SI
SI
discoversearch

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 : AMD, ARMH, INTC, NVDA
AMD 252.00-2.6%9:40 AM EST

 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext  
To: neolib who wrote (15938)5/10/2016 3:51:54 PM
From: fastpathguruRead Replies (4) of 72317
 
I meant on the performance side. If they can't get more than 20% increased performance between 28nm planar and 16nm FinFET then there are problems. But some of the other claims had much greater increases in performance, so I guess we must wait until actual reviews come out. My comment is wrt to the fact that people stopped regularly upgrading CPUs when the performance increase between generations slowed to 10% and its starting to look like GPUs are headed for the same problem. Tis true that most CPUs now use much less power than 5 years ago as well, but that is not so much of a selling issue unless its for mobile.
I think this is a very conscious decision on AMD's part; Given the embarrassingly parallel nature of GPUs, they could have tailored the # of CUs to whatever performance/power ratio they wanted. And, in time, they will.

IMHO AMD wants to put R9-level performance in every near-entry-level PC, and you can't (and now don't need to) do that with a 200-300w GPU.

fpg

BTW, I have this dream where some company like AMD designs a small, modular CPU/GPU/xbar unit that is tiled and pre-integrated on the wafer itself... With a matching suite of interposers & HBM, you could build true, scalable, wafer-scale SOCs. Want 4 "units" of perf/power? Simply don't cut the units apart on the wafer. Mate the block with the cheap-as-sin 4-unit-interposer, and attach memory... Want a very boutique 16 units? Find a good wafer and cut out a block of 16. Price decouples from cost, and you don't have to throw away half your wafer due to defects...
Report TOU ViolationShare This Post
 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext