AI

Xolotl (Strata based) - DeepSeek-V4.1-Flash (552B) on a single 4090 at ~27 tok/s, Uncensored 1017 quant coming in a couple of days

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 18/10/2026, 08:47:26 UTC.

Briefings

A user named Indras-Mirror built an inference engine called Xolotl to run DeepSeek-V4.1-Flash on a single RTX 4090, achieving 27 tokens per second decode and 180-320 tokens per second prefill with 512K context. The setup utilizes 90GB of DDR4 RAM and an NVMe drive for weight storage, running multiple models including GLM-5.3-Flash and Qwen3.8-Flash-Next. The 1017 quantization update is currently being downloaded and is expected to be installed within a couple of days.

Key points

• The inference engine Xolotl successfully runs DeepSeek-V4.1-Flash at 27 tokens per second decode on a single RTX 4090 with 24GB VRAM.
• The system achieves 180-320 tokens per second prefill with a 512K context window using 90GB of DDR4 RAM and an NVMe drive.

Context

This finding demonstrates the feasibility of running large language models like DeepSeek-V4.1-Flash on consumer-grade hardware, specifically a single RTX 4090, with high throughput and context handling capabilities.

Limitations

The evidence is limited to a single user's setup and does not represent a general benchmark. The 1017 quantization update is currently being downloaded and is not yet installed, so its performance impact is unknown. The user's internet connection is described as slow, which may affect download speed but not necessarily the final model's performance.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 18/10/2026, 17:23:43 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeia5jwtexkjvedkdck3w3ek2mrm4waj7zwrr5lr42bjwlnafuh2m5e