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 : Advanced Micro Devices - Moderated (AMD) -- Ignore unavailable to you. Want to Upgrade?

To: fastpathguru who wrote (255194)	8/4/2008 7:47:00 PM
From: combjelly	Read Replies (1) \| Respond to of 275872

"Anyone share the same feeling of incredulity I had at viewing this chart from page 6?" It should tell you everything you need to know. They are cooking the books. "Fixed memory bandwidth... Ring bus... Driver/algorithmic scaling... Doesn't matter, just keep throwing more cores in there!" It isn't impossible. However, it is hard to believe that everything is cache resident, or that the resources don't get saturated well before 64 cores are reached.

To: fastpathguru who wrote (255194)	8/4/2008 10:58:20 PM
From: Saturn V	Read Replies (1) \| Respond to of 275872

Gotta wonder what magic blue crystals Intel has invented to solve one of the hardest problems in contemporary computer science. I think that the Larabee 'blue crystals' are only applicable to a subset of computational problems. Four threads per core means that a lot of the common data will reside in the same cache. The Ring Bus also means that the data in the adjacent core can be accessed at a high speed. I do not think that same "data locality" will hold in most situations. Finite Element Analysis will obviously benefit as does graphics. But even then it is surprising that the performance scales as well as it does. Obviously if the cores are on different chips, the scaling will be a lot worse due to the delay in accessing data . Two Larabees will have a performance < 2* Single Larabee performance.