cise4832

AMD成日比N記同Intel挖角 /_\

聽人講Intel QPI都有幾個人來自AMD HT team
Kelper後N記都挖左AMD唔少人

yjfoeg

希望真係唔係比我口臭 講中  AMD RIP @ 2016
dom 發表於 2015-10-14 17:21

係你衰
你睇下你用咩顯卡玩STAR WARS BETA

Stiggy930

Sadly AMD is not VIA ... which have a large"backup"  coporate to support it
MAYBE , just maybe  M ...
dom 發表於 2015-10-14 17:18

Matrox is still alive and it does not have a big "bank" behind it.

If AMD end up becoming another Matrox, it will be very sad indeed. But at the end, is better than nothing. If M$ is really into making high end devices, then they may. They just released S4 Pro and S4 Book, whether they are successful is still too early to be seen.

As of your other post saying that AMD is losing market share everywhere, I guess the board, Lisa Su, Mark Papermaster are all aware of this as all bets are on Zen and Greenland. They do not have the resources to have multiple development teams going to refresh product line. If Zen and Greenland are at least acceptable to the market, then AMD's life will not be too bad. If not, then yea, is GG.

Kundera

nv is big in its own realm but in the world of HSA there are only AMD and intel.

I am not quit sure what they can do by hiring Phil. It's not likely IBM and nv will produce a SoC together, any kind of integration can only be done on platform level through a faster than PCIe bus. Whether joining OpenPower or not won't give any fundamental change to nv's HPC business. Perhaps the differentiation between CUDA for x86 & Power is just adding complexity on the path to HSA.

Meanwhile if you look at intel's upcoming Purley platform/ xeon phi then you'll know nv's GPGPU kingdom will not last for long. It's doing everything nv dreamed to do, in a unified x86 platform. All I see here is nv's struggle of encapsulated self.

Ksec

@Stiggy930 @dom

Well it is more like AMD is dead, long live AMD. It might just be a very different AMD from what we know. As per @Stiggy930 example, yeah Matrox is technically alive, but i would call it dead. At least in the sense it no longer make any difference to the market.  Sigh

Stiggy930

nv is big in its own realm but in the world of HSA there are only AMD and intel.

I am not quit sur ...
Kundera 發表於 2015-10-14 18:51

C hing is right on most things but one.

Power.org is still the governing body around the Power Architecture instruction set but specific implementations are now free to use under a liberal license granted by IBM.

...

IBM is looking to offer the POWER8 core and other future cores under the OpenPOWER initiative[2] but they are also making previous designs available for licensing.[5] Partners are required to contribute intellectual property to the OpenPOWER Foundation to be able to gain high level status.

   
https://en.wikipedia.org/wiki/OpenPOWER_Foundation

In a way, nVidia can design a APU with IBM reference POWER 8 core. Just that, they have to share to all other members in the consortium. Therefore, nVidia is just sharing their NV-Links for now.

I think nVidia want to at least achieve virtual unifying address space between CPU and GPU at close to no performance penalty and want to make the future CUDA be way more easy to use. (OpenCL 2.0  has two modes of unifying address space. A lot faster than OpenCL 1.0 but still no where close to what HSA offer as HSA is built around a MMU that shared by both CPU and GPU that can do address space conversion on the fly.) At fast enough speed, NV-Link can reduce this performance burden further. I guess this is what nVidia is trying to do.

AMD actually has a similar project going on to speed up virtual unifying address pace for both CPU and dGPU by using Sea Micro IP.

Stiggy930

@Stiggy930 @dom

Well it is more like AMD is dead, long live AMD. It might just be a very different  ...
Ksec 發表於 2015-10-14 18:58

Let's just hope Zen and Greenland and the HPC APU did not fail.

16 Zen cores + Greenland based graphic HPC APU does sound very promising.

Kundera

C hing is right on most things but one.


   


In a way, nVidia can design a APU with IBM refere ...
Stiggy930 發表於 2015-10-14 22:05

Power APU is one possibility but so far we havn't heard of any code name or roadmap for this. There's a lot to do to achieve hardware level hUMA, I assume a modification in cache system/ memory interface of Power cores? Since it would be a HPC product that development will take much longer. If nv just started doing this we won't see an actual SoC in 3 years.

Ksec

AMD New Results are out... and they are shit....
http://www.anandtech.com/show/97 ... es-hurt-bottom-line

Read it, and you can properly search for Lisa's comment at the end of the Results.

The only good news is AMD has  $371 million from their Remaining ATMP Fab Spin OFF, which means they can last a little longer again.... LOL

http://www.anandtech.com/show/97 ... ure-for-371-million