all systems operational
Get started
256 GB DDR4 ECC · 6-8 channels

256 GB RAM dedicated servers

Standard on our Xeon Platinum 8168 tier. 256 GB DDR4-2666 ECC across 6 channels — for big databases, ML inference, and virtualization density.

256 GB
DDR4-2666 ECC RDIMM
6 channels
per socket (Platinum)
~128 GB/s
memory bandwidth
from $350
1 Gbit tier

About 256 GB RAM dedicated server

256 GB RAM lets you keep an entire mid-size database in memory (a 200 GB Postgres cluster fits with room for OS cache), run 30-50 mid-size VMs on one host, or hold multi-billion-vector embeddings for ML retrieval workloads.

Our Xeon Platinum 8168 tier ships standard with 256 GB DDR4-2666 ECC across 6 memory channels — that's ~128 GB/s aggregate bandwidth, nearly 2× the E5-2690 v3 platform. AVX-512 on the CPU + high memory bandwidth make this a strong platform for real-time analytics, video transcoding, and quantized ML inference.

Typical workloads: large MySQL/Postgres masters (500 M+ row tables in buffer pool), ClickHouse analytics on multi-TB datasets, Elasticsearch clusters with fat indexes, big Redis caches, dense KVM/LXC hosts, private ML serving stacks (LLM inference with quantized 70B models).

Platform expandable to 1.5 TB total RAM (24 DIMM slots × 64 GB). NVMe Gen3 storage standard with 4 configurable bays. Uplink: 1 / 10 / 20 Gbit unmetered.

/faq

Frequently asked questions

Can I run a 70B LLM in 256 GB RAM?

Yes, quantized. Llama 3 70B at FP16 needs ~140 GB VRAM/RAM, at INT4 quantization ~35-40 GB. On CPU inference (no GPU), 256 GB RAM comfortably serves quantized 70B with reasonable throughput.

Is 256 GB overkill for a WordPress site?

For a single WordPress site — yes. But if you're running 50+ WordPress instances (agency, multi-tenant), 256 GB gives each site meaningful cache headroom.

How does 256 GB perform vs SSD-backed swap?

Vastly better for hot data. Random 4K reads: RAM ~100 ns, NVMe ~50 μs — 500× slower. If your working set fits in RAM, keep it in RAM.

Chat with us@hostfory