http://news.harmony-central.com/Newp/20 ... Ships.html
food for thought
PCI is dead?
It's too early to say that PCI is dead. The new technologies are as yet unproven as far as stability is concerned. PCI-X is fairly back-compatible with PCI as far as I know, as long as the legacy PCI card can run at the correct voltage (someone please correct me if I'm wrong here).
I have not had good experiences with PCI Express, although I've only tried it with an Nforce4 chipset.
I think Firewire can work, if it is approached in the correct way (with the RME Fireface, they programmed their own firewire chip on the device I believe - there's an article in this month's Sound on Sound about it). I haven't tried the Fireface though, and in my experience 'ancient' PCI has outperformed firewire every time.
Also, as far as I can remember, PCI audio cards only started becoming reliable in around 1997-1998, and started to get really stable with low latency even later (of course, CPU power also contributes to this).
IMO you would get much more out of improving system bus design (even the most expensive Mac G5s have problems in this respect) than the expansion slot technology. Multiple PCI buses (those Firewire/USB2 solutions all work over PCI after all) and a decent architecture to make it run reliably is the way ahead IMHO.
I have not had good experiences with PCI Express, although I've only tried it with an Nforce4 chipset.
I think Firewire can work, if it is approached in the correct way (with the RME Fireface, they programmed their own firewire chip on the device I believe - there's an article in this month's Sound on Sound about it). I haven't tried the Fireface though, and in my experience 'ancient' PCI has outperformed firewire every time.
Also, as far as I can remember, PCI audio cards only started becoming reliable in around 1997-1998, and started to get really stable with low latency even later (of course, CPU power also contributes to this).
IMO you would get much more out of improving system bus design (even the most expensive Mac G5s have problems in this respect) than the expansion slot technology. Multiple PCI buses (those Firewire/USB2 solutions all work over PCI after all) and a decent architecture to make it run reliably is the way ahead IMHO.
dArKr3zIn,
Most older motherboards have PCI slots that are 32bit/33Mhz and support only 5.0v signalling. Newer boards that meet PCI v2.3 proposed spec support both 3.3v and 5v signaling, but 5v PCI v2.2 is still the current spec for most PC motherboards not used for servers or high end workstations.
32Bit / 33Mhz with 5V signal
Several years ago a 64bit/5volt slot was used by to server motherboards to support faster scsi and LAN cards. This was backwards compatible with 32bit/33Mhz cards but if a 32bit card was used all cards would have to use 32bits only. From what I recall this was usually referred to as '64bit PCI slots' in literature, although I'm sure there's a PCI version number attached to it.
64Bit / 33Mhz with 5V signal
A few years later Revision 1.0 of the PCI-X spec defined PCI-X 66 and PCI-X 133 devices with the introduction of 3.3v signalling. I have also seen 100Mhz offered on some of the more recent server motherboards. When 32bit/33Mhz cards that support 3.3v signalling are used the whole PCI-X bus is locked at 32bits/33Mhz and all higher speed cards are thus locked to this rate. Many server boards include several PCI busses (often 2 different PCI-X busses and sometimes even a PCI bus).
64Bit / 66Mhz with 3.3V signal
Also there is an upcoming PCI-X 2.0 specification that includes PCI-X 266 and PCI-X 533 from the PCI-Sig. It also uses 1.5v signalling to allow the higher speeds, although the I/O buffers support both signalling levels for backwards compatibility.
Apple's recent G5 (and Xserve) line had the option for PCI or PCI-X, but to simplify the motherboard design their PCI bus was based on PCI v2.3 and supported ONLY 3.3v (since the same VRM would supply PCI or PCI-X).
PCI-Express 1.0 became an official spec in 2002, and 1.1 should be with us soon. While AGP/PCI are parallel, PCI Express is a serial array interface. Multiple PCI Express lanes can be connected to one device, giving x1, x2, x4, x8, x16 and x32. While x4, x8, and x12 are likely to be reserved only for the server market, x1, x2, and x16 remain for the consumers for now.
