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


To: graphicsguru who wrote (239612)8/30/2007 7:46:03 PM
From: pgerassiRead Replies (1) | Respond to of 275872
 
Dear Graphicsguru:

Actually as the hops go up, speculation averages worse. Rewind costs get multiplied by a factor of four (or so) for every hop speculated because the number of corner cases grows exponentially as the hops increase. So do the possible targets involved and that includes I/O targets (DMA, RAID, and Network Controllers plus GPUs to name a few). They too have to unwind and they usually have longer latencies due to slower clocks and longer pipelines.

Soon the savings are buried under the costs when things don't go right. I would think Intel would add a flag to turn off coherency speculation until its widely considered to be "safe". That way, if a mission critical user doesn't want to worry, the iffy parts can be deactivated for a performance penalty. Intel will have a lot of egg on its face if, like Hyperthreading, turning it off actually increases performance in most cases.

Pete



To: graphicsguru who wrote (239612)8/30/2007 8:04:07 PM
From: combjellyRespond to of 275872
 
"In fact the speculation should yield a bigger payoff for
large NUMA systems."

Can't argue with that. But the complexities make it unlikely.