|[News] What a PPC linux kernel hacker thinks about the Articia
|Posted on 14-Jun-2004 23:52 GMT by JohnV
Here's what Benjamin Herrenschmidt, a respectable linuxPPC kernel hacker
thinks about the Articia.
is some interesting thread about the Articia. Note that Benjamin Herrenschmidt
is a clued linuxPPC kernel hacker with lot of
experience. Here's what he thinks about the Articia "features":
It's basically incompetent northbridge design.
No, that means the HW is a Piece Of Shit !
Cache coherency is a basic feature of anything claiming to be used as a
|List of all comments to this article
|Sorted by date, most recent at bottom
|not this again...
|14-Jun-2004 22:07 GMT
|Joe "Floid" Kanowitz
|14-Jun-2004 22:37 GMT
|14-Jun-2004 22:58 GMT
|Joe "Floid" Kanowitz
|14-Jun-2004 23:36 GMT
|Ben Hermans (Hyperion)
|15-Jun-2004 00:08 GMT
|15-Jun-2004 00:14 GMT
|15-Jun-2004 02:47 GMT
|15-Jun-2004 03:18 GMT
|15-Jun-2004 03:26 GMT
|15-Jun-2004 03:30 GMT
|What a PPC linux kernel hacker thinks about the Articia : Comment 11 of 68
|Posted by Anonymous on 15-Jun-2004 03:35 GMT
|If we believe for a moment that "it doesn't support cache coherency" is the magic Articia feature, that means Ben and Alan both spat nonsense over the past year about it. Not having cache coherent DMA means paying a high price for all DMA transfers, and adding complexity to your OS. It's silly even to call that a feature, let alone claim as Ben Hermans did that it makes MAI's hardware faster than x86 chipsets.
Worse, as is pointed out in that thread, Linux _does_ support architectures without cache-coherent DMA. If MAI wanted the Articia included in that group it should have been the work of moments to do so. So why did they insist on trying to treat it as if they had cache-coherent DMA? Incompetence? A deliberate attempt to mislead people into thinking it was (in Benh's thoughts) a real desktop chip rather than just cheap embedded tech?
Some people asked, when the Articia "feature" was last the subject of an ANN thread, if we would see performance indicators from a fixed Linux kernel, to know how much performnace difference MAI's feature really is. I think you have there your explanation for the foot-dragging at MAI. If the "feature" is the absence of cache-coherency it would be a LOT slower and it remains to be seen how much that cripples OS4 once the PowerPC port is complete.
|List of all comments to this article (continued)