Performance improvement with faster SDRAM (and ultimately DDR-RAM)
Some days ago there was a discussion here about the expected performance improvements
with DDR-RAM compared to SDRAM RAM. Someone suggested that -- if the improvements scale nearly
linear -- somebody should benchmark PC100 and PC133 RAM (both with CAS2) on a VIA and Irongate
platform and compare the performance difference. Well, that's exactly what c't (respected German
PC magazine) did in its latest issue. I only quote i815, KT133 and Irongate 850 here:
Platform Games Business Linux
i815 (PC100-333) 87,4 159 125,0
i815 (PC100-222) 90,7 164 122,3
i815 (PC133-333) 93,3 168 119,3
i815 (PC133-222) 95,6 172 116,8
Irongate (PC100-333) 80,7 147 145,3
Irongate (PC100-222) 82,5 150 142,9
KT133 (PC100-333) 88,3 161 126,0
KT133 (PC100-222) 90,5 163 124,2
KT133 (PC133-333) 91,3 164 123,9
KT133 (PC133-222) 93,0 166 122,2
Test platforms: i815 with Pentium III 800, Irongate with Classic Athlon 800, KT133 with
Thunderbird 800. 128 MB RAM, Asus AGP V3800 Ulra graphic card.
Benchmarks: Quake 3, 3DMark 2000 (Games); BAPCo SysMark 2000 (Business); Linux Kernel
Compilation (Linux)
My take: On the Athlon KT133 platform we have improvements of ca. 3,5% in games, 3,1% in
business and 3,1% in Linux. This is roughly 3,2% more performance overall. This
equals ca. 1% performance increase per 10 MHz (memory). If DDR-RAM scales linear we would
have additional performance increase of ca. 10% (100 MHz DDR) or 13% (133 MHz DDR) via
PC-100) -- assuming the latency stays the same. This doesn't sound much but keep
in mind that a 100 MHz faster CPU compares to ca. 8% more performance overall. So if all assumptions
are right DDR systems should have a performance advantage of ca. 100 MHz via systems with
SDRAM. So the 1 Ghz Thunderbird with DDR-RAM should easily surpass the 1.13 GHz P3 with SDRAM.
Andreas |
