

This page describes the instance families, sizing options, and underlying hardware available for Zenlayer Elastic Compute (ZEC) instances. Use it to choose the right instance shape for your workload and understand the CPU and GPU hardware behind each family.

***

## 1. Instance Families

![ZEC instance families at a glance](/files/cU7u01F1OH5w7PCXooCl)

ZEC offers six instance families, split by CPU vendor and by whether they back standard compute or GPU-attached instances:

| Family | Role         | CPU Platform                                  | CPU Model(s)                    |
| ------ | ------------ | --------------------------------------------- | ------------------------------- |
| z2a    | Standard CPU | AMD EPYC 2nd Gen (Rome)                       | EPYC 7H12                       |
| z4a    | Standard CPU | AMD EPYC 4th Gen (Genoa)                      | EPYC 9654                       |
| za4h   | Standard CPU | AMD EPYC 4th Gen (Genoa)                      | EPYC 9554                       |
| z2i    | Standard CPU | Intel Xeon Scalable 2nd Gen (Cascade Lake)    | Xeon Gold 6254 · Xeon Gold 6248 |
| z3a    | GPU instance | AMD EPYC 3rd Gen (Milan)                      | EPYC 7543                       |
| z4i    | GPU instance | Intel Xeon Scalable 4th Gen (Sapphire Rapids) | Xeon Gold 6430                  |

> **GPU instance** families (z3a, z4i) are deployed exclusively in GPU-enabled instances and are not available as standard compute.
>
> **za4h** is currently available only in the ZEC Mumbai region.

***

## 2. vCPU-to-RAM Ratios

Each family supports three ratio types so you can match your workload's memory profile:

| Ratio Type       | vCPU:RAM | Best For                                                           |
| ---------------- | -------- | ------------------------------------------------------------------ |
| CPU Intensive    | 1:2      | Compute-bound workloads — batch processing, encoding, simulation   |
| Standard         | 1:4      | General-purpose workloads — web servers, application servers, APIs |
| Memory Intensive | 1:8      | Memory-bound workloads — caches, in-memory databases, analytics    |

Instance sizes scale from 1 vCPU up to 64 vCPU within each ratio type. For example, a `z2a.std.16` is a 2nd-gen AMD instance with the 1:4 standard ratio and 16 vCPUs (16 vCPU / 64 GiB RAM).

***

## 3. vNIC and Networking Limits

vNIC capacity scales with instance size. The following limits apply across all instance families:

| vCPU Count | Max vNICs | vNIC Multi-Queue | Private IPv4 per vNIC |
| ---------- | --------- | ---------------- | --------------------- |
| 1–2        | 2         | 1                | 6–10                  |
| 4–8        | 2         | 2–4              | 15                    |
| 16         | 4         | 8                | 30                    |
| 32         | 8         | 16               | 30                    |
| 64         | 16        | 32               | 30                    |

* **vNICs**: Maximum number of virtual network interfaces (including the primary vNIC) you can attach to an instance.
* **vNIC Multi-Queue**: Number of transmit (Tx) and receive (Rx) queues per vNIC. Higher counts improve network throughput and reduce latency.
* **Private IPv4 per vNIC**: Maximum private IPv4 addresses that can be assigned to a single vNIC.

***

## 4. Regional Availability

Available instance sizes vary by location depending on inventory. Check the [Zenlayer console](https://console.zenlayer.com/zec/virtualMachine) when creating an instance to see what's available in your target region.

***

## 5. CPU Specifications

### 5.1 z2a — AMD EPYC 2nd Gen (Rome)

#### AMD EPYC 7H12

| Property            | Value                       |
| ------------------- | --------------------------- |
| Generation          | AMD EPYC 2nd Gen (Rome)     |
| Microarchitecture   | Zen 2                       |
| Socket              | SP3                         |
| Cores / Threads     | 64 / 128                    |
| Base Clock          | 2.60 GHz                    |
| Max Boost Clock     | 3.30 GHz                    |
| L2 Cache            | 32 MB (512 KB × 64)         |
| L3 Cache            | 256 MB                      |
| Memory Controller   | 8-channel DDR4              |
| Max Memory Speed    | DDR4-3200                   |
| Max Memory Capacity | 2 TB (4 TB with 3DS RDIMMs) |
| PCIe Version        | PCIe 4.0                    |
| PCIe Lanes          | 128 per socket              |
| TDP                 | 280 W                       |
| Process Node        | TSMC 7 nm (N7)              |

