Blog
Written versions of our YouTube videos, so you can check the numbers and tables at your own pace.
-
DeepSeek V4.1 on RTX 4090 24GB: 27.9 tok/s, 15.0 with RAM capped at 64GiB (youtube.com)
DeepSeek V4.1 Flash with Strata on one RTX 4090 24GB server: 27.9 tokens/s, and 15.0 when only the usable RAM was capped at 64GiB. Korean multiple choice: 106/150.