Hi Havoc,

Reading your reply, it's clear to me that you and I have subtly different models of how the hardware works. Your position on the PCIe problem flows from your model... and mine flows from mine.

The truth may be a mixture. In this reply I'm not trying to argue for or against either view, just expose the differences so we can think about it.

Quote Havoc:

---

Quote:

---

You are quite right that the drivers must sort out the mess of shared resources, but they are doing that fine right now on a genuine PCI bus.

---

But there is a difference between a standard pci bus and a pci bus behind a pci-e bus. In a normal pci bus, there are at least 4 real hardwire physical interupts. And we do see that sometimes shared interupts cause problems, even on normal pci busses. But on a pci bus behind a pci-e channel, there is only the interupt from the pci-e channel. It has to be since that is the connection between the data and the cpu. It can be that the bridge groups the interupts from the pci and carries them forward to the pci-e switch.

---

Yes, there is clearly a difference "between a standard pci bus and a pci bus behind a pci-e bus", but it seems the difference is invisible to software. PCI cards work on a PCIe mobo without any driver changes and with an OS that is far too old to have any special PCIe code embedded in it.


IRQs have their origin in real hardware events, but these days they are mapped by both hardware and software, and are handled as data within the hardware (AFAIK). I don't believe the IRQs are allocated in the rigid fashion you describe. The fact that the PCI bus is connected via a single PCIe link through a PCIe switch does not (to my mind) mean that all the PCI cards have to share the same IRQ.

And the problems reported appear to occurr even when only one PCI device is installed... I'm not convinced IRQ sharing is the issue.

Quote:

---

But on the Tyan, due to the strange way the slots are divided between the 3 northbridges it can be less of a problem. If you plug your card into the single pci slot, it doesn't matter if there is a bridge. It is a 1-to-1 relation. And if you plug it into one of the pci-x slots, it is part of a normal pci bus since the pci-x bus is backward compatible and just slows down. Also the pci-e and pci-x are on different hyperlinks to the cpu.

---

Can't argue with any of that... but I find it strange that the problems (apparently) disappear completely on the Tyan however you set it up, and are present on all single-CPU PCIe mobos.

I feel that if this were are simple driver or IRQ interaction issue we would see much more variation in behaviour with different mobo + soundcard combos.

Quote:

---

But even then, I doubt there are much designs with a bridge. All this electronic design is done with vlsi libs. So if they made northbridges with several pci busses (the intel chipsets did that), then swapping one of them with a pci-e switch isn't a big difference. The way data is handled to the cpu stays the same, you use the fsb or hyperlink.

---

I agree the bridge does not seem a likely culprit.

The comparison between a pre-PCIe Northbridge with multiple busses (nForce 2 and 3 did the same thing) and a PCIe chips is clearly valid and instructive.

I'm not trying to be picky, but I think we may have an important difference here. I think that the PCIe switch corresponds to the entire northbridge chip, and each PCIe link (with one or more lanes) corresponds to a PCI (or other) bus.

My model of a nF4 chip is that it consists essentially of one huge multiplexor (the PCIe switch) talking via a HT link to the CPU/Memory controller on one side and to a whole series of PCIe links on the other. Severeral of those PCIe links have embedded bridges. These are esentially protocol converters - a PCIe <-> USB bridge, a PCIe <-> IEEE 1394 bridge, a PCIe <-> Ethernet, and of course a PCIe<-> PCI bridge. The Firewire and USB might be connected to the PCI bus rather than to PCIe links, of course.

In this model, PCIe can be seen as a rationalization of previous Northbridge designs, since any Northbridge with multiple busses is esentially a huge multiplexor also. However, PCIe also adds a new bus protocol - the PCIe link. This provides consitancy and standardization, but the PCIe link also has massively greater bandwidth than any previous bus, and can run at lower latencies.

Quote:

---

So I stay convinced this is a driver issue. Be it the very agressive video card or too much priority to the pci-e switch. I guess the last, exagerated by the first. Do not forget that if you design a bus that can handle 2Gbit/s (16 lane pci-e), then you have to write drivers that can handle that load. So whatever peep the pci-e switch makes, it is served. And since video cards ask a lot of attention, they get plenty of attention.

---

So here you are pointing the fingure at the PCIe graphics cards drivers, rather than the soundcard drivers? Possible...

I do agree that something seems to be swamping the PCIe switch and preventing it from servicing the PCIe link to the PCI bus properly. The PCIe graphics cards seems the only candidate. We differ in that I think the reasons lie in the hardware (or setup parameters/bios) of the PCIe switch, rahter than the graphics card driver.

It would be nice if it were the bios - but I'd have expected more variation and more movement on this problem if that were the case.





Peter

--------------------
PaQ

Quote Jim Wright:

---

Although Martin just wrote a column on why this is a perilous time to upgrade, this is one time I have to disagree. I think PCIe is going to muddy the waters for a while, and if you want a DAW using PCI sound cards, the window for building one is starting to close.... audio is too small a market segment to keep nForce3 motherboards in production, and many other market segments will happily migrate to PCIe.

