You may like to delude yourself that the entire market of the earth do as yourself - savouring the benchmarks, then scouting around on the net for the best deal on your chosen chips, but most PC purchasers I have ever met usually go "I want a new PC, it may cost $X. You may find yourself looking at one $800 Intel and one $800 AMD system, because they bundled different brands of HDDs and keyboards and whatnot with it.įurther, the individual situation has a greater impact for the purchasing decision - shall I get the $820 rig now or come back next week and get the $800 one? Is my time and bother worth $20? If even that, because differences in other specifications and involved brands make for a bigger difference than $20. Who knows?Īnd who cares? The typical PC customer does not hold up one $200 CPU next to a $220 CPU and think "10% more expensive, I don't want that one!" because most people buy a whole system where the difference is more likely $800 vs $820 upon which the price difference becomes 2.5%. Today the 6th of December 2007 the AMD offerings might have bad bang/buck, next week they might drop the price and become better. The question is if you think the bang/buck is right, and that depends solely on the day-to-day fluctuations in price. In any situation that you consider the Intel2-5 alternatives, there is an AMD above it. The ranges overlap, if you make a list of the individual offerings in performance order it looks like this: You are looking at several ranges of CPU offerings that go between two endpoints of performance. Let me put it in unfunny words for you to understand: the market does not base their purchasing decisions on the same values as computer geeks do. I have been here for many years, pointing at things and having people stare at my fingertip over and over again.
#What is amd k10 Patch#
The patch for the Linux kernel essentially works around it by forcing page faults after updating PTEs, keeping track of these states in other parts of the PTEs, and ensuring that the Accessed and Dirty bits are never modified.Īlso, I want to point out absurdities in the attitudes of some people, but I ought to know better. These race conditions are aggravated as the cores run faster than the 元 cache (since, from the perspective of the cores, the 元 cache takes longer and longer to complete the change to the Accessed and Dirty bits, providing an increasing probability that an interloper will arrive), which is probably one reason why the Phenom 9700's launch has been delayed. Now two different processors have different copies of the PTE and don't know it. It sees that the entry is "clean", and it goes ahead and writes a PTE, having a "dirty" local copy that differs from the "dirty" copy on this processor. Another processor can snoop this TLB entry while the bits are being updated.So threads running on the other cores will end up accessing or writing to the wrong physical page frame, off a stale PTE. Now when this inconsistent core's copy is evicted, it goes back to the 元 cache, dirty, and yet inconsistent with the other cores on the processor. Other cores that have cached this PTE believe this is its value. Now when the 元 cache's version of the TLB is evicted, it is written out to the page table's portion in the 元 cache. It modifies the PTE and marks this TLB entry dirty. Say this core later on needs to change the same PTE.This core now has a "writeback" PTE in its highest-level TLB that it thinks is "clean". Another core can come in and grab the PTE out of 元 cache while the bits are being updated.
#What is amd k10 update#
The gist of the issue is that the update of the Accessed and Dirty bits in the 元 TLB is not atomic. The supposed frequency sensitivity can be explained by the fact that the 元 cache's clock rate stays at 1.8 GHz while the cores' and L2 caches' clock rate rises, I think.