Run Gemma 4 on a VPS: sizing and speed
Which Gemma 4 tag fits your VPS RAM, what each one costs in tokens per second on CPU, and the sizing arithmetic to check it before you pull 20GB.
Filtering by topic #quantization · clear
Which Gemma 4 tag fits your VPS RAM, what each one costs in tokens per second on CPU, and the sizing arithmetic to check it before you pull 20GB.
Community Ollama tags still run DeepSeek V4 Flash locally. Get the real GGUF disk and RAM numbers, and see which VPS can actually load the model.
GLM 5.2 is cloud only in Ollama's library. Here is the GLM model that actually fits a VPS, and the RAM each quantisation needs on your own box.
Qwen 3.8 27B is now on Ollama as qwen3.8:27b. Here is the RAM and speed arithmetic for running it on a CPU only VPS, and what fits in 8 to 64 GB.
Raised num_ctx and Ollama no longer fits? The KV cache grew, not the weights. Set OLLAMA_KV_CACHE_TYPE with flash attention and watch the change in ollama ps.
Pick an Ollama quantization with arithmetic instead of guesswork: what q4_K_M, q8_0 and fp16 cost in RAM, and where the quality actually drops.
Size an LLM for a CPU server without guessing: weights in bytes, KV cache growth with context, runtime overhead, and why fitting is not the same as usable.