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: graphicsguru who wrote (239619)	8/30/2007 9:10:10 PM
From: Tenchusatsu	Read Replies (2) \| Respond to of 275872

GG, > Everything in your post that's decipherable is wrong. It's amazing how much Pete supposedly knows about the risks of speculative data usage. The funny thing is that Intel has been speculatively fetching data for quite some time now. It's called "memory disambiguation," a feature AMD is only beginning to implement with Barcelona. (Yes, it's a little different than what you are talking about, but I'm sure the risks are similar.) Tenchusatsu

To: graphicsguru who wrote (239619)	8/31/2007 1:29:47 AM
From: pgerassi	Read Replies (2) \| Respond to of 275872

Graphicsguru: You are so wrong in your posts outside of graphics, its obvious you don't understand what you are talking about. There are a lot of things that sounded quite good on paper that didn't work in the real world. I have been around a long time and I have seen this time and time again. Getting a pixel wrong doesn't hurt a graphics frame with millions all that much. Getting a bad result in a mission critical system may cost a life. People with those systems are quite rightly risk adverse. So unless it can be proven (mathematically or otherwise) that speculative cache processing doesn't lead to processing errors, they won't allow it. They have to consider every possible case. And if there is a near infinite number of different cases, either they will want something to limit the scope to what can be considered or deactivate it. So what you said is wrong. I tried to show where you were wrong, but since it seems to be beyond your comprehension, you think its drivel. TOUGH! Well it isn't. The reason why the rewind is exponential is that the CPU can affect x different things in one "hop". In two hops, it can affect x more different things and the x other things can affect x^2 more things, not just the 2x things you seem to assume. A rewind requires you to stop everything that could have been touched (that takes even more time), rewind what happened from all stop points back to when they weren't touched and then allow all of it to go on. At three hops, its x^3 and so forth. Soon x^n gets to be bigger than n*Y, even when Y is far larger than x. Given this, I assume that Intel will limit n to a small number by some means. Like one active speculation in flight per core or until a write is about to be done. Most of the benefits will be there with few costs and much lower complexity overall. Also note that they likely use a different coherency protocol within a die than between dies over CSI. AMD uses different cache coherency methods for on die (SRQ) and over HT (cHT). Pete