Just my 2 bits --

Jim

---

I think you've reinforced my point there Jim

What with the current PCIe issues, all the warnings here (and elsewhere) about nForce4 (and yes, I am writing something for PC Notes to warn non-forum readers about this) plus dual-core versions of almost everything imminent, but with no real guarantees about which current motherboards will be able to accommodate these new processors (certainly not the Intel ones, but I suspect not all Athlon 64 boards will get BIOS updates), plus the singular lack of 64-bit music application and plugin support at the moment, it's an incredibly confusing time for anyone to buy or build a new PC.

Only those (like you) who are in possession of all the facts can make an informed decision to (in your case) go for an Athlon 64 and nForce3 chipset, which gives you a fighting chance of being able to upgrade to a dual-core model later on if you wish, and compatibility with all the 64-bit stuff later on as well, to keep your latest investment useful for as long as possible.

I do however concede Peter's earlier point that in practice a minority of musicians actually seem to upgrade any more than their hard drives and RAM (largely I suspect because buying a faster processor is nearly always so much more than any other component that it's often worth upgrading your motherboard at the same time and getting the benefit of lots of other new features).

You say that audio is too small a market segment to keep nForce3 boards in production, so your choice seems to be partly to buy one of the 'older' models before they disappear for good. This is sensible considering the other options at the moment, but few people normally decide to deliberately buy something thats already been 'superceded', because they are desperate to make sure their new purchase stays 'current' for as long as possible.

It's a confusing time for all of us, but I'm very pleased that you've seen through the confusion to make your decision


Martin

--------------------
YewTreeMagic

Hi,


Yes... it is perilous to upgrade during times of change; but much less so if you do the research/thinking.

nF3 socket 939 seems to me to be the safest bet, and the Neo2 is one of the three mobos I'm looking at for X2. MSI are definately working on an X2 bios.

The risks are that you might want a new soundcard or an FX processor in a year or two, and by then the hardware might be all PCIe. However, I think that's an unlikely scenario - getting the Audio manufacturers to react at all to PCIe is the current issue; and I can see no way (even if my crystal ball worked) of making decisions to protect against that sort of potential upgrade problem.

A socket 939 version of the Tyan S2895 would be the ideal answer - working PCI and PCIe support... but I very much doubt that will ever exist.



Peter

--------------------
PaQ

Quote Peter C:

---

A socket 939 version of the Tyan S2895 would be the ideal answer - working PCI and PCIe support... but I very much doubt that will ever exist.



Peter

---

The single cpu Opteron series is going over to socket 939, so such a board may well appear in the not too distant future. PCI-X is a requirement in workstation boards and none of the nForce4 boards offer this. My only guess is that someone will make one albeit a little pricy compared to 'gamer' boards.

Edited by spjessop (21/06/05 02:39 PM)

Is PCI-E a problem in general - or just on NForce 4 equipped mobos for Athlons.

Are people using Intel-based pc's on any of the mobos using
chipsets such as 915P Express, 925X Express etc having problems with their pci soundcards? These mobos support
PCI-E - so the older (and very common!) PCI architecture is obviously supported in some kind of compatibilty mode.

NOT trying to start an Intel/AMD war here.....just want to know if PCI-E should be avoided on ALL pc platforms or not!!

MikeN

Hi Mike,

Yes.

Intel PCIe has the same problem.

Or a similar one anyway.


On Intel chipsets the problem goes away if you fit a sufficiently modest graphics card (X300). I don't regard this as a fix - merely a circumvention - on the grounds that without knowledge of what the problem really is you don't know whether adding some other PCIe card, or changing graphics card, or updating some drivers, or... will make the propblem re-appear.

We are assuming (and it is an assumption) that the problems are basically the same in Intel and nForce4 chipsets.

If you think about it, you will see that this is consistant with my view that the problem is something to do with priorities and/or buffer allocation inside the PCIe switch.

But that's as much as we know...



Peter

--------------------
PaQ

Quote Peter C:

---

Quote Rick Taylor:

---

What would be wrong with putting a dual core processor in the Tyan and putting another in the other slot at some point in the future? What would be wrong with using a single processor for a while?

---

One dual-core Opteron is significantly more expensive than two equivalent single-core Opterons, each of which is similalrly priced to the eqivalant dual-core X2.

---

Yes... they are. By the time you really need them they will be old hat and the prices should be considerably lower.

Take Havoc's machine for instance... If, for some reason, the next version of Cubase requires something more powerful than dual Opterons in order to run in real time... Havoc can simply plug in two dual core Opterons {AMD supposedly has many of them in the works. {If you'll look at most of the entries in Tyans catalog they say, in big red letters... "DUAL CORE READY". We can probably assume that this means the company has tested a few processors and found them to be good. Otherwise they'd have to be morons to make those claims.}}

Given that it will quite likely be at least a year before this amount of power becomes an absolute necessity and given the rate at which the cost of processors drop... Havoc will simply be able to drop in two "affordably priced" dual core Opterons. In the meantime Havoc has the benefit of running a Dual Opteron machine with all of the goodies and extras and hugely wide pipes of his present Tyan.

