Martin's PCIe Article

[Peter C]

Hi,

[Rant]

Well... there I was, waiting for the epxoy resin to dry on the first pre-productiuon model of my new quiet case.

When the latest SoS dropped through my letterbox.

So I opened it to read the wise words of the soundcard manufactures on the vexed question of PCIe.

I read, with mounting horror and amazement. This group of apprently informed technical guys seemed to have entered a collective ostrich emulation mode.

Only ESI thought that PCIe lay inside the planning horizon. Actually, it's been here for over a year.

Maybe they were all spouting the company line or had been told not to say anything to avoid giving their plans away to the competition.

Either that, or they are living in a time warp.

At present, all PCIe mobos have legacy PCI slots; but if you buy a machine now, I don't believe your next upgrade will have any PCI capability.

The PCI bus will dissapear far faster than the ISA bus did, because these days few systems have any PCI expansion cards - everything is integrated onto the mobo - so no legacy equipment to create a demand for mobos with a PCI bus. Plus PCIe is a fully agreed and undisputed international standard with all the major players pushing it.

And technically far superior.

It's not like nVidia is going to go on manufacturing nF3 chips just to satisfy the needs of the DAW industry.

So nobody - but nobody - should still be buying PCI soundcards, or indeed a PCI anything if it costs more than $10. A soundcard is an expensive investment, and typically lasts through two or three PC systems.

If you build a new DAW and use your existing soundcard, no problem. But if you are planning a totally new DAW including a new soundcard do you really want to invest in old technology?

Needs must...some people can't afford to wait, but it's getting to the ponit where my advice is this:

If you can wait, hold fire, go play tiddly-winks for a year, and wait till these cowboys decide they are in business to satisfy the needs of their customers. A radical idea, I know...

[/Rant].

Peter

[cc.]

The other possibility is that there never will be any PCI-e soundcards.

This is what happened with MIDI interfaces when ISA went away - they all went USB even though USB gives worse performance than ISA. No one bothered to make a PCI MIDI interface.

Not a nice thought, but certainly possible

[*INACTIVE USER*]

Well, there is something I do not understand: this forum is the only one where you will find that there are problems with pci slots on pci-e motherboards for audio. All other audio/sound forums I visit never mention the problem, and when I did once in a thread, there was unbelieve all around.

As for the ostrich mode...I said it before and that is the same as CC: audio cards will go usb and firewire. Could be that they are still in the last century, audio card manufacturers are not exactly known for following pc standards at any speed. Neither would it take much to switch, they could use a bridge (like the video manufacturers did) to start. And getting native pci-e interfaces wouldn't be difficult either (as stndard cells).

[Peter C]

Hi,

[rant] aside, my personal belief is that at least one of the people in Martin's article was holding out because they did not want to pre-release a product.

I still think we will see PCIe soundcards early in 2006.

Failing that, someone will produce a PCIe to PCI slot converter + bridge. That will cause merry hell with the size/positioning of the PCI slots in the back of the case; but I, for one, will modify my case design to cope.

We know that works - it costs about an extra $10 ($40 street price) to add a PCIe-PCI bridge onto an existing PCI soundcard...

And the variation in the retail price of an emu1212m is more than $40...

Peter

[Dishpan]

> I still think we will see PCIe soundcards early in 2006.

Of course we will Peter, either native or with a bridge.

[Martin Walker]

Hi All!

Well, I suspect that these manufacturers don't see a lot of point in investing large amounts of time and money in developing custom chips when (as Matthias Carsten of RME said) a complete solution will be available sooner or later from specialised sources.

After all, some of the companies who contributed to my feature spent several years designing their own FireWire chips, and one of them only finally launched FireWire products a few months ago as a result. Doing the same thing all over again when the number of musicians with PCI Express slots is currently so small could be financial suicide.

I fully understand your points about PCI slots disappearing, but cc. and Havoc also have a good point about PCI Express audio interfaces not being taken up at all by (possibly) a significant proportion of manufacturers. After all, if FireWire and USB 2 already do what most musicians want in the way of track and I/O count, why do a lot more work in developing a new range of products using a (for now) totally unfamiliar interface.