It has been suggested here before that Creamware should transition to Firewire (or USB 2.0), and since many percieve that TC's Powercore has supposedly already 'moved on'. It seems to me that TC releasing their new Powercore II as a PCI *first* stands as an interesting example. I would guess that TC will eventually follow up with a Firewire version of thier new card as well, but the fact that they haven't given up on the PCI bus after all might calm a few nerves around here.
dArKr3zIn refers to the life cycle of PCI sound cards, and I'm fairly sure that the future of our peripherals will follow a similar trend, taking several years to evolve. Intel's roadmap for PCIe in their own products has had this mapped out clearly for several years now, and for the forseeable future they generally only have PCIe slots intended for graphics with the occasional smaller array thrown in for good measure on later model lines. PCI-X has been with us for several years now and has not penetrated the consumer market at all beyond the Mac G5's.
Now this isn't to suggest that if Creamware released a new PCI based family of cards 5 years from now it would still fit the market, that's quite a ways away. I do suggest though that our current cards have a bit of life left in them, and perhaps Creamware can pull something out of their sleeve with profits from the new hardware.
<font size=-1>[ This Message was edited by: valis on 2005-07-02 10:46 ]</font>
Most older motherboards have PCI slots that are 32bit/33Mhz and support only 5.0v signalling. Newer boards that meet PCI v2.3 proposed spec support both 3.3v and 5v signaling, but 5v PCI v2.2 is still the current spec for most PC motherboards not used for servers or high end workstations.
32Bit / 33Mhz with 5V signal
Several years ago a 64bit/5volt slot was used by to server motherboards to support faster scsi and LAN cards. This was backwards compatible with 32bit/33Mhz cards but if a 32bit card was used all cards would have to use 32bits only. From what I recall this was usually referred to as '64bit PCI slots' in literature, although I'm sure there's a PCI version number attached to it.
64Bit / 33Mhz with 5V signal
A few years later Revision 1.0 of the PCI-X spec defined PCI-X 66 and PCI-X 133 devices with the introduction of 3.3v signalling. I have also seen 100Mhz offered on some of the more recent server motherboards. When 32bit/33Mhz cards that support 3.3v signalling are used the whole PCI-X bus is locked at 32bits/33Mhz and all higher speed cards are thus locked to this rate. Many server boards include several PCI busses (often 2 different PCI-X busses and sometimes even a PCI bus).
64Bit / 66Mhz with 3.3V signal
Also there is an upcoming PCI-X 2.0 specification that includes PCI-X 266 and PCI-X 533 from the PCI-Sig. It also uses 1.5v signalling to allow the higher speeds, although the I/O buffers support both signalling levels for backwards compatibility.
Apple's recent G5 (and Xserve) line had the option for PCI or PCI-X, but to simplify the motherboard design their PCI bus was based on PCI v2.3 and supported ONLY 3.3v (since the same VRM would supply PCI or PCI-X).
PCI-Express 1.0 became an official spec in 2002, and 1.1 should be with us soon. While AGP/PCI are parallel, PCI Express is a serial array interface. Multiple PCI Express lanes can be connected to one device, giving x1, x2, x4, x8, x16 and x32. While x4, x8, and x12 are likely to be reserved only for the server market, x1, x2, and x16 remain for the consumers for now.
It has been suggested here before that Creamware should transition to Firewire (or USB 2.0), and since many percieve that TC's Powercore has supposedly already 'moved on'. It seems to me that TC releasing their new Powercore II as a PCI *first* stands as an interesting example. I would guess that TC will eventually follow up with a Firewire version of thier new card as well, but the fact that they haven't given up on the PCI bus after all might calm a few nerves around here.
dArKr3zIn refers to the life cycle of PCI sound cards, and I'm fairly sure that the future of our peripherals will follow a similar trend, taking several years to evolve. Intel's roadmap for PCIe in their own products has had this mapped out clearly for several years now, and for the forseeable future they generally only have PCIe slots intended for graphics with the occasional smaller array thrown in for good measure on later model lines. PCI-X has been with us for several years now and has not penetrated the consumer market at all beyond the Mac G5's.
