Messages in this thread | | | Date | Sat, 24 Jul 1999 17:32:39 +0200 | From | David Olofson <> | Subject | Cache, x86, DSP... (Re: real-time threaded...) |
| |
Ove Ewerlid wrote: > > David Olofson wrote: > > Just 4 59309's in the basic configuration, though... Those DSPs are > > efficient, but it wouldn't take a monster Alpha to beat that > > performance. Might even be possible with Celerons or Xeons. (Well, > > P-II/III too, but they have slower caches. The cache size doesn't help > > much with small buffers/low latency.) > > Celeron/Xeon/PII/PIII all have 16+16 kb L1 cache.
Yes, I was thinking L2 here. That's very important for most kind of processing done in multitrack audio systems. The plug-in architecture requires lots of intermediate buffers in some cases. However, when a buffer is smaller than 128 bytes, this isn't much of a problem even with the 128 kB L2 of Celeron. (Real life lests show that Celerons beat P-IIs on audio processing under Windoze, even though Windoze needs significantly bigger buffers.)
> There are many data intensive DSP tasks you need the L1 speed to > match the needs of the CPU. FIR filtering for instances. > (particularly with KNI)
BTW, what's your experience of signal processing performance on Celeron/P-II/etc vs DSPs? Alpha or other architectures?
Price/performace of workstation hardware vs DSP cards? (What's the trend nowadays?) Provided workstations can do the whole job, including guaranteeing that there are no drop-outs, I guess DSP cards will have to be VERY affordable to be a viable option, as they still need a host workstation for many applications, and are more complicated and expensive to develop for. Comments?
//David
> Ove > > -- > Ove Ewerlid > Ove.Ewerlid@syscon.uu.se or Ove.Ewerlid@signal.uu.se > Phone: +46 70 666 23 63, Fax: +46 18 503 611, +46 18 555 096
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |