Serverworks chip sets

[Donald B. Kinghorn]

We've had good luck with the Supermicro 370DLE. Using high quality CAS 2
memory we saw significant speedup in our quantum chemistry codes. I'm also
interested in the HE version but I don't think I can talk the guys here into
spending the money to check it out.

We've also had good luck with ATA/100 ide hard drives (ibm) on the Promise
controller. We've gotten 50MB/sec reads and writes with "real" code. I like
SCSI but it's hard to justify the expense when you consider that you can get
a 64GB ide drive and an ATA/100 controller for under $350. My experience is
that good SCSI drives fail just as often as good quality  ide drives. Biggest
drawback is -- for performance you should only use one drive per controller.

Hope this helps
-Don

> I am seeking advice on some alternative Supermicro Motherboards/Chipsets
> for a cluster.
>
> We are looking at the Supermicro 370DLE board, with ServerWorks
> ServerSet III LE chipset for the nodes, and a 370 DL3 for the head.
> The nodes would then use inexpensive IDE drives and fit in a 2U cabinet.
>
> Alternatively, it has been suggested to me that I should consider the
> Supermicro 6010H 1U solution, which uses the 370 DER board with
> ServerWorks III HE-SL chipset.  Apparently, this motherboard has 2X
> bandwidth to memory, and the board provides 2 100BaseT network
> connections, but it requires more expensive (and/or smaller) SCSI
> drives and is more expensive (an possibly noisier, to cool the 1U
> case).
>
> Earlier discussions by this group suggest that there is not a
> significant disadvantage to using IDE drives on nodes, but presumably
> there is not much disadvantage (other than cost) to using SCSI disks.
>
> My questions:
>
> (1) Does anyone have experience with the HE-SL chipset?  Are there any
> benchmarks that indicate relative performance the HE-SL and LE Serverworks
> chipsets?
>
> (2) Alternative motherboard/system suggestions?
>
> Bill Pearson
>
> _______________________________________________
> Beowulf mailing list
> Beowulf at beowulf.org
> http://www.beowulf.org/mailman/listinfo/beowulf