However, I suspect at least some of the companies are already working on PCI Express 'behind the scenes' and don't want to give any advance knowledge of their plans to competitors. We'll see

I hope you all enjoyed reading the feature though, even if you didn't agree with some of its sentiments

Martin

[Peter C]

Hi,

Yes I did enjoy the article... useful and interesting - all you can possibly do is report what people say.

I did want to ask about the reference to "a complete solution will be available sooner or later from specialised sources".

Surely the only customers for such a solution would be the soundcard manufacturers. It's not clear to me who these "specialised sources" are, and:
1. Why the soundcard manufacturers are not in closer communication with them - the "sooner or later bit" was somewhat laissez-faire.
2. Why the "specialised sources" were not also invited to the discussion.

The article (i.e the people) did make a clear case for the continuance of internal soudcards, but to me it seemed they ducked the issue that PCI will disappear. Saying "our PCI cards ar great, we see no reason to change" is not a response to the key question:
What are you proposing to do when PCI slots are no longer avaialble".

I'm looking at/for a PCIe to PCI bridge card...

Peter

[Gav]

I would expect in theory that it would also be possible to create a firewire to PCI bridge (as in have a small rack of PCI cards connected to a PC with firewire), or even using Gigabit Ethernet. Yes it would need drivers on the PC side and some clever embedded electronics, but would be entirely possible.

But why use firewire or ethernet for that? What if the future brings us hypertransport (or some other high speed interconnect) that allows PCs to become several smaller boxes (or several connected boards in a single case)? Need ISA, or PCI or PCIE or AGP or something else - buy a bridge that connects direct to the CPU/northbridge.

PCIE is the next step of interconnects, and so far for musicians it's not looking rosy. If we didn't have USB2 and firewire then I'm sure you'd have PCIE audio interfaces already, but we have them. I would expect my next interface to be firewire, but I have PCI cards that do the job fine for now.

[Jim Y]

I should think Via may well be working on PCIe audio controllers similar in capability to their current PCI Envy24 chips. M-audio, Esi and Terratec for example, rely on these to do most of the work in their PCI cards.

As an aside, I wonder what Esi used in the Maxi I/O? It's well beyond the channel count of any Envy24 that I'm aware of.

[PaulD]

Hi
Since the Mac world is already PCIe only for new models, and ProTools will have to be migrated to that to sell to Mac users, then no doubt others like MOTU will develop PCIe kit (all of which presumably will have to be useable in the next generation of Intel Macs).

[*INACTIVE USER*]

I didn't read the article, but I don't get the point Martin quotes from RME. AFAIK rme uses Xilinx programmable logic on their pci cards (from the pics at their site, both my cards and the way they answer to a pci query). If I look at this marketing blurb: "Royal Philips Electronics and Xilinx, Inc. demonstrated a programmable PCI Express endpoint silicon solution offered at 1/10th the cost than traditional solutions." from some internet news site then it gets weird. If you do a search on the xilinx site you get 3170 hits for "pci express". If there is one company that would be able to switch fast then it should be rme...

[Peter C]

I should think Via may well be working on PCIe audio controllers similar in capability to their current PCI Envy24 chips. M-audio, Esi and Terratec for example, rely on these to do most of the work in their PCI cards.

So VIA is the misterious "specialist supplier" then?

As an aside, I wonder what Esi used in the Maxi I/O? It's well beyond the channel count of any Envy24 that I'm aware of.

I beleive ESI make their own cores.

Peter

[Gav]

I didn't read the article, but I don't get the point Martin quotes from RME. AFAIK rme uses Xilinx programmable logic on their pci cards (from the pics at their site, both my cards and the way they answer to a pci query). If I look at this marketing blurb: "Royal Philips Electronics and Xilinx, Inc. demonstrated a programmable PCI Express endpoint silicon solution offered at 1/10th the cost than traditional solutions." from some internet news site then it gets weird. If you do a search on the xilinx site you get 3170 hits for "pci express". If there is one company that would be able to switch fast then it should be rme...