Now this isn't to suggest that if Creamware released a new PCI based family of cards 5 years from now it would still fit the market, that's quite a ways away. I do suggest though that our current cards have a bit of life left in them, and perhaps Creamware can pull something out of their sleeve with profits from the new hardware.
<font size=-1>[ This Message was edited by: valis on 2005-07-02 10:46 ]</font>
Because it is a serial topology...
Also in regards to north/south chips, the PCI-e spec falls halfway in between a bus interconnect specification (like the Hypertransport that AMD uses now) and a peripheral spec (PCI, IDE etc) meaning that it should replace both. From what I've read the PCI-e spec looks like a switched fabric for the backplane and PCI-e slots are direct connections into this, meaning that north & south are a single chip in the spec.
The part that caught my eye in Intel's documentation shows the controller still acting as the memory controller, meaning that Intel still hasn't got near-term plans to integrate a memory controller, and that PCI-e on AMD motherboards has to interface with Hypertransport to get access to memory.
Also something that will please audio users, PCI-e allows for prioritizing bus traffic so that data for realtime and streaming applications (like audio) will be treated as higher priority than other less important data. At least for a spec this is more elegant than allowing each peripheral to set the latency timer as it feels it should, although it remains to be seen how this will play out in reality. As it stands now it seems to me that at least with Nvidia their motherboards seem to 'prefer' the graphics data from the PCI-e graphics slot over our dsp cards (hence the problems with the Nforce4 boards).
Also in regards to north/south chips, the PCI-e spec falls halfway in between a bus interconnect specification (like the Hypertransport that AMD uses now) and a peripheral spec (PCI, IDE etc) meaning that it should replace both. From what I've read the PCI-e spec looks like a switched fabric for the backplane and PCI-e slots are direct connections into this, meaning that north & south are a single chip in the spec.
The part that caught my eye in Intel's documentation shows the controller still acting as the memory controller, meaning that Intel still hasn't got near-term plans to integrate a memory controller, and that PCI-e on AMD motherboards has to interface with Hypertransport to get access to memory.
Also something that will please audio users, PCI-e allows for prioritizing bus traffic so that data for realtime and streaming applications (like audio) will be treated as higher priority than other less important data. At least for a spec this is more elegant than allowing each peripheral to set the latency timer as it feels it should, although it remains to be seen how this will play out in reality. As it stands now it seems to me that at least with Nvidia their motherboards seem to 'prefer' the graphics data from the PCI-e graphics slot over our dsp cards (hence the problems with the Nforce4 boards).
Oh, I forgot to mention that the last 3 motherboards that I've owned all had both 32bit and 64bit slots. The one I run right now has 2x 64-bit 66MHz PCI (3.3V) slots and 4x 32-bit 33MHz PCI(5V) slots. The 64bit slots are really PCI-X slots, but I bought this motherboard before the spec was officially "PCI-X". The 32bit PCI bus is connected to the system bus through the 64bit slots, meaning that in theory if I swamped the 64bit bus it could affect my audio cards. In reality though this afforded me enough bandwidth to run 3 (non-RAID) high speed 10000 rpm SCSI drives (plus 2 IDE data drives) and left plenty of bandwidth for my 2 Scope cards. I no longer use the cards in this system, but for the 3 years that I did I never saw a single PCI overflow, and IRQ sharing has never been an issue. Of course the motherboard (including dual onboard adaptec u160 SCSI and intel Pro-100 managed LAN) ran me nearly $800 USD new back in 2001.
