[GH-ISSUE #489] 老哥们,4090太贵了,有没有试过2080ti这种垃圾卡? 多卡下,能跑起来么? ddr4内存管够 #346

Closed
opened 2026-10-29 14:38:17 +00:00 by bsakaguchi · 8 comments
Owner

Originally created by @acodervic on GitHub (Feb 19, 2025).
Original GitHub issue: https://github.com/kvcache-ai/ktransformers/issues/489

Originally created by @acodervic on GitHub (Feb 19, 2025). Original GitHub issue: https://github.com/kvcache-ai/ktransformers/issues/489
Author
Owner

@TTThanos commented on GitHub (Feb 19, 2025):

可以的,理论上2张2080Ti就够了

<!-- gh-comment-id:2668031765 --> @TTThanos commented on GitHub (Feb 19, 2025): 可以的,理论上2张2080Ti就够了
Author
Owner

@yansiyu550 commented on GitHub (Feb 19, 2025):

有14GB显存就行

<!-- gh-comment-id:2668100685 --> @yansiyu550 commented on GitHub (Feb 19, 2025): 有14GB显存就行
Author
Owner

@SongXiaoMao commented on GitHub (Feb 20, 2025):

2080TI要魔改吧 B站有人发视频了 估计稳定了 会官方合并吧

<!-- gh-comment-id:2670297280 --> @SongXiaoMao commented on GitHub (Feb 20, 2025): 2080TI要魔改吧 B站有人发视频了 估计稳定了 会官方合并吧
Author
Owner

@lvyufeng commented on GitHub (Feb 20, 2025):

Image

Image

<!-- gh-comment-id:2670356535 --> @lvyufeng commented on GitHub (Feb 20, 2025): ![Image](https://github.com/user-attachments/assets/5b655671-5b0e-4d3a-9fff-da61dd10ec38) ![Image](https://github.com/user-attachments/assets/7c1e2ffd-da03-4896-b878-45fd9ebe5114)
Author
Owner

@ffhelly commented on GitHub (Feb 20, 2025):

B站有人测试过 魔改4x2080ti的推理。 5tks左右。

<!-- gh-comment-id:2670366381 --> @ffhelly commented on GitHub (Feb 20, 2025): B站有人测试过 魔改4x2080ti的推理。 5tks左右。
Author
Owner

@Azure-Tang commented on GitHub (Feb 20, 2025):

可以,我记得有其它issue跑起来过,可以找一下

<!-- gh-comment-id:2670462708 --> @Azure-Tang commented on GitHub (Feb 20, 2025): 可以,我记得有其它issue跑起来过,可以找一下
Author
Owner

@344303947 commented on GitHub (Mar 6, 2025):

5000 就能6/s了 上2W哈哈。

<!-- gh-comment-id:2703799656 --> @344303947 commented on GitHub (Mar 6, 2025): 5000 就能6/s了 上2W哈哈。
Author
Owner

@lvyufeng commented on GitHub (Mar 6, 2025):

5000 就能6/s了 上2W哈哈。

那是因为这机器早就配了,而且配的原因是要最低成本跑tp➕pp

<!-- gh-comment-id:2703805237 --> @lvyufeng commented on GitHub (Mar 6, 2025): > 5000 就能6/s了 上2W哈哈。 那是因为这机器早就配了,而且配的原因是要最低成本跑tp➕pp
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
kvcache-ai/ktransformers#346
No description provided.