I would wonder what the cost of that IP is though. Can a sound business decision be made to go with PCIE right now? How about 3 months ago? If only it were as simple as you suggest...

[*INACTIVE USER*]

I would wonder what the cost of that IP is though.

Don't know, but it could be far less than you suspect. By providing the IP for a low cost, they get their chips designed into the card and bind a customer. Give a finger, take an arm.

[Peter C]

... AFAIK rme uses Xilinx programmable logic on their pci cards ... "Royal Philips Electronics and Xilinx, Inc. demonstrated a programmable PCI Express endpoint silicon solution offered at 1/10th the cost than traditional solutions."

I would wonder what the cost of that IP is though. Can a sound business decision be made to go with PCIE right now? How about 3 months ago? If only it were as simple as you suggest...

Business would be simple if you could just wait til the marked demanded a product and then go develop it.

In fact, you have to take a view, and balance the risk that a market does not take off against the risk of being left behind.

The thing is though, that PCIe is not a "maybe", it's a done deal. The only question is how soon the supply of mothboards that do have a PCI bus will dry up. If the future holds any internal soudcards, or FX processors like the UAD-1, at all they will be PCIe devices. Period. Developing a PCIe soundcard would not be a leap into the unknown, it just make you first into a developing market.

Because there is already a demand. Just look at the problems with PCI bus support on nF4 mobos. Every DAW manufacturer in the know Universe would heave a sigh of relief if a PCIe soundcard turned up.

Peter

[*INACTIVE USER*]

Right so. The decision to go with pci-e was taken before the first motherboards with it came to market.

[Mook]

Why release a PCI-E device when 32-bit and 64-bit drivers need to be written?

Don't you think the manufacturers will hold on until Vista is released so only one set of drivers need to written (and thus only one lot of support etc) - forcing us all to upgrade the OS in a big conspiracy!

[Peter C]

Hi,

You don't need to rewrite the drivers for PCIe - the drivers only see the device, not the connection to it.

AFAIK

Peter

[Jim Y]

Esi WaveTerminal 192 and Juli@ use the Envy24HT. Maybe they're using the same solution as RME in the Maxio? Would they design their own for a new product with a max five years life in a niche market?

Although it's been claimed that PCIe devices look the same to the machine as old PCI, according to someone from Creative, there is a significant difference to the way transfers occur which makes timely transfer a problem (as I read it, I don't have a link).

The way soundcards currently communicate has always baffled me. It doesn't seem to work using buffer sized burst transfers because the cards don't have a buffer to match the DMA one used by the driver (although consumer chips do appear to have hardware buffers). Rather, it looks like one sample for each channel is sent together via the DMA controller as needed by the cards audio clock. That's a lot of little packages transfered very slowly compared to the bus speed. Wouldn't it be better if the entire DMA buffer was sent in one burst? The soundcard can sync it to the audio clock. The host program shouldn't care as long as samples are dealt with in the right order.

It also seems that there is no system to detect lost transfers as there is with video where you can have "dropped frames" reported. Should the user have to wait until a recordings played back to discover there's been gapping? This shouldn't even be able to happen, yet current technology allows it, which is surely crazy.
If anything, it's this that I'd like to hear a soundcard developer explain "Why doesn't your driver know it's lost samples?".

The opportunity to improve the efficiency and integrity of audio data transport is maybe there with PCIe, it will be a great shame if it isn't.

[spjessop]

Whilst it was an interesting article, I found the discussion on testing to be mildly annoying.

I find it amazing that these sort of products need mollycoddling at all. You pay a premium* for them, and then you find out it doesn't work in your new premium motherboard. But wait, your FW800 hard drive is coping with massive amounts of data, why all the fuss?

They seemed rather proud of their testing methods, yet people do still have problems getting these premium* products to work properly on some systems. We will see real progress when you plug a USB or FW soundcard into any current generation motherboard and it just works with minimal preperation of windows.

By minimal I mean disabling a few unwanted startup tasks, and setting priority to background services.

