[NTLUG:Discuss] Video card recommendations

Jay Urish j at unixwolf.net
Tue Aug 23 10:22:26 CDT 2005



Bryan J. Smith wrote:
> On Tue, 2005-08-23 at 08:50 -0500, Jay Urish wrote:
> 
>>too much... I am on a budget..
> 
> 
> On Tue, 2005-08-23 at 09:47 -0500, Bryan J. Smith wrote:
> 
>>Then you're not going to find quality PCI-X for under $300.
> 
> 

I am on the horn at the moment with Jim Grant over at advance PC talking 
about AMD procs and motherboards that have a PCI-X slot AND X-16 maybe.

I think I am going to dump the lame Adaptec card and pick up a 3ware card.



> BTW, understand the "main problem" right now is the lack of PCIe storage
> controllers.  The only card I've seen is the $600 LSI Logic MegaRAID
> 320-2E (2 channel U320, 400MHz XScale, 128MB SDRAM) and it's not as good
> as its 320-2X cousin (it's the same IOP331 with a PCIe-to-PCI-X bridge
> in the IOP332).
> 
> The _only_ PCIe ATA cards are cheap, "dumb" ATA channels with
> "FRAID" (fake RAID).  They have *0* RAID logic other than the 16-bit
> BIOS Int13h disk services and use the CPU for the software driver.
> Because the software driver is proprietary, they aren't typically
> available for Linux.  The GPL reverse engineered drivers are just as
> crap, and you're better off with LVM2/MD.
> 
> 
>>So get a $100 Socket-939 nForce4 Ultra mainboard and use the 4 on-board
>>SATA with Linux LVM (RAID-0) and MD (RAID-1) organization.  The new
>>Athlon 64 x2 3800+ dual-core is approaching $300 and kicks the crap out
>>of anything Intel's got at ALU/FPU operations.  The only time Intel
>>competes is when "lossy" SSE is used, and even these latest Rev. E
>>Athlon 64 / Opterons have SSE3 microcoded in their FPU (just as fast,
>>better precision using the FPU, not a "lossy" SSE pipe).
> 
> 
> Use LVM2 if you want RAID-0.  Hold off on the LVM2 Device Mapper (DM)
> work like RAID-1.
> 
> If you want RAID-1 atop, then make 2-sets of 2 disks of RAID-0 using
> LVM2, and then mirror them with MD = RAID-0+1.  Don't do more than 1
> LVM2 operation on disks, there can be a race condition.
> 
> 




More information about the Discuss mailing list