z***t
发帖数: 10817 | 1 【 以下文字转载自 Stock 讨论区 】
发信人: zhangfei (Leo Kamis), 信区: Stock
标 题: Re: 谁给讲讲NVDA和INTC的关系
发信站: BBS 未名空间站 (Fri Nov 11 12:53:28 2016, 美东)
1. Intel still has better Silicon process. NVidia depends on TSMC on
manufacturing.
2. Intel missed and behind Nvidia AI several years. Right now is doing catch
up.
* buy Nervana @about 400m.
* buy Movidius.
* change Xeon Phi roadmap to include deep learning specific instruction.
* buy Altera @16.7B, trying co-processing concept.
* All the other AI investment through intel capital.
3. Nvidia is way ahead of Intel on AI eco-system. Cuda, cudnn lib on lower
level. All exiting AI software library, theano, torch, tensorflow, caffe,
mxnet..., application layer like language translation, image reorganization
and tagging, self driving car/truck, automatic drone (Jetson), robot ....
Very good relationship with internet company like Baidu, Alibaba, AWS,
Google...
=================
My personal think on critical point for Intel/Nvidia.
* Whether Intel can release Xeon Phi on 2017 as Intel promised. And in
hardware raw performance level, and in price level can compete with Nvidia
GPU offering.
* Whether Intel can and has the deep pocket to move the huge software
community to port existing AI software to Intel platform.
* Whether Nvidia can release its Volta platform on 2017 to widen its gap
with Intel.
* Nvidia needs to solve the PCIE bottom neck to scale up its GPU offering.
There is a hugh conflict of interests here. NVLINK vs new CPU co-processing
interface. Intel is not supporting NVLINK, IBM power 9 will be released in
2017, and support both.... Nvidia cannot count on Power 9, since more than
90% data center market is still Intel.
* Whether the whole AI is a hype bubble, or it is something real this time.
* 2017 will be a critical year. | N*****m
发帖数: 42603 | 2 差不多吧
不过intel问题不大，完全统治data center。
catch
【在 z***t 的大作中提到】
: 【 以下文字转载自 Stock 讨论区 】
: 发信人: zhangfei (Leo Kamis), 信区: Stock
: 标 题: Re: 谁给讲讲NVDA和INTC的关系
: 发信站: BBS 未名空间站 (Fri Nov 11 12:53:28 2016, 美东)
: 1. Intel still has better Silicon process. NVidia depends on TSMC on
: manufacturing.
: 2. Intel missed and behind Nvidia AI several years. Right now is doing catch
: up.
: * buy Nervana @about 400m.
: * buy Movidius.
| s*****l
发帖数: 7106 | 3 以后datacenter都是gpu cloud了
【在 N*****m 的大作中提到】
: 差不多吧
: 不过intel问题不大，完全统治data center。
:
: catch
| N*****m
发帖数: 42603 | 4 不可能，通用功能，gpu还是差远了
【在 s*****l 的大作中提到】
: 以后datacenter都是gpu cloud了
| s*****l
发帖数: 7106 | 5 有cpu 但计算主要是gpu
不是可能不可能的问题
是什么时候普及的问题
amz已经有了
【在 N*****m 的大作中提到】
: 不可能，通用功能，gpu还是差远了
| N*****m
发帖数: 42603 | 6 我当然知道有gpu cluster
我说的是gpu cluster用途有限
data center的主流还是xeon
【在 s*****l 的大作中提到】
: 有cpu 但计算主要是gpu
: 不是可能不可能的问题
: 是什么时候普及的问题
: amz已经有了
| l*******t
发帖数: 1430 | 7 看ai估值。几个暴涨nv吃下intel也未可知。 |
|
