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: jcholewa who wrote (20342)11/23/2000 4:12:01 PM
From: fyodor_Respond to of 275872
 
<JC: Dean Kent of Real World Technologies actually made me aware that "legacy" doesn't even mean what you think it does in the first place. Observe: ...>

The definition Dean Kent uses matches my understanding very well. The "problem" in this case is that legacy apps will, per definition, virtually never use the latest technology and optimizations. I'll admit that this puts a somewhat harsh spin on the word "legacy" ;).

-fyo



To: jcholewa who wrote (20342)11/24/2000 1:42:12 PM
From: dale_laroyRead Replies (2) | Respond to of 275872
 
I don't think Palimino will have SSE, but it might have hardware prefetch, and not the brain damaged hardware prefetch of P4.

I suspect that much of the explanation for both the great and poor results of P4 has to do with hardware prefetch. Bandwidth just isn't an important enough factor in Linpack to explain P4's stellar results. I think what we are seeing is that well behaved benchmarks, such as Linpack, Q3, and SPECfp are getting a large boost from hardware prefetch, while less well behaved benchmarks are taking an equal hit from the tendency of hardware prefetch to prematurely flush the cache.

AMD has patented a predictive cache architecture that associates prefetch with code execution patterns, presumably using bits in the branch history table. If AMD has improved the branch prediction of Palomino, they may very well have added the hardware necessary for predictive hardware prefetch.