<font size=-1>[ This Message was edited by: valis on 2005-07-03 01:23 ]</font>
<font size=-1>[ This Message was edited by: valis on 2005-07-03 01:23 ]</font>
You know my current PC does come with a PCI-E X16 graphics card - an ASUS Extreme, because I wanted a system for graphics as well as music (I do art too). I also thought, maybe mistakenly, that PCI-E would not hog the bandwidth on the PCI bus. Of course I do have an Intel chipset so maybe thatcounts for a lot but I haven't reall had a problem with the PCI-E graphics card interfering with my audio at all. The main problem I've had was due to the fact that I only have 3 legacy PCI slots left to play with and even though it's an ASUS made mobo and graphics card the PCI-E slot is positioned so close to the first PCI slot that the graphics card blocks putting anything in that first slot so effectively I have just 2. And both share IRQ's with network adapters - talk about bad design!!Also something that will please audio users, PCI-e allows for prioritizing bus traffic so that data for realtime and streaming applications (like audio) will be treated as higher priority than other less important data. At least for a spec this is more elegant than allowing each peripheral to set the latency timer as it feels it should, although it remains to be seen how this will play out in reality. As it stands now it seems to me that at least with Nvidia their motherboards seem to 'prefer' the graphics data from the PCI-e graphics slot over our dsp cards (hence the problems with the Nforce4 boards).
The upshot of this is that I can't have the network on and play music at the same time - if I do I get system freezes, especially if downloading something. But it's easy enough to get round - I just have to disable the network while playing music - just a bit of a pita only having those 2 slots.
So bear in mind it might not even be 3 useable slots.
They do have some system bus bandwidth issues (like virtually all motherboards these days) but I've found them to be very stable and run glitch-free out of the box. I cannot say the same for current PC motherboards, and I've been building PCs for many years for audio purposes.
Regarding PCI-X, IMHO expansion card slot specs will always change.. it's inevitable. I trust Mac boards in this respect much more than PC boards though (as a result of experience in using them).
<font size=-1>[ This Message was edited by: dArKr3zIn on 2005-07-05 08:19 ]</font>
Regarding PCI-X, IMHO expansion card slot specs will always change.. it's inevitable. I trust Mac boards in this respect much more than PC boards though (as a result of experience in using them).
<font size=-1>[ This Message was edited by: dArKr3zIn on 2005-07-05 08:19 ]</font>
Valis - incidentally, I think those Powercore mk2 cards are the same spec as the Powercore Firewire (the full rack box, not the compact version).
I've used the Firewire box, and I remember reading in the documentation that it is not recommended to use multiple Powercore Firewire boxes, due to firewire bandwidth issues.
I've used the Firewire box, and I remember reading in the documentation that it is not recommended to use multiple Powercore Firewire boxes, due to firewire bandwidth issues.
Agreed
<br>
Unstable pc's are built on cheap mobos or with
an unfortunate (or thoughtless) choice of components. I reckon a lot of people also tend to pop down to their local crappy pc store for some memory or a HD & in honesty, though i've done it myself, it's stupid. ....full stop.
I'm thinking back to the Atari Falcon & the fact that all those years ago, you could record Digital Multitrack Audio into it, on a slow processor & HD (by 2day's standards, anyway) It's hilarious, actually
<br>
Unstable pc's are built on cheap mobos or with
an unfortunate (or thoughtless) choice of components. I reckon a lot of people also tend to pop down to their local crappy pc store for some memory or a HD & in honesty, though i've done it myself, it's stupid. ....full stop.
I'm thinking back to the Atari Falcon & the fact that all those years ago, you could record Digital Multitrack Audio into it, on a slow processor & HD (by 2day's standards, anyway) It's hilarious, actually
Ah so you're saying that the firewire box preceded the new PCI card. That may be, I don't use Poco here. Still stands as a decent example for the PCI naysayers.On 2005-07-05 08:42, dArKr3zIn wrote:
Valis - incidentally, I think those Powercore mk2 cards are the same spec as the Powercore Firewire (the full rack box, not the compact version).
I've used the Firewire box, and I remember reading in the documentation that it is not recommended to use multiple Powercore Firewire boxes, due to firewire bandwidth issues.