Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Qualcomm Moderated Thread - please read rules before posting -- Ignore unavailable to you. Want to Upgrade?

To: engineer who wrote (133304)	12/19/2016 1:42:01 AM
From: Qurious	Respond to of 197030

AMP for $100M R&D? Were that the case, would have been done long ago. It is not a problem which can be solved with dollars. MP in general can be looked at as a matter of granularity in the parallelism. SMP granularity is stuck at the thread level, which is perfectly acceptable and viable -- for now. Otoh, AMP granularity is still stuck at functional block level (comm, graphics, dsp etc.), and so only suitable for dedicated/embedded applications, not servers. Going below those levels in either case has been mainly academic research, and not particularly impressive research at that. The problem with SMP is scalability. I raised Amdahl's law mainly in light of Q's breezy intro of its 48-core server cpu. In this case, there is no question it would run SMP Linux. The Linux kernel has thousands of spin locks of various granularity, and has never been commercially proven above 24 cores (using an Intel proprietary version). 48 core SMP Linux is uncharted territory. Beyond the thousand of locks, and the obvious memory bank contention issues, every application will need to be hand-tuned in order to scale in a remotely sensible manner to 48 cores. And those optimizations are generally not obvious in the least, and take great insight to identify and realize. Not saying we can't scale to 48 cores. But it will take an awful lot of very smart programming to take advantage of it. I'd be surprised if early trials do not max out at circa 8 cores.

To: engineer who wrote (133304)	12/19/2016 1:55:43 AM
From: Qurious	Respond to of 197030

Here's an overview of the scalability issue: halobates.de

To: engineer who wrote (133304)	12/19/2016 2:09:53 AM
From: Qurious	Read Replies (2) \| Respond to of 197030

A more detailed look: pdos.csail.mit.edu Note: the "stock" software (kernel+app) scalability looks ugly.

To: engineer who wrote (133304)	12/21/2016 10:57:41 AM
From: Jim Mullens	Read Replies (1) \| Respond to of 197030

Curious / Engineer re: Software issue / Qualcomm Demos 48-Core Centriq 2400 Server SoC “...We have a real catch 22 here. People can't take $100M to build a new processor wihtout sales. Sales takes SW and applications. Without Software, no speed up is shown. Making the software for asymetic cores can take $100M itself....” >>>>>>>>>>> Unfortunately your discussion is way over my head but the article below discusses software issues and why QCOM is working with / targeting GOOG / FB / AMZN. Who’s developing the software?.... it appears that QCOM is targeting the very large web service companies (Data Centers) that develop their own software. Snip>>>> *Qualcomm is currently testing the processor with large web service companies and looks to gain some market share from Intel.* *Competing with Intel in the server space* However, competing with the giant in its own market is not child’s play. As seen from the chart above, Intel’s x86 technology commands 99.3% of the server market. The remaining 0.7% is held by IBM’s ( IBM) Power9 processors and other ARM processors. All commercial software available in the market supports the x86. Even Advanced Micro Devices ( AMD) scrapped plans for an ARM processor and is now launching Zen, its x86 processor, in 2017. It’s difficult for an ARM processor to survive in an x86 world. Unlike other ARM chipmakers that are still struggling in the server space, Qualcomm is moving strategically by planning its entry. First, it waited for several ARM server chipmakers to exit the market. Then, it waited for applications to emerge that can run on all processors. However, that did not happen. So, Qualcomm ( QCOM) is targeting very large web service companies like Google ( GOOG), Facebook ( FB), and Amazon ( AMZN) that develop their own software and are willing to experiment with server processors other than the x86. These companies make bulk purchases of server chips and then use accelerators to speed up processing. A cheaper server chip would be attractive for them, as it would bring significant cost savings. marketrealist.com Qualcomm’s 2017 Resolution: Automotive, Mobile PCs, Data Center PART 5 OF 13 Qualcomm Could Enter the Server Market in 2017