Unsloth UD-quants - Qwen 3.8 27b for example - worth using 8-bit or stick with faster 6 bit for coding?

For those using these models for coding in larger projects where things can get complex, do you find yourself using the 8-bit quants if you have enough memory? Or do you stick with UD-Q6_K_XL? The 6-bit is faster, noticeably so on my setup. And I keep seeing people say it's imperceptible. I've been

中文全文翻译正在补齐，可先查看下方英文原文。

For those using these models for coding in larger projects where things can get complex, do you find yourself using the 8-bit quants if you have enough memory? Or do you stick with UD-Q6_K_XL? The 6-bit is faster, noticeably so on my setup. And I keep seeing people say it's imperceptible. I've been