To: combjelly who wrote (63808 ) 11/16/2001 12:52:05 PM From: Petz Read Replies (1) | Respond to of 275872 Athlon XP outperforms P4-2GHz in new Sandra benchmarks>hexus.net They used a beta version of the benchmarks which enable for Athlon XP's SSE optimizations among other things and compared an Athlon XP 1900+ (plus others) on a KT266A board vs. a P4 on the ubiquitous Asus i850 board. Processor XP1900+ P4-2000 Dhrystone_ 4428....3830 Whetstone 2217....2440 MultiM-Int_ 8816....7875 MultiM-FP 10161....9500Finally the Memory Benchmark. Before I show you the numbers, this new version of Sandra calculates the memory benchmark figures differently from all previous versions. It makes extensive use of data prefetch and buffering (not the data prefetch in Quantispeed, but it's own algorithm), uses buffering which is a common technique in memory intensive applications to ensure a steady flow of data and uses extra instruction sets like SSE heavily. Therefore using the the new Sandra, the current AMD processors and supporting chipsets turn out numbers a lot closer to the current leader in Sandra's memory bench, the P4 + i850(RAMBUS) combination. This reflects a real world situation where while the P4 + i850 has phenomenal memory bandwidth, it doesn't give it a huge lead that the Sandra numbers might indicate. These new numbers are more indicative of real world performance. But for some reason, Hexus.net used an Epox motherboard with the 760 chipset for the memory bench comparison. Even so, it came amazingly close to the P4-2000 i850 results: Processor XP1900+ P4-2000 MemoryInt 1997Mb/s 2177Mb/s MemoryFP 1906Mb/s 2188Mb/s Then these comments:These new Sandra numbers will become commonplace as soon as sites update their copies when the new version is released and you'll see a closer match between AMD and Intel system as far as the memory bench is concerned. We're happy to show you numbers from this beta version of Sandra as Adrian, the Sandra author notes that the beta we used was a genuine release candidate for the next stable version and the numbers will not change significantly. As a final point of reference, I've seen in the Labs a 1.2MP running @ 1548Mhz (172Mhz FSB) on an EPoX 8KHA+ (KT266A) and giving memory benchmark numbers of 2607 ALU / 2365 FPU, overtaking the P4 by a large margin. However, that's on a heavily overclocked system, but shows the potential performance of such a system running that FSB. Does this really make any difference? I personally discount ANY benchmarks like SiSoft that don't use real applications, but it looks like the one "benchmark" that embarrassed AMD will soon be neutral in the benchmark wars. Petz