高盛在NVDA上赚了50亿美元。。。。。。
时间：12-12
整理：3721RD
点击：
小小散户永远处于信息链的最底层 --- GS开始套现了，我们才知道为什么NVDA从去年12月中开始狂飙。人家都开始卖了，我们才刚刚知道它为什么涨了
November 25, 2013， GS bought NVDA @ $27/share; more than 10% of NVDA.
December 12, 2016, GS decided to terminate the warranty.
（NVDA Mkt cap52.28B; this warranty is more than $5B)
NVDA will be severely diluted.
We do not know GS will sell the shares to the public market or keep them for
long run.
Jan. 2.17 | About: NVIDIA Corporation (NVDA)
来源： NVIDIA Corporation (NASDAQ:NVDA) | Seeking Alpha
Long/short equity, deep value, special situations, contrarian
Summary
In November 2013, Nvidia sold about 64 million warrants to Goldman Sachs.
Today, both parties are entering in a termination agreement, whereby Nvidia
will settle in shares.
The termination agreement will cost Nvidia shareholders anywhere from 8% to
9.2% in dilution.
This is a major dilutive transaction and therefore negative for the stock.
November 25, 2013， GS bought NVDA @ $27/share; more than 10% of NVDA.
December 12, 2016, GS decided to terminate the warranty.
（NVDA Mkt cap52.28B; this warranty is more than $5B)
NVDA will be severely diluted.
We do not know GS will sell the shares to the public market or keep them for
long run.
Jan. 2.17 | About: NVIDIA Corporation (NVDA)
来源： NVIDIA Corporation (NASDAQ:NVDA) | Seeking Alpha
Long/short equity, deep value, special situations, contrarian
Summary
In November 2013, Nvidia sold about 64 million warrants to Goldman Sachs.
Today, both parties are entering in a termination agreement, whereby Nvidia
will settle in shares.
The termination agreement will cost Nvidia shareholders anywhere from 8% to
9.2% in dilution.
This is a major dilutive transaction and therefore negative for the stock.
高盛为啥卖？还不是因为intel开始用FPGA搞这人工智能？而且直接开放面向对象语言接口，速度快功耗低，狗剩当然要出货
10%而已，怎么能说白打工呢？
要看开发速度的。。
现在google tensorflow直接支持cuda。
啥时候支持一把FPGA？
FPGA搞人工智能必须要上量。而且是大公司搞。
不过都是google玩剩的，google都直接ASIC了。
也不一定，调试算法用cuda，快糙猛。真正部署生产环境还是用FPGA好。显卡功耗太大，一般电信商的机位提供的功率放不了几台
所以我说要上量。而且玩FPGA，intel不提供好轮子，个人是很难玩的。
而且技术的应用是分阶段的。
一开始用CUDA抢占市场，等做大了再考虑FPGA，如果能继续做大再上ASIC。
好像intel现在提供caffe的FPGA版本了，tensorflow不清楚
现在人工智能用的领域最多的是图像识别，大数据分析和广告算法吧，intel应该会提供好轮子的。不过还是CUDA快糙猛