The upgrade should be simple and require a minimum of effort. I can generally pop in a new AMD processor in about 5 minutes. Popping in two should take him ~10 minutes. This means that he need not change most of his system and can simply go back to work with no real expenditure of time or energy.

--------------------
www.RTaylor-Design.Com

Quote:

---

it's clear to me that you and I have subtly different models of how the hardware works. Your position on the PCIe problem flows from your model... and mine flows from mine.

The truth may be a mixture. In this reply I'm not trying to argue for or against either view, just expose the differences so we can think about it.

---

No problem with that! Two know more than one. Truth will very likely be in a middle ground. But the big problem I have is that nowhere I can find something representing what goes on inside one of those nForce chips. This makes all we do nothing more than speculation. Nothing wrong with a bit of discussion, is there?

What I find totaly strange is the post that it works with a Creamware card! My experience with a creamware Pulsar was such as to decide to never let any one of those cards come inside the house before I die. Normally, they only work well in a complete Intel system and even then... When they came out, they would not even work in an Intel dual cpu, and it took until hyperthreading to have creamware do something about that. They don't even mechanicaly fit the pci standard (ask some apple users).

Diametrally opposed to that is that it is RME that is the most vocal about their cards not working! A company well known for their excellent drivers.

--------------------
Expert in non-working solutions

I have a creamware pulsar 2. They are pretty pci bus hungry. It didn't like my first mb (some Via chipset thang, I can't remember the exact board, it died anyway) but its run solid and stable on my Asus MB and 2gig Athlon processor for years now.

Quote Rick Taylor:

---

Quote Peter C:

---

Quote Rick Taylor:

---

What would be wrong with putting a dual core processor in the Tyan and putting another in the other slot at some point in the future? What would be wrong with using a single processor for a while?

---

One dual-core Opteron is significantly more expensive than two equivalent single-core Opterons, each of which is similalrly priced to the eqivalant dual-core X2.

---

Yes... they are. By the time you really need them they will be old hat and the prices should be considerably lower.

...

---

But that still leaves you with the Tyan and one Opteron 275 as your upfront cost in order to match the power of the X2 solution, which comes to $2100 vs $800 for the X2.

My objection is simple - the investement and future proofing do not justify the extra upfront cost (IMO).



Peter

--------------------
PaQ

Quote Havoc:

---

... But the big problem I have is that nowhere I can find something representing what goes on inside one of those nForce chips.

---

Yes... I looked for details too, and found the PCIe specs on a site which required a password to access them. It seems we have an international standard that is not public domain! Really helpful.

Which is why we are guessing and discussing; and why I bothered to put forward my model of how a PCIe switch/chip works. It may be the best we have.

Quote:

---

What I find totaly strange is the post that it works with a Creamware card! My experience with a creamware Pulsar was such as to decide to never let any one of those cards come inside the house before I die. Normally, they only work well in a complete Intel system and even then... When they came out, they would not even work in an Intel dual cpu, and it took until hyperthreading to have creamware do something about that. They don't even mechanicaly fit the pci standard (ask some apple users).

---

I'm not necessarily convinced it really does work with the Creamware. The tests done by jschild on the Nuendo forum show that the issues turn up under heavy load.

There are four possible reasons why the Creamware might [appear to] work:
1. Not pushing the machine very hard
2. Pure blind luck
3. Running relatively high latencies
3. Something specific to the Creamware hardware

I think it would be very interesting to find out which; because if it's either the third or fourth posibility it's important information.

So: @ jonboyblue and vanity,
as much as you can tell us please?

Quote:

---

Diametrally opposed to that is that it is RME that is the most vocal about their cards not working! A company well known for their excellent drivers.

---

My hunch is that the lower the latency you try for, the worse the problem gets. In other words... the better quality the soundcard the more likely it is to have problems.

This is consistant with the fact the ordinary PCI cards do work on PCIe mobos, and might give us an angle on why the Creamwares work.


Peter

--------------------
PaQ

Quote:

---

My hunch is that the lower the latency you try for, the worse the problem gets.

---

Might be. The creamware works the machine hard. You have the "OS" for it running and every change etc must be passed on to the card. And this is the only card I know of that can saturate a pci bus. If you use a single channel that comes into the card and then back out of the card, you just add reverb until the pci maxes out. This is on a dsp card that is supposed to offload your cpu...

--------------------
Expert in non-working solutions

Hmmmm

Strange design.

That's maxing out the PCI bus on bandwidth. It does not tell us anything directly about the Creamware's PCI latency requirements.

Hard to tell what's significant here.



Peter

--------------------
PaQ

Quote Peter C:

---

Quote Rick Taylor:

---

Quote Peter C:

---

Quote Rick Taylor:

---

What would be wrong with putting a dual core processor in the Tyan and putting another in the other slot at some point in the future? What would be wrong with using a single processor for a while?

---

One dual-core Opteron is significantly more expensive than two equivalent single-core Opterons, each of which is similalrly priced to the eqivalant dual-core X2.

---

Yes... they are. By the time you really need them they will be old hat and the prices should be considerably lower.

