Anand的模拟desktop测试实际上高估了K10的游戏提升

zaarath

我突然想起了NUMA对于单线程游戏的副作用，所以用一个K10比较两个K8实际上高估了K10的游戏性能，我在xtrememsystems发了相同的帖子，懒得翻译过来了：

If anything, Anand's K10 desktop preview overestimated K10's gaming advantage against K8.

People kept insisting the K10 was crippled by the server memories while conveniently ignoring the fact that the dual K8's were crippled as well because they were tested in the exactly same server box. So Anand's preview gave a very good indication how Phenom would fare against desktop K8 in single/dual thread test.

However, the gaming performance with dual K8s was actually further crippled by NUMA memory architechture. In Anand's own review for AMD 4x4, the 2,8G FX62 outperforms 3.0G FX74 in Oblivion by over 10%.

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2879&p=10

FX62: 88.4
FX74: 78.3

So K10's gaming performance advantage over K8, at least in Oblivion, was over estimated, more like 5% instead of 16%.

[ 本帖最后由 zaarath 于 2007-9-11 15:57 编辑 ]

Prescott

这么几个百分点，无所谓了，用不着斤斤计较。:lol:

zaarath

宜将剩勇追穷寇，不可沽名学霸王

itany

Vista应该是支持NUMA的，损失的性能还是可以找回来的……