$英伟达(NVDA)_老虎社区_美港股上老虎

赌狗终将一无所有

01-27

$英伟达(NVDA)$ 底层逻辑未变！

1. 人类是不是要全面进入ai时代？是！更是了！

2. 英伟达芯片牛不牛？太牛了！

3. 人类对ai芯片需求大不大？大！目前除了llm，其它场景基本真空！

4. deepseek省了多少倍芯片？准确数据无从得知，r1为纯文本模型，训练量仍为v3三倍，可见scaling law依然坚强。Alexander wang称deepseek拥有5万张h100，如果接近这个数量那这并不是个小数目。

免责声明：上述内容仅代表发帖人个人观点，不构成本平台的任何投资建议。

精彩评论

KEINS
01-27
KEINS
Dep的芯片是不多，但是他训练的结果很烂，还得加大芯片的用量才行。所以他现在说吹牛说自己用少量的芯片训练的更好，这简直就是骗人。我刚才实际体验了地真的很难用，给的答案都不准确。说明他的芯片用少了，用的低端了还是不行，还得用高端芯片，还得加大芯片用量。
- 赌狗终将一无所有:
  我也试了，我出了一个工程题，deepseek算错两个小数点，chatgpt算对了。虽然说我们自己这种测试意义也不太大，但哪怕是对比deepseek自己v3和r1，训练量也是大幅上升。
穷不达
01-27
穷不达
对。而且日后如果需要将llm和视频图像进行整合。大模型所需的更复杂gpu场景英伟达家的硬件必定还是硬通货。而且小公司通过这波降本增效即使大公司不再需要更多硬件但是小公司仍然需要还会吸引更多玩家入场
wuxxx
01-27
wuxxx
而且为了AGI这些多模态模型一样会吃算力，现在deepseek好像还不可以识别图像