...

---

But that still leaves you with the Tyan and one Opteron 275 as your upfront cost in order to match the power of the X2 solution, which comes to $2100 vs $800 for the X2.

My objection is simple - the investement and future proofing do not justify the extra upfront cost (IMO).



Peter

---

I understand your objection. With the Tyan board you can start with 2 cheaper Opterons and upgrade to dual cores when the need arises in the future.

Some series 200 Opterons are pretty cheap at the moment It looks to me like you could do the Tyan and processors for right at $900.

Pricewatch lists the K8WE for ~$465.oo... They have a list of processor prices if you care to look. In a year or so dual core Opterons should be considerably cheaper. A Tyan board with 2 dual core Opterons is still going to be a really hardcore machine for some time yet. A dual Opteron with two mid-level processors should do anyone fine for the time being.

--------------------
www.RTaylor-Design.Com

Hi Rick,

Quote Rick Taylor:

---

...
I understand your objection. With the Tyan board you can start with 2 cheaper Opterons and upgrade to dual cores when the need arises in the future.

Some series 200 Opterons are pretty cheap at the moment It looks to me like you could do the Tyan and processors for right at $900.

---

Yes - you could, though it would be closer to $1000 at UK prices.

What that gets you is a system costing perhaps $200 more than the X2, with significantly less total CPU power than the X2 but the quality etc benefits of the Tyan.

At that price difference you are probably talking two 1.8Ghz Opterons vs the 2.2Ghz X2, so the performance defecit is around 25%. It's still a faster sytem than you can get out of any single single-core solution, though: so it has its place, clearly.

Because you are no longer comparing like for like its a tough call. In terms of bang/buck on day one it's a no-brainer; but the long term maps out completely differently.

If your software can (or will) use 4 cores effectively and you are going to need that sort of power inside 18 months (but not initially) the Tyan is the right choice. The upgrade path is there, but it's not cheap now and I don't think it will get cheaper very quickly. The Opterons are sold into the server market which is not particlalrly price sensitive. Reliability, upgrade potential and TCO are the key considerations.

If you are likely to need four cores, but only further down the line, the upgrade might get cheaper. On the other hand the upgrade path might go away entirely, as the Opterons will be on socket M2 by then. Tricky.

If two cores will be sufficient in the longer term the Tyan has nothing to offer in terms of CPU upgrades that the X2 can't do cheaper.

The Tyan clearly offers the prospect of runing a PCI soundcard alongside PCIe versions of the UAD-1 (for instance); or of upgrading to a PCIe soundcard. As this stuff does not yet exist there cannot be any guarantees. Tricky.

The other technology on the Tyan certainly looks very nice, but actually putting a value on it all for a DAW is hard. For a long time people have been using server technology to make high-end workstations. The two are not the same. Dual-core may make the distinction much clearer.

My perception of a DAW is basically a very straighforward machine - with an emphasis on CPU power, Low Noise, RAM and HD (in that order).

It's a judgement call, but in the end I'd either go straight for the X2 (if I needed the power) or start with a single-cored socket 939, forget PCIe entirely, hold on to my $$$, and upgrade to an X2 when I needed to.

I'd be looking for a guarantee that the mobo would accept the X2, of course. That's what I'm testing right now.


Peter

Edited by Peter C (22/06/05 02:27 PM)

While I understand both Peter's and Rick's reasoning, it all boils down to preferences and use. And I don't think that after some time -let say 10 years- and both doing the wanted changes/upgrades the price difference would be significant.

--------------------
Expert in non-working solutions

Very conciliatory. ANd fair. Can we leave it at that?


Peter

--------------------
PaQ

I'll drink to that

A drink? Then we can start on the respective merits of Highland vs Island vs Speyside? A spot of Caol Ila anyone?

--------------------
Expert in non-working solutions

I'm very fond of Kahula actually.


Peter

--------------------
PaQ

Martin -- In most respects, I think we're violently in agreement It's definitely a confusing time for making PC purchasing choices.

However -- I do think it's going to get worse before it gets better.

• PCIe boards will be increasingly common; PCI/AGP boards increasingly scarce. As time passes, PCI/AGP will be viewed more and more as a legacy/low-end item, and high-performance PCI boards will vanish sooner than the commodity-market variety.
  
• A mature PCIe audio card is probably a considerable ways off. The technology is relatively new; drivers and protocols for sharing the PCIe bus "fairly" (preserving audio quality, not boosting graphics card scores) will likely take time to work out, and I suspect many vendors will prefer to focus on Firewire (and perhaps USB 2.0) products, where the core technology is much better understood (especially w.r.t to glitch-free, low-latency audio protocols).

Considering both points, it seems likely that within 6-12 months, it will be a lot easier to put together a system around Firewire (or perhaps USB 2.0) peripherals than around PCI/PCIe audio peripherals. If one is content to use Firewire, it does make sense to wait some time for dual-core processors to become cheaper, for BIOS/motherboard issues to be resolved, and for initial teething pains with PCIe graphics cards to be sorted out.