I think in an ideal world most mid-level soundcards should be USB2 or FW (which is more than enough throughput) to maximise compatibility, then those precious PCIe slots can be used for DSP and MADI where the real power is needed. Then Universal Audio and TC Electronic could (should?) supply you with a patcher that optimises any other settings needed to get them to run well.

Thing is this utopian situation won't happen. No-one is listening.

* by premium I mean compared to many serial peripherals we pay over the odds for audio equipment.

[little person]

Hi
Since the Mac world is already PCIe only for new models, and ProTools will have to be migrated to that to sell to Mac users, then no doubt others like MOTU will develop PCIe kit (all of which presumably will have to be useable in the next generation of Intel Macs).

We have G5s and Pro Tools HD at work and have a PCI expansion chassis connected to each system already.I agree, I think digidesign will pave the way for others to follow....

[Wurlitzer]

So nobody - but nobody - should still be buying PCI soundcards, or indeed a PCI anything if it costs more than Peter C0. A soundcard is an expensive investment, and typically lasts through two or three PC systems.

If you build a new DAW and use your existing soundcard, no problem. But if you are planning a totally new DAW including a new soundcard do you really want to invest in old technology?

Well, people have different attitudes to such things, but I personally would not have a problem with buying a PCI card now.

Some soundcards might last through three systems, many will not. While the soundcard itself might still be working perfectly well in seven year's time (with each system lasting say three years), it's highly likely that the user's needs will have developed to the point that they want to revise their audio setup as they upgrade their computer. It's also virtually certain that they will be using a different OS by that time, and highly probable their software configuration will have changed and developed, opening up new possibilities and imposing new demands. Will there even by drivers for the soundcard that work under the new OS, regardless of its connection protocol?

Personally I've long ago given up worrying about what will be compatible with what in that kind of time scale. It's hard to make systems that not only work together happily, but also fufill all one's musical needs with the best possible interface and workflow. By the time you consider everything that needs to be considered, you're doing well if you can just get everything right NOW. Worrying about how it will look in five or ten years' time just makes it impossible - there are too many variables.

Others may think differently, but if the right soundcard for my needs happened to be PCI, and that suited the rest of my system, I'd buy it. It MIGHT be rendered obsolete when I next upgrade my system, or it might not. PCI cards might still be around by then, or there might be a perfectly well-functioning bridge or adaptor that can do the job. I might want to replace the soundcard by then anyway, or my studio might have blown up rendering the whole question academic. Then there's the fact that there are no PCI-E soundcards on the market now, and when the first ones come out, we have no idea what they will be like or how well they will work. The cutting edge is not the best place to be when it comes to something as delicate and temperamental as computer-based audio.

Like I said, too many unknowns. PCI soundcards in known good mobos with PCI slots work now. That's good enough for me.

[Jon Jon Jon]

I have found a similar rant going on over here
http://forums.anandtech.com/messageview.cfm?catid=27&threadid=1701961

there is also a link to why EMU arnen't manufacturing PCIe cards.

[*INACTIVE USER*]

Some soundcards might last through three systems, many will not. While the soundcard itself might still be working perfectly well in seven year's time (with each system lasting say three years), it's highly likely that the user's needs will have developed to the point that they want to revise their audio setup as they upgrade their computer.

Lots of this depends on how you plan and invest. When I decided it was time for a new soundcard I went totally over the top. So after 6 years my rme 9652 is still more than ample.

But at the same time I regret buying a pci-x raid controller, I should have bought a pci-e one. This one might get a second live, but it won't get 3 like the soundcard.

So if I change soundcards, it will be usb, whatever version is then current.

[ghr]

Meh, whatever, if they do bring out PCIe soundcards, then clearly the price of PCI ones will drop so much that we can buy for example, 3 M-Audio Delta 1010's, for around £50, and have them on our old Nforce 3 motherboards, and be well happy that everyone else is getting ripped off with PCIe and the latest stuff. Hey, we might even be able to get a current (as in current NOW) Protools HD system for not so much in the next few years. Bring on PCIe