发表看法

{"i18n":{"language":"zh_CN"},"detailType":1,"isChannel":false,"data":{"magic":2,"id":397125189472936,"tweetId":"397125189472936","gmtCreate":1737977027907,"gmtModify":1737977030162,"author":{"id":3516069222588608,"idStr":"3516069222588608","authorId":3516069222588608,"authorIdStr":"3516069222588608","name":"赌狗终将一无所有","avatar":"https://static.tigerbbs.com/5dea5b60703e4a675b801d078999d34d","vip":1,"userType":1,"introduction":"","boolIsFan":false,"boolIsHead":false,"crmLevel":2,"crmLevelSwitch":0,"individualDisplayBadges":[],"fanSize":518,"starInvestorFlag":false},"themes":[],"images":[],"coverImages":[],"html":"<html><head></head><body><a href=\"https://laohu8.com/S/NVDA\">$英伟达(NVDA)$&nbsp;</a> 底层逻辑未变！\n1. 人类是不是要全面进入ai时代？是！更是了！\n2. 英伟达芯片牛不牛？太牛了！\n3. 人类对ai芯片需求大不大？大！目前除了llm，其它场景基本真空！\n4. deepseek省了多少倍芯片？准确数据无从得知，r1为纯文本模型，训练量仍为v3三倍，可见scaling law依然坚强。Alexander wang称deepseek拥有5万张h100，如果接近这个数量 那这并不是个小数目。</body></html>","htmlText":"<html><head></head><body><a href=\"https://laohu8.com/S/NVDA\">$英伟达(NVDA)$&nbsp;</a> 底层逻辑未变！\n1. 人类是不是要全面进入ai时代？是！更是了！\n2. 英伟达芯片牛不牛？太牛了！\n3. 人类对ai芯片需求大不大？大！目前除了llm，其它场景基本真空！\n4. deepseek省了多少倍芯片？准确数据无从得知，r1为纯文本模型，训练量仍为v3三倍，可见scaling law依然坚强。Alexander wang称deepseek拥有5万张h100，如果接近这个数量 那这并不是个小数目。</body></html>","text":"$英伟达(NVDA)$ 底层逻辑未变！ 1. 人类是不是要全面进入ai时代？是！更是了！ 2. 英伟达芯片牛不牛？太牛了！ 3. 人类对ai芯片需求大不大？大！目前除了llm，其它场景基本真空！ 4. deepseek省了多少倍芯片？准确数据无从得知，r1为纯文本模型，训练量仍为v3三倍，可见scaling law依然坚强。Alexander wang称deepseek拥有5万张h100，如果接近这个数量 那这并不是个小数目。","highlighted":1,"essential":1,"paper":1,"likeSize":1,"commentSize":4,"repostSize":0,"favoriteSize":0,"link":"https://laohu8.com/post/397125189472936","repostId":0,"isVote":1,"tweetType":1,"viewCount":3342,"commentLimit":10,"likeStatus":false,"favoriteStatus":false,"reportStatus":false,"symbols":["NVDA"],"verified":2,"subType":0,"readableState":1,"langContent":"CN","currentLanguage":"CN","warmUpFlag":false,"orderFlag":false,"shareable":true,"causeOfNotShareable":"","featuresForAnalytics":[],"commentAndTweetFlag":false,"andRepostAutoSelectedFlag":false,"upFlag":false,"length":318,"optionInvolvedFlag":false,"xxTargetLangEnum":"ZH_CN"},"commentList":[{"id":397126960030304,"commentId":"397126960030304","gmtCreate":1737977230160,"gmtModify":1737977233990,"authorId":3440540550401156,"author":{"id":3440540550401156,"idStr":"3440540550401156","authorId":3440540550401156,"name":"KEINS","avatar":"https://static.tigerbbs.com/70c84e27d76e3a3450fb952380b9db3e","vip":1,"crmLevel":2,"crmLevelSwitch":0,"individualDisplayBadges":[]},"repliedAuthorId":0,"objectId":397125189472936,"objectIdStr":"397125189472936","type":1,"supId":0,"supIdStr":"0","prevId":0,"prevIdStr":"0","content":"Dep的芯片是不多，但是他训练的结果很烂，还得加大芯片的用量才行。所以他现在说吹牛说自己用少量的芯片训练的更好，这简直就是骗人。我刚才实际体验了地真的很难用，给的答案都不准确。说明他的芯片用少了，用的低端了还是不行，还得用高端芯片，还得加大芯片用量。","text":"Dep的芯片是不多，但是他训练的结果很烂，还得加大芯片的用量才行。所以他现在说吹牛说自己用少量的芯片训练的更好，这简直就是骗人。我刚才实际体验了地真的很难用，给的答案都不准确。说明他的芯片用少了，用的低端了还是不行，还得用高端芯片，还得加大芯片用量。","html":"Dep的芯片是不多，但是他训练的结果很烂，还得加大芯片的用量才行。所以他现在说吹牛说自己用少量的芯片训练的更好，这简直就是骗人。我刚才实际体验了地真的很难用，给的答案都不准确。说明他的芯片用少了，用的低端了还是不行，还得用高端芯片，还得加大芯片用量。","likeSize":1,"commentSize":1,"subComments":[{"id":397128348545448,"commentId":"397128348545448","gmtCreate":1737977478225,"gmtModify":1737977482202,"authorId":3516069222588608,"author":{"id":3516069222588608,"idStr":"3516069222588608","authorId":3516069222588608,"name":"赌狗终将一无所有","avatar":"https://static.tigerbbs.com/5dea5b60703e4a675b801d078999d34d","vip":1,"hat":"https://static.tigerbbs.com/f84859c65fc2d7681e28f6b8e9b0535a","crmLevel":2,"crmLevelSwitch":0,"individualDisplayBadges":[]},"repliedAuthorId":0,"objectId":397125189472936,"objectIdStr":"397125189472936","type":1,"supId":397126960030304,"supIdStr":"397126960030304","prevId":0,"prevIdStr":"0","content":"我也试了，我出了一个工程题，deepseek算错两个小数点，chatgpt算对了。虽然说我们自己这种测试意义也不太大，但哪怕是对比deepseek自己v3和r1，训练量也是大幅上升。","text":"我也试了，我出了一个工程题，deepseek算错两个小数点，chatgpt算对了。虽然说我们自己这种测试意义也不太大，但哪怕是对比deepseek自己v3和r1，训练量也是大幅上升。","html":"我也试了，我出了一个工程题，deepseek算错两个小数点，chatgpt算对了。虽然说我们自己这种测试意义也不太大，但哪怕是对比deepseek自己v3和r1，训练量也是大幅上升。"}],"verified":10,"allocateAmount":0,"commentType":"hot","coins":0,"score":0},{"id":397133836947864,"commentId":"397133836947864","gmtCreate":1737979273353,"gmtModify":1737979277206,"authorId":4195272735879932,"author":{"id":4195272735879932,"idStr":"4195272735879932","authorId":4195272735879932,"name":"穷不达","avatar":"https://static.tigerbbs.com/af68781b8c44eade49ee4fce2cecabf8","vip":1,"crmLevel":1,"crmLevelSwitch":0,"individualDisplayBadges":[]},"repliedAuthorId":0,"objectId":397125189472936,"objectIdStr":"397125189472936","type":1,"supId":0,"supIdStr":"0","prevId":0,"prevIdStr":"0","content":"对。而且 日后如果需要将llm和视频图像进行整合。大模型所需的更复杂gpu场景 英伟达家的硬件必定还是硬通货。 而且小公司通过这波降本增效 即使大公司不再需要更多硬件 但是小公司仍然需要 还会吸引更多玩家入场","text":"对。而且 日后如果需要将llm和视频图像进行整合。大模型所需的更复杂gpu场景 英伟达家的硬件必定还是硬通货。 而且小公司通过这波降本增效 即使大公司不再需要更多硬件 但是小公司仍然需要 还会吸引更多玩家入场","html":"对。而且 日后如果需要将llm和视频图像进行整合。大模型所需的更复杂gpu场景 英伟达家的硬件必定还是硬通货。 而且小公司通过这波降本增效 即使大公司不再需要更多硬件 但是小公司仍然需要 还会吸引更多玩家入场","likeSize":2,"commentSize":0,"subComments":[],"verified":10,"allocateAmount":0,"commentType":"hot","coins":0,"score":0},{"id":397298604806784,"commentId":"397298604806784","gmtCreate":1738018968562,"gmtModify":1738018972290,"authorId":4091593421306780,"author":{"id":4091593421306780,"idStr":"4091593421306780","authorId":4091593421306780,"name":"wuxxx","avatar":"https://static.laohu8.com/default-avatar.jpg","vip":1,"crmLevel":2,"crmLevelSwitch":0,"individualDisplayBadges":[]},"repliedAuthorId":0,"objectId":397125189472936,"objectIdStr":"397125189472936","type":1,"supId":0,"supIdStr":"0","prevId":0,"prevIdStr":"0","content":"而且为了AGI这些多模态模型一样会吃算力，现在deepseek好像还不可以识别图像","text":"而且为了AGI这些多模态模型一样会吃算力，现在deepseek好像还不可以识别图像","html":"而且为了AGI这些多模态模型一样会吃算力，现在deepseek好像还不可以识别图像","likeSize":1,"commentSize":0,"subComments":[],"verified":10,"allocateAmount":0,"commentType":"valid","coins":0,"score":0}],"isCommentEnd":false,"isTiger":false,"isWeiXinMini":false,"url":"/m/post/397125189472936"}