However, if one wants to use PCI-based audio cards, I think it makes more sense to buy a system now, rather than in a few months (as long as one buys a system with reasonable upgrade potential). Looking at the costs of the DAW I'm building -- the CPU (Athlon 64 3700+ "San Diego") was about $330, the RAM (2x1G "EL") about $300, the motherboard $130, the power supply $90 (Seasonic SP-12 430W), and $400 for the 1820M (at Page Computers, a discount source). A pair of new HDs (2x160G Samsung 7200RPM) totalled about $170. Other parts were largely on hand, or scavenged from my current DAW; all told, I will have around $1500 invested in the system (excluding LCD monitors). The motherboard is the least expensive item, and if I have to replace it down the line, it won't bother me much (except for the hassle of reinstalling WinXP OEM). But even the CPU is only about 20% of the system cost.

For me, it seemed best to invest in the "mature" end of the PCI generation of systems (even though they're clearly on the way out), rather than riding the bleeding edge of the PCIe generation. Several factors influenced this: my current system is an aging P4 that is starting to hit the wall with current memory-intensive sample-based soft synths (and I can't expand it beyond the current 1G of expensive-but-slow RDRAM). Futhermore -- my second daughter is starting college in about a year, and purchasing a new system then (or during the next few years) would be tough to do. Finally -- I've been waiting about 18 months for an Athlon 64 CPU with 1M cache, dual-channel memory support, and a price-tag I could afford. The "San Diego" is the first offering that met all the criteria -- and it seemed time to just do it, and get on with making music.

And as far as the 64-bit world is concerned -- I know of people running WinXP 64, 64-bit 1820M drivers and Sonar 4 (64 bit beta), with excellent results, on systems very similar to my new DAW. So I'm fairly confident I'll be able to do that as well, perhaps this Fall.

If your current system still basically meets your needs, you're willing to wait perhaps 12 months before upgrading, and you're willing to use Firewire peripherals when you do -- then it does make sense to wait. But, if you want to use PCI peripherals (either one you already own, or something like the EMU DAS series), or you're pushing the limits of your current system -- then upgrading this summer (rather than the end of 2005) makes more sense to me.

Best regards,

Jim

Edit: I just reread Martin's PC Musician column; he's only recommending waiting a month or two for the dual-core processor situation to settle down. That's quite reasonable, and still within my suggested window of "before the end of the summer". I'm still content with my decision to get the "San Diego" 3700+; the X2 4400+ (the least expensive version with 1M cache) will cost around $581, vs. $330 for the 3700+, and it may take a while for drivers and software to get sorted for dual-core operation (I'm thinking of many products that had issues with HyperThreading, but perhaps dual-core is less disruptive to the run-time environment -- I haven't investigated).


Edited by Jim Wright (23/06/05 12:12 AM)

Quote Peter C:

---

At that price difference you are probably talking two 1.8Ghz Opterons vs the 2.2Ghz X2, so the performance defecit is around 25%. It's still a faster sytem than you can get out of any single single-core solution, though: so it has its place, clearly.

---

Everything about it's tried and true and you get a motherboard that will do pretty much anything you want as well as being stable, capable, yadda, yadda, yadda...

Quote:

---

Because you are no longer comparing like for like its a tough call. In terms of bang/buck on day one it's a no-brainer; but the long term maps out completely differently.

If your software can (or will) use 4 cores effectively and you are going to need that sort of power inside 18 months (but not initially) the Tyan is the right choice. The upgrade path is there, but it's not cheap now and I don't think it will get cheaper very quickly. The Opterons are sold into the server market which is not particlalrly price sensitive. Reliability, upgrade potential and TCO are the key considerations.

If you are likely to need four cores, but only further down the line, the upgrade might get cheaper. On the other hand the upgrade path might go away entirely, as the Opterons will be on socket M2 by then. Tricky.

---

Tyan makes the appropriate boards. They're {mostly... they've put out a couple of expensive lemons} solid as a rock and the performance is outstanding. They have lots of experience with servers and have been putting out Opteron compatible boards since the beginning. You could always do the same with one of their cheaper boards or a different configuration so long as it's got the "dual core tag" on it somewhere. You could probably get by for $6-700.oo without memory. The system's still going to be sufficient for anything on the market now and you've got the upgrade kicker...

The processors in stock now will still be there for several years {you can still buy p3s}. AMD looks to be re-targeting the Opteron market. 6+ plus new processors in the queue right now... who knows what's eventually going to come down the pike.

Quote:

---

If two cores will be sufficient in the longer term the Tyan has nothing to offer in terms of CPU upgrades that the X2 can't do cheaper.

---

Dual cores aren't necessary now. They will be once the software folk get on the bandwagon and start adding additions onto the software version of Barbie's Dream House and MS adds "Supersize" skins to the Office Paper Clip. It will probably take a few years before the type of power we're talking about with the dual system is obsolete {or even lacking at all}. Dual core Opterons will have dropped in price considerably by then. {I'm sure there will be additional options}

Quote:

---

The Tyan clearly offers the prospect of runing a PCI soundcard alongside PCIe versions of the UAD-1 (for instance); or of upgrading to a PCIe soundcard. As this stuff does not yet exist there cannot be any guarantees. Tricky.

