thanks for the link to the product, i've ordered (for another purpose) the GigaByte iRAM ...
ok, 64 GB = 4 x 16 GB = 4 X $ 1.000.- (for the board) + 32 x $ 200.- = $ 10.400 (excl. VAT) for a poor-man's SSD (4 free sATA ports needed, system may remain 160 minutes without power, otherwise ... total data loss).
no need to mention for the symphonic cube you'd need 6-7 of them (besides about 50 free sATA ports of course) and i'm not sure this would be very welcome by users who sometimes consider the SC for $ 10.000.- to be expensive ...
technical details: this page says to have tested the hyperdrive on random read (without providing too many details like size of requested data packets) and mentions a latency of 0,2 ms. this would be a factor of 20 compared to the fastest sATA and SCSI drives and so calculating some overhead (but not other system latencies)buffer size could be reduced to 1/16 to about 4 kB. so buffering all 800.000 samples from the SC would need a computer with 3 GB accessible RAM - nice, should work. but - oops - what is this? average CPU usage 31% (on a not specified CPU) with a maximum of 67% ... seems to be unusable for a sampling computer.
the data thoughput is IMO not the real issue, because an average 40 MB/s like with many current harddrives would allow about 230 voices (stereo, already taking the realtime-decompression into consideration) to be streamed.
so all in all - to go this route seems to be not very practicable for me.
christian
ok, 64 GB = 4 x 16 GB = 4 X $ 1.000.- (for the board) + 32 x $ 200.- = $ 10.400 (excl. VAT) for a poor-man's SSD (4 free sATA ports needed, system may remain 160 minutes without power, otherwise ... total data loss).
no need to mention for the symphonic cube you'd need 6-7 of them (besides about 50 free sATA ports of course) and i'm not sure this would be very welcome by users who sometimes consider the SC for $ 10.000.- to be expensive ...
technical details: this page says to have tested the hyperdrive on random read (without providing too many details like size of requested data packets) and mentions a latency of 0,2 ms. this would be a factor of 20 compared to the fastest sATA and SCSI drives and so calculating some overhead (but not other system latencies)buffer size could be reduced to 1/16 to about 4 kB. so buffering all 800.000 samples from the SC would need a computer with 3 GB accessible RAM - nice, should work. but - oops - what is this? average CPU usage 31% (on a not specified CPU) with a maximum of 67% ... seems to be unusable for a sampling computer.
the data thoughput is IMO not the real issue, because an average 40 MB/s like with many current harddrives would allow about 230 voices (stereo, already taking the realtime-decompression into consideration) to be streamed.
so all in all - to go this route seems to be not very practicable for me.
christian
and remember: only a CRAY can run an endless loop in just three seconds.