***

### 5.2 z3a — AMD EPYC 3rd Gen (Milan)

#### AMD EPYC 7543 *(GPU instances only)*

| Property            | Value                       |
| ------------------- | --------------------------- |
| Generation          | AMD EPYC 3rd Gen (Milan)    |
| Microarchitecture   | Zen 3                       |
| Socket              | SP3                         |
| Cores / Threads     | 32 / 64                     |
| Base Clock          | 2.80 GHz                    |
| Max Boost Clock     | 3.70 GHz                    |
| L2 Cache            | 16 MB (512 KB × 32)         |
| L3 Cache            | 128 MB                      |
| Memory Controller   | 8-channel DDR4              |
| Max Memory Speed    | DDR4-3200                   |
| Max Memory Capacity | 2 TB (4 TB with 3DS RDIMMs) |
| PCIe Version        | PCIe 4.0                    |
| PCIe Lanes          | 128 per socket              |
| TDP                 | 225 W                       |
| Process Node        | TSMC 7 nm (N7P)             |

***

### 5.3 z4a — AMD EPYC 4th Gen (Genoa)

#### AMD EPYC 9654

| Property            | Value                    |
| ------------------- | ------------------------ |
| Generation          | AMD EPYC 4th Gen (Genoa) |
| Microarchitecture   | Zen 4                    |
| Socket              | SP5                      |
| Cores / Threads     | 96 / 192                 |
| Base Clock          | 2.40 GHz                 |
| Max Boost Clock     | 3.70 GHz                 |
| L2 Cache            | 96 MB (1 MB × 96)        |
| L3 Cache            | 384 MB                   |
| Memory Controller   | 12-channel DDR5          |
| Max Memory Speed    | DDR5-4800                |
| Max Memory Capacity | 6 TB                     |
| PCIe Version        | PCIe 5.0                 |
| PCIe Lanes          | 128 per socket           |
| TDP                 | 360 W                    |
| Process Node        | TSMC 5 nm (N5)           |

***

### 5.4 za4h — AMD EPYC 4th Gen (Genoa)

#### AMD EPYC 9554 *(ZEC Mumbai only)*

| Property            | Value                    |
| ------------------- | ------------------------ |
| Generation          | AMD EPYC 4th Gen (Genoa) |
| Microarchitecture   | Zen 4                    |
| Socket              | SP5                      |
| Cores / Threads     | 64 / 128                 |
| Base Clock          | 3.10 GHz                 |
| Max Boost Clock     | 3.75 GHz                 |
| L2 Cache            | 64 MB (1 MB × 64)        |
| L3 Cache            | 256 MB                   |
| Memory Controller   | 12-channel DDR5          |
| Max Memory Speed    | DDR5-4800                |
| Max Memory Capacity | 6 TB                     |
| PCIe Version        | PCIe 5.0                 |
| PCIe Lanes          | 128 per socket           |
| TDP                 | 360 W                    |
| Process Node        | TSMC 5 nm (N5)           |

***

### 5.5 z2i — Intel Xeon Scalable 2nd Gen (Cascade Lake)

#### Intel Xeon Gold 6254

| Property            | Value                                      |
| ------------------- | ------------------------------------------ |
| Generation          | 2nd Gen Intel Xeon Scalable (Cascade Lake) |
| Microarchitecture   | Cascade Lake                               |
| Socket              | LGA 3647 (P0)                              |
| Cores / Threads     | 18 / 36                                    |
| Base Clock          | 3.10 GHz                                   |
| Max Turbo Frequency | 4.00 GHz                                   |
| L2 Cache            | 18 MB                                      |
| L3 Cache (LLC)      | 24.75 MB                                   |
| Memory Controller   | 6-channel DDR4                             |
| Max Memory Speed    | DDR4-2933                                  |
| Max Memory Capacity | 1 TB per socket                            |
| PCIe Version        | PCIe 3.0                                   |
| PCIe Lanes          | 48 per socket                              |
| TDP                 | 200 W                                      |
| Process Node        | Intel 14 nm++                              |