---

Havoc's experiences as well as the Nuendo forum folk's experiences have been pretty positive.

Everything involves a certain amount of risk.

Quote:

---

The other technology on the Tyan certainly looks very nice, but actually putting a value on it all for a DAW is hard. For a long time people have been using server technology to make high-end workstations. The two are not the same. Dual-core may make the distinction much clearer.

---

Bigger pipes are bigger pipes. More ram is always good. Niceities are niceities. Jaguar XKEs are Jaguar XKEs.

Quote:

---

My perception of a DAW is basically a very straighforward machine - with an emphasis on CPU power, Low Noise, RAM and HD (in that order).

---

Simplicity, bandwidth, power... you can always stick on more drives and stick the machine in the closet should it prove to be noisy.

Quote:

---

It's a judgement call, but in the end I'd either go straight for the X2 (if I needed the power) or start with a single-cored socket 939, forget PCIe entirely, hold on to my $$$, and upgrade to an X2 when I needed to.

I'd be looking for a guarantee that the mobo would accept the X2, of course. That's what I'm testing right now.

---

http://www.tyan.com/products/html/opteron.html

"DUAL CORE READY"

Quote:

---

Peter

---

I'm not saying it's a bad choice. I am trying to point out that the Tyan is by no means a lesser choice. Both have their perks.

--------------------
www.RTaylor-Design.Com

Are the issues you have been chatting about in relation to PCI audio cards only. IE can I avoid these issues by using a firewire audio interface? Has anyone used a beefy PCIe GPU with a FW audio box on a s939 mobo and not had any issues?

Quote Inoshishi:

---

Are the issues you have been chatting about in relation to PCI audio cards only. IE can I avoid these issues by using a firewire audio interface? Has anyone used a beefy PCIe GPU with a FW audio box on a s939 mobo and not had any issues?

---

Very good question.

I'm not sure what test have been done.

However, I certainly would not assume Firewire or USB will work fine. Firewire and USB may be connected via the PCI bus on some PCIe mobos. On others, even if they are attached directly to the chipset/PCIe switch, the means of attachment is analagous to the way the PCI bus is attached - so the same problems are likely to occurr.


Peter

--------------------
PaQ

Users of RME's Fireface 800 are certainly reporting problems with a selection of motherboards, as it's included on the list linked to earlier in this thread.


Martin

--------------------
YewTreeMagic

Quote Jim Wright:

---

Martin -- In most respects, I think we're violently in agreement It's definitely a confusing time for making PC purchasing choices.

---

I think that makes three of us in violent agreement, then.

This is certainly a confusing time to be upgrading - but it turns out most of the issues have robust answers, so there are good choices to be made if one thinks it (though like you have).

Quote:

---

However -- I do think it's going to get worse before it gets better.

• PCIe boards will be increasingly common; PCI/AGP boards increasingly scarce. As time passes, PCI/AGP will be viewed more and more as a legacy/low-end item, and high-performance PCI boards will vanish sooner than the commodity-market variety.

---

The big question is this:
Just how quickly will PCI cards disappear.

Last time we had a comperable change was when the PCI bus took over from the ISA bus. Legacy ISA slots hung around for a couple of years. It will be faster this time, I think, because so much stuff is now integrated onto the mobo. Most PCs just have a graphics card and nada in their expansion slots...

Quote:

---

• A mature PCIe audio card is probably a considerable ways off. The technology is relatively new; drivers and protocols for sharing the PCIe bus "fairly" (preserving audio quality, not boosting graphics card scores) will likely take time to work out, and I suspect many vendors will prefer to focus on Firewire (and perhaps USB 2.0) products, where the core technology is much better understood (especially w.r.t to glitch-free, low-latency audio protocols).

---

I (sadly) think you may be right - though not for technical reasons. I think the soundcard manufacturers just don't want to know about PCIe.

PCIe does not intrinsically favour the graphics card (fairly or unfairly). We have no specific reason to believe that there will be any problems if you simply take a PCI soundcard, add a PCI<->PCIe bridge and plug it in to a PCIe slot. After all, that's how some of the first generation PCIe graphics cards were made - so we know the bridge chip exists and is cheap. That may not be the best way to do the job; but it has to be a quick cheap way of getting the first prototype built to start testing.

Quote:

---

Considering both points, it seems likely that within 6-12 months, it will be a lot easier to put together a system around Firewire (or perhaps USB 2.0) peripherals than around PCI/PCIe audio peripherals. If one is content to use Firewire, it does make sense to wait some time for dual-core processors to become cheaper, for BIOS/motherboard issues to be resolved, and for initial teething pains with PCIe graphics cards to be sorted out.

---

Actually... I disagree here. I think that the PCI mobos (for the latest CPUs) will have gone in 12 months, forcing the soundcard manufacturers to react. Neither Firewire nor USB seem to me to be a terribly clever bus for audio, because they are shared and not designed for low latency operation. The dual core bios issues will fade quickly, not least because the AMD dual-cores have actually been around for nearly a year (as Opterons engineering samples) and the entire A64 range was designed as dual/multi-core from day one.

