Kairntech platform – no GPU
All engines and vectorizers are supported, with the exception of Transformers, BGE-M3 vectorizer…
| Component | Specifications |
|---|---|
| CPU | At least 8 cores, preferably more Base clock speed ≥ 3.0 GHz Single Thread Rating (STR) Index ≥ 2200 High core count is prioritized over single-thread speed |
| RAM | 64 GB |
| Disk | 400 GB SSD with high read IOPS (≥ 10,000) |
Refer to CPU Benchmark for reference.
Example:
- CPU: AMD Ryzen 7 7700X (8 cores / 16 threads, STR index: 4185, clock speed: 5.4 GHz)
- RAM: 64 GB
- SSD: Samsung 980 Pro 500 GB (PCIe 4.0 NVMe, high IOPS: 690 000 read / 620 000 write)
Kairntech platform – with GPU (recommended)
All engines and vectorizers supported. Training and inference times can be significantly reduced.
| Component | Specifications |
|---|---|
| CPU | At least 8 cores, preferably more Base clock speed ≥ 3.0 GHz Single Thread Rating (STR) Index ≥ 2200 High core count is prioritized over single-thread speed |
| GPU / VRAM | ≥ 8 GB VRAM CUDA compute capability ≥ 7.5 |
| RAM | 64 GB, 128 GB recommended |
| Disk | 400 GB SSD with high read IOPS (≥ 10,000) |
Refer to CPU Benchmark for reference.
Example:
- CPU: AMD Ryzen 7 9800X3D (8 cores / 16 threads, STR index: 4427, clock speed: 5.2 GHz)
- GPU/VRAM: NVIDIA RTX 3060 (12 GB VRAM)
- RAM: 128 GB
- SSD: Samsung 980 Pro 500 GB (PCIe 4.0 NVMe, high IOPS: 690 000 read / 620 000 write)
Local LLM inference server (GPU required)
The goal is to run an LLM locally to ensure data privacy, maintain full control, and reduce long-term costs by moving AI operations from the cloud to your own hardware.
| Component | Specifications |
|---|---|
| CPU | At least 16 cores, preferably more Base clock speed ≥ 3.0 GHz Single Thread Rating (STR) Index ≥ 2000 |
| GPU / VRAM | ≥ 48 GB VRAM (16 GB VRAM with quantized models) to run GPT-OSS-20B, Mistral 24B… |
| RAM | 128 GB |
| Disk | 2 TB NVMe SSD (PCIe Gen4) |
Refer to CPU Benchmark for reference.
Example:
- CPU: AMD Threadripper 7970X (32 cores / 64 threads, STR index: 4170, clock speed: 5.3 GHz)
- GPU/VRAM: NVIDIA RTX 6000 ada (48 GB VRAM)
- RAM: 128 GB
- SSD: Samsung 980 Pro 2 TB (PCIe 4.0 NVMe, high IOPS: 690 000 read / 660 000 write)
Entity-fishing server (no GPU required)
The entity-fishing server aims to automatically annotate content using Wikidata and the Wikipedia knowledge base. It is mainly combined with a NER model within a pipeline.
| Component | Specifications |
|---|---|
| CPU | 8 physical cores, CPUs with high sustained all-core turbo frequencies, ideally > 3 GHz |
| RAM | 128 GB recommended |
| Disk | 400 GB SSD with high read IOPS (≥ 10,000) |
Refer to CPU Benchmark for reference.
Example:
- CPU: AMD Ryzen 7 5700G (8 cores / 16 threads, STP index: 3273, clock speed: 4.6 GHz)
- RAM: 128 GB
- SSD: Samsung 980 Pro 1 TB (PCIe 4.0 NVMe, high IOPS: 690 000 read / 660 000 write)