#### Intel Xeon Gold 6248

| Property            | Value                                      |
| ------------------- | ------------------------------------------ |
| Generation          | 2nd Gen Intel Xeon Scalable (Cascade Lake) |
| Microarchitecture   | Cascade Lake                               |
| Socket              | LGA 3647 (P0)                              |
| Cores / Threads     | 20 / 40                                    |
| Base Clock          | 2.50 GHz                                   |
| Max Turbo Frequency | 3.90 GHz                                   |
| L2 Cache            | 20 MB                                      |
| L3 Cache (LLC)      | 27.50 MB                                   |
| Memory Controller   | 6-channel DDR4                             |
| Max Memory Speed    | DDR4-2933                                  |
| Max Memory Capacity | 1 TB per socket                            |
| PCIe Version        | PCIe 3.0                                   |
| PCIe Lanes          | 48 per socket                              |
| TDP                 | 150 W                                      |
| Process Node        | Intel 14 nm++                              |

***

### 5.6 z4i — Intel Xeon Scalable 4th Gen (Sapphire Rapids)

#### Intel Xeon Gold 6430 *(GPU instances only)*

| Property            | Value                                         |
| ------------------- | --------------------------------------------- |
| Generation          | 4th Gen Intel Xeon Scalable (Sapphire Rapids) |
| Microarchitecture   | Sapphire Rapids                               |
| Socket              | LGA 4677                                      |
| Cores / Threads     | 32 / 64                                       |
| Base Clock          | 2.10 GHz                                      |
| Max Turbo Frequency | 3.40 GHz                                      |
| L2 Cache            | 40 MB                                         |
| L3 Cache (LLC)      | 60 MB                                         |
| Memory Controller   | 8-channel DDR5                                |
| Max Memory Speed    | DDR5-4800                                     |
| Max Memory Capacity | 4 TB per socket                               |
| PCIe Version        | PCIe 5.0                                      |
| PCIe Lanes          | 80 per socket                                 |
| TDP                 | 270 W                                         |
| Process Node        | Intel 7 (10 nm ESF)                           |

***

## 6. GPU Specifications

### NVIDIA GeForce RTX 4090

| Property                | Value                                            |
| ----------------------- | ------------------------------------------------ |
| Architecture            | Ada Lovelace                                     |
| GPU Die                 | AD102                                            |
| CUDA Cores              | 16,384                                           |
| Tensor Cores            | 512 (4th Gen)                                    |
| RT Cores                | 128 (3rd Gen)                                    |
| GPU Memory              | 24 GB GDDR6X                                     |
| Memory Bus Width        | 384-bit                                          |
| Memory Bandwidth        | 1,008 GB/s                                       |
| Memory Speed            | 21 Gbps                                          |
| L2 Cache                | 72 MB                                            |
| FP32 Performance        | 82.6 TFLOPS                                      |
| FP16 Performance        | 165.2 TFLOPS (non-sparse)                        |
| TF32 Tensor Performance | 82.6 TFLOPS (non-sparse) / 330.3 TFLOPS (sparse) |
| Boost Clock             | 2,520 MHz                                        |
| Interface               | PCIe 4.0 x16                                     |
| TDP                     | 450 W                                            |
| Process Node            | TSMC 4 nm (N4)                                   |
| Launch Date             | October 2022                                     |

### NVIDIA Tesla T4

| Property         | Value                     |
| ---------------- | ------------------------- |
| Architecture     | Turing                    |
| GPU Die          | TU104                     |
| CUDA Cores       | 2,560                     |
| Tensor Cores     | 320 (2nd Gen)             |
| GPU Memory       | 16 GB GDDR6               |
| Memory Bandwidth | 320 GB/s                  |
| FP32 Performance | 8.1 TFLOPS                |
| FP16 Performance | 65 TFLOPS (with sparsity) |
| INT8 Performance | 130 TOPS                  |
| INT4 Performance | 260 TOPS                  |
| ECC              | Supported                 |
| Interface        | PCIe 3.0 x16              |
| Form Factor      | Low-profile, single-slot  |
| TDP              | 70 W                      |
| Process Node     | TSMC 12 nm                |
| Launch Date      | September 2018            |


---

# Agent Instructions: Querying This Documentation

If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter:

```
GET https://docs.console.zenlayer.com/welcome/elastic-compute/instance-specification.md?ask=<question>
```

The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.