I don't think the PCIe problem is actually anything to do with the gaphics cards. I also don't think that the problem with PCI soundcards on PCIe mobos will necessarily ever get fixed. I think the game plan is to ditch PCI completely by Christmas (this year).

So if it turns out that genuine PCIe soundcards work fine, I think that's what we'll be using in 6-12 months; but probably some new names will have gotten bigger, and some familiar ones will have gone. Adapt or die.

On the other hand, if genuine PCIe soundcards have the same problems as PCI cards on PCIe mobos, then you are surely right. We'll all be messing with USB and Firewire in a year's time.

And the forums will be full of posts from users complaining about gapping in the audio if they leave their digital camera reader and USB memory stick plugged in.

Quote:

---

However, if one wants to use PCI-based audio cards, I think it makes more sense to buy a system now, rather than in a few months (as long as one buys a system with reasonable upgrade potential).

Looking at the costs of the DAW I'm building... The motherboard is the least expensive item ... even the CPU is only about 20% of the system cost.

For me, it seemed best to invest in the "mature" end of the PCI generation... rather than riding the bleeding edge of the PCIe generation.

---

Absolutely.

The only real uncertainty is PCIe. Eliminate that, make sure you get a mobo that will have an X2 bios, and the downside risk is that you might want to add some card that (by then) is only available in PCIe in a few years time. Something else will have turned up by then forcing a new mobo on you anyway.

If your CPU upgrade to X2 is delayed until next Summer it wouldn't surprise me if you change mobo to socket M2 and DDRII at that point as well... depending, of course on how the PCIe issue plays out.

Quote:

---

And as far as the 64-bit world is concerned -- I know of people running WinXP 64, 64-bit 1820M drivers and Sonar 4 (64 bit beta), with excellent results, on systems very similar to my new DAW. So I'm fairly confident I'll be able to do that as well, perhaps this Fall.

---

The only question about 64-bit is when to make the switch. The important thing is to get a 64-bit capable CPU and choose your moment.

Quote:

---

... and it may take a while for drivers and software to get sorted for dual-core operation (I'm thinking of many products that had issues with HyperThreading, but perhaps dual-core is less disruptive...)

---

I believe dual core will prove entirely benign in this respect. An X2 should (and apparently) does look just like a dual Opteron to the software (though apparently a bit faster).

Hyperthreading was always a bit of an oddball because some aspects of the execution unit are shared between the two virtual execution units; and also because competition between the virtual CPUs was not subject to task priority.




Peter

--------------------
PaQ

Thanks Martin, I didn't notice the Fireface was on the RME hompage listed as having problems. I did also think at the back of my mind (As the other poster mentioned) that the way mobo's were engineered meant that the usb/FW sockets somehow funnelled into the PCI bus.

Is it possible that the North/South bridge (The one not to the GPU) on some NF4 mobo's might be implemented in such a way that this does not happen?

I like RME's advice to use a low power GPU. Why would anyone want to use NF4 other than to fit a beast of a GPU? It does have some features that NF3 doesn't but surely that's the main one. I've just bought a 24" 1920x1200 dell monitor and theres no way anything less than a 7800 gtx is going to keep those pixels busy for game playing.

I can't believe we're having these problems again.

Oops, just switched personalities. Must organise my forum logins so I don't appear Schizophrenic!!

Quote Bryan Blessed:

---

...
Is it possible that the North/South bridge (The one not to the GPU) on some NF4 mobo's might be implemented in such a way that this does not happen?

---

It's certainly possible. I suspect the dual CPU Tyan S2895 that Rick likes would also work fine with Firewire audio. But I also suspect the Tyan ($500) is the only mobo where this is true, currently.

Quote:

---

I like RME's advice to use a low power GPU. Why would anyone want to use NF4 other than to fit a beast of a GPU? It does have some features that NF3 doesn't but surely that's the main one... game playing.

---

Indeed. Of course, PCIe was not invented just for graphics cards. It's supposed to fix all the problems with PCI - lack of bandwidth, no scaling, poor sharing, poor low latency opertation etc. But the first use has been for graphics cards, and until all the other stuff we used PCI for has converted to PCIe the only real use for PCIe is for mobster graphics cards.

And anyway, fitting a low power GPU does not fix the soundcard problem on nF4. It circumvents the problem on Intel PCIe chipsets, but even that's just a circumvention, not a robust fix.




Peter

--------------------
PaQ

Good stuff Peter,

Thinking about the bigger picture these incompatibilities are a bit ridiculous.

As far as I can tell NF4 is the mainstream for PC buyers who are likely to be interested in audio (IE want fast, good value PC's which are expandable) Basically what RME are saying is : if you want to use your PC for audio with our products buy an outdated PC which won't be that expandable in future and may not do the other things you may, quite legitmately, want/expect it to do now.

I have no beef with RME. They are the best sound card company i've dealt with and their products/drivers are as good as they can get them. Also they seem to be alone in being up front about these issues.

