Int_2 mod done!

[patrik]

@Crusher:

I tried a one year old 10krpm 72GB IBM U160 disk on my CSPPC and got 34MB/Sec when doing raw reads from the disk. Ofcourse the CSPPC SCSI-interface was set to use the synchronous 20MHz 16-bit transfer mode... this mode is the fastest available mode and thus 40MB/Sec is the theoretical maximum for the CSPPC SCSI.

Just as a sidenote, even if the CSPPC SCSI-chip could handle faster SCSI transfer modes, like 80MB/Sec, it probably would not be able to transfer data to/from memory any faster. If you own a CSPPC, download bustest and execute it with the argument 'fast'. The results you will get from bustest is the bandwidth at which the 68k cpu can access the CSPPC onboard fastmem. You can assume that the CSPPC SCSI-chip wont be able to do any better multiple longword transfers than the 68k cpu. If you consider that, you will see that there is not much bandwidth headroom when using 40MB/Sec SCSI.

//end of long sidenote

(edit):

When using DiskSpeed to test what transferrates that could be achieved via the filesystem (as in everyday use), I got  a maximum of 32MB/Sec... not too shabby.


/Patrik

[patrik]

@Crusher:

The 68k doesnt determine the memspeed between the CSPPC scsi and CSPPC fastmem. The SCSI-chip on the CSPPC does the memory transfers by itself as it uses dma which means it accesses the memory in kindof the same way as the cpu. I was only implying the the SCSI-chip probably cant talk much faster to the memory than the 68060 as they both transfer at most 32-bits (a longword) during a buscycle (the ppc transfer at most 64-bit during a buscycle and should get atleast the double bandwidth towards fastmem) and thus a faster scsi-chip would not be able to perform much better... maybe a 10MB/Sec more with a 80Mb/Sec SCSI-chip... it was just a sidenote . The CSPPC scsi might even be able perform better in cpu-memory-bandwidth-intensive situations with lower spec cpu's as then there will be more fastmem bandwidth free for the SCSI-chip to use .


/Patrik