# Qwen 3.8 on 3090 comparisons — retained record

> Retained reddit record from reddit:LocalLLaMA.

Canonical: https://highsignal.app/data/records/4d36e38406625b85

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:LocalLLaMA)  published 2026-08-31 17:56 ingested 2026-09-01 02:39

[original source ↗](https://www.reddit.com/r/LocalLLaMA/comments/1w3lg41/qwen_38_on_3090_comparisons/)

## retained text



```
I’m getting \~35 tok/s with Qwen3.8-27B on my RTX 3090.

Another setup gets \~65 tok/s on the same GPU.



Is mine slow?



Mine:

• Q5 weights  
• Q8 KV cache  
• 32K context  
• 4 parallel slots  
• MTP off  



The faster setup:

• Q4 weights  
• Q4 KV cache  
• 131K context  
• 1 slot  
• MTP on    
  
The trade-offs:  
Q5 vs Q4: Different quantization precision  
  
Q8 vs Q4 KV: cache precision versus context capacity  
  
4 slots vs 1: concurrency versus single-user speed  
  
MTP (Multi-token prediction) on vs off: speculative speed versus memory and verification work  
  
Is there anything I'm missing? Any other variables that I should consider? 

For context, I'm using Llama.cpp 
```



document fetched 2026-09-01 00:41 · canonical https://reddit.com/r/LocalLLaMA/comments/1w3lg41/qwen_38_on_3090_comparisons