But what about the "pro" sound card makers as a whole. Shouldn't they have seen this coming? Did no-one tell them about PCIe?

I don't think its acceptable for them to say "Sorry, buy NF3/non PCIe". They must make products that are rock solid when working with it.

Edited by Bryan Blessed (24/06/05 11:56 AM)

Hi Bryan,


I completely agree...

The entire PC audio industry seems to be in a state of denial abouit PCIe.

Everybody told them. They just don't want to know (it seems).

I did ask RME, and their technical response is that "PCIe sucks". It seems they think that the problems with PCI soundcards on PCIe mobos necessarily tell us something about genuine PCIe soundcards running native on PCIe mobos.

If you read what they say on their site it's clear whoever wrote it fundamentally mis-understood how PCIe works. Their e-mail to me confirmes this.

They quote the problems with PCI cards on PCIe as a reason not to develop PCIe soundcards at all; effectively asserting that PCIe itself is the problem. I take the opposite view i.e. that PCI support under PCIe is the problem, and this makes it imperative to develop a PCIe card.

If they are right they'll save a bit on R&D. If I'm right they will vacate the entire PCIe soundcard market to some newcomer.




Peter

--------------------
PaQ

What about more workstation chipsets like the nForce Pro 2200, as used by this Supermicro board:

http://www.acmemicro.com/estore/merchant.ihtml?pid=2015&step=4

I know it's not cheap, but if it works wouldn't it be worth it?

*Falls over self with disbelief about emerging consensus

PCIe is a fact that sound card makers must deal with. If they don't like it they can, as you say, lump it.

The sound card maker with with the most muscle is Creative but even they are small pathetic & insignificant fry when ranged against Intel, ATI, NVidia, AMD etc all of whom have decided that the future is PCIe.

Not sure where this leaves me. Most likely buying a good NF4 system now and waiting an annoying while for a card maker to come up with a product that works properly with it.

A Tyan based system would be overkill/too pricey for me.

Quote csl:

---

What about more workstation chipsets like the nForce Pro 2200, as used by this Supermicro board:

http://www.acmemicro.com/estore/merchant.ihtml?pid=2015&step=4

I know it's not cheap, but if it works wouldn't it be worth it?

---

Well... the Tyan S2895 apparently works, so this one might as well. The rteal propblem is price - and it's the higher price of the socket 940 CPUs as well as the mobo cost. But Rick' line throughout this thread is that it's worth the cost, whereas my basic arguement is that it isn't, and you shold just stick with nF3.



Peter

--------------------
PaQ

Hi Bryan,

I simply see no need for nF4 on a DAW at present. There are no PCIe audio cards of any sort yet.

So for now I believe nF3 is the way to go - anf the manufacturers will simply have to react when the mopbhos supply dries up.


I suspect that behind their PR wall of denial there is a bit more happening, and that sometime early next year...




Peter

--------------------
PaQ

Quote Bryan Blessed:

---

A Tyan based system would be overkill/too pricey for me.

---

No such thing as overkill. In a year it won't be.

The price isn't really that bad... you can buy different boards for considerably less. {Onboard graphics...}

This is linked in a roundabout way from the RME pages. I can't seem to get the database to come up {whether that's my firewall or not...} It does give a decent idea of relative performance. {Ok... so it's a trifle off-topic.}

http://www.informationarchitecture.it/cubasetest/results.php?order=perform ance

--------------------
www.RTaylor-Design.Com

Quote:

---

They quote the problems with PCI cards on PCIe as a reason not to develop PCIe soundcards at all;

---

Uh? Look what I found on the Dell website in a white paper on pci-e:

Quote:

---

Support for Real-Time Data Traffic

Unlike PCI, PCI Express includes native support for isochronous (or time-dependent) data transfers and various QoS levels. These features are implemented via "virtual channels" that are designed to guarantee that particular data packets arrive at their destination in a given period of time. PCI Express supports multiple isochronous virtual channels&#8212;each an independent communications session&#8212;per lane. Each channel may have a different QoS level. This end-to-end solution is designed for applications that require real-time delivery such as real-time voice and video.

---

This looks like a godsend if designing near-real-time systems

--------------------
Expert in non-working solutions

Edited by Havoc (24/06/05 04:13 PM)

Ah, for some reason I thought the Tyan board used the nForce4 chipset, not the Pro. Whoops...

I think the pro is just the nF4 re-branded for workstations.


Peter

--------------------
PaQ

Hi Havoc,

Quote Havoc:

---

...

Support for Real-Time Data Traffic

Unlike PCI, PCI Express includes native support for isochronous (or time-dependent) data transfers and various QoS levels. ... Each channel may have a different QoS level. This end-to-end solution is designed for applications that require real-time delivery such as real-time voice and video.

This looks like a godsend if designing near-real-time systems

---

It also suggests the problem might be that nF4 (either per se or because of mobo design decisions) has too low a QoS for PCI. Maybe the QoS is set at a level which simply meets the "official" obligations of PCI - almost certainly not nearly good enough for a pro soundcard to work.



Peter

--------------------
PaQ