Charles/Constantine, "What I see is that currently the P4 with its much superior Sandra scores sucks in real world performance. So, improving the already great Sandra scores does not tell me anything."
Your line of reasoning is flawed. First, the Sandra STREAMs are designed to generate 100% off-chip traffic regardless of the size of caches; therefore no, even significant, improvement in cache miss rate performance will be unveiled by this benchmark. However, this alleged cache improvements will show off in real world performance (I mean primarily SPEC scores. BAPCO/ winbenches will enjoy about 1/3 of those improvements).
Also, I would not apply the term "sucks" so boldly. It is apparent that the P4 _platform_ is fairly competitive, which simply demonstrates that if you are doing half a job but twice as fast, the result is generally the same. However, the potential for improvements seems to be slightly higher on P4 than for Athlons, IMHO.
"p.s. The 512K Cache will probably pull the SDRAM P4 up closer to the RDRAM/DDR version."
That's exactly the point: if the 512k will bring 20% performance improvement to SDRAM P4 platform (which the the current perceived performance gap), similar improvement can be anticipated for DRDRAM platforms too, which will give Intel more bragging rights.
- Ali |