[NTLUG:Discuss] HDTV card for Linux

[Paul Ingendorf]

10GB / Hour ~= 3MB/s well within the specs for ATA-4.  The devices your
sourcing like tivo currently work with drives in excess of 300GB (kernel
patch required).  As for the IO constraints of a 33Mhz 32 bit interface I
think that 125MB/s is plenty fast.  The biggest problem you have on these
architectures is memory io and the lack of a DMA controller.  They are still
adequate though and in consumer electronics adequate is perfect, it means
your not wasting money on something that does more than what it is intended
to do.  Bringing me somewhat to your conclusion that with a consumer device
like most of the PVRs you can buy today you will not see them supporting
HDTV in their current incarnations.  There are a few HDPVRs on the market I
don't know anything about them other than they do exist.

-----Original Message-----
From: discuss-bounces at ntlug.org [mailto:discuss-bounces at ntlug.org]On
Behalf Of Bryan J. Smith
Sent: Tuesday, September 07, 2004 5:39 PM
To: NTLUG Discussion List
Subject: [NTLUG:Discuss] Re: HDTV card for Linux


FYI, NTSC MPEG-2 storage is typically 1GB/hour, HDTV is 10GB/hour.  So
to get even a decent amount of storage for HDTV, you want 300GB+.

So I'd definitely recommend a regular PC for HDTV PVR.

- Old PVR solutions and the ATA-4 limitations

The problem being the common IBM PowerPC 405 and various MIPS consortium
4000/5000-series variants have chipsets and reference designs that only
support ATA-4.  Although they have have upped the uC/EuP
(microcontroller/embedded microprocessor) clock and put in a more
powerful tuner/encoder, the platform remains the same in these black box
HDTV PVRs from what I've seen.  So they will not support more than
128GiB (133GB) of storage in an ATA device.

At 120GB, that's barely 12 hours.  24 hours for two disks on a single
ATA channel.  Not idea when you're pushing that much data around --
possibly for 2 tuners as well.

- Watch the I/O contention on a single, legacy, shared 0.125GB PCI bus

Also consider the I/O contention of a chipset that puts ATA and PCI on
the same, shared, 0.125GBps (32-bit at 33MHz) PCI bus.  Even with the
tunner/codec only sending 3-5MBps to the I/O, when you are pushing
multiple streams out to the I/O, there could be some contention between
I/O-memory-storage and frame drop.

PCI-Express will address this nicely, as even forthcoming, legacy AGP
mainboard will have (2) 0.25GBps PCI-Express x1 slots on the southbridge
-- which as a full 0.5-1GBps connection to the northbridge (either using
PCI-X, like the Intel ICH or older VLink, or 8+8-bit at 500MHz
HyperTransport like on all other chipsets, even for Intel CPUs,
including ViA's newer offering).  These (2) PCI-Express x1 slot
chipsets, both on the southbridge, from SiS and ViA should be out on
mainboard within the next 4-6 weeks.

As far as right now, consider a mainboard with at least a bridged PCI
solution into a 0.5GBps (64-bit at 66MHz) PCI/PCI-X master.  A ServerWorks
chipset mainboard is ideal for P3/P4 (the i7500-series is an Intel
rebranded), and don't knock the aging AMD AthlonMP using the 760MPX
(762+768 north+south).  Otherwise, and possibly overkill, a
HyperTransport system with something like the AMD8131 dual-PCI-X tunnel
-- but I haven't seen a single Socket-940 with it.  I'd wait for
PCI-Express and put either the tuner (not likely to see a card soon) or
ATA (more likely to see PCI-E x1 cards for ATA) on it.

-- Bryan

P.S.  In case you did not know, PCI-Express is _serial_ I/O and not
backward compatible with PCI.


--
Bryan J. Smith                                  b.j.smith at ieee.org
------------------------------------------------------------------
"Communities don't have rights. Only individuals in the community
 have rights. ... That idea of community rights is firmly rooted
 in the 'Communist Manifesto.'" -- Michael Badnarik



_______________________________________________
https://ntlug.org/mailman/listinfo/discuss