Osaker2025-05-28 17:59:25不是开玩笑...真TM的强. 我的是12GB内存的老旧三星N20U, N年前的机子, 跑最大的那个Gemma 3n E4B int4, 可以读图, 有2token/s, 1st token 才2秒左右. 这个模型已经可以解决生活中的很多问题, 如果读图解答, 离线翻译等等... 太强了吧#老旧设备#性能提升#智能模型
大希沙织2025-01-12 21:42:48这次的NV也是大失所望 还在玩5070(+dlss4)=4090(gaming)这种低级文字游戏，cuda数2w的5090比4090快30%这不是废话吗？功耗也涨了40%，结果是性能倒吸 最关键的是2000刀的显卡居然只给了32G显存，现在二手A100都只要2500了 整个发布会就只有DLSS4能看，还是50系专属，40系是跑不动transformer吗？ #NV#性能提升#DLSS4
Figen2025-01-04 00:35:12There are very good improvements in the algorithm. It was working very badly for 6 months. But now since January 1st the algorithm has been working very well. Great! Has an update been made to the a#算法改进#更新#性能提升
