PCI Express - Coming Soon to a PC Near You

Max Romantschuk writes "I've been following the emerging of PCI Express for some time now. PCI Express, previously known as "Third Generation I/O" or "3GIO", is the technology set to replace PCI. PCI has been with us for around ten years now, and is rapidly running out of bandwidth. Last week Anandtech ran an interresting story on PCI Express. The techology has previously been covered by Hexus and ExtremeTech aswell. I feel this technology looks all set to replace PCI, and we really do need some new bus technology to keep up with the bandwidth demands of today's applications. Or is this just yet another way to force us into a new upgrade cycle?"

Comments

can't please everybody I guess

[Anonymous Coward]

I feel this technology looks all set to replace PCI, and we really do need some new bus technology to keep up with the bandwidth demands of today's applications. Or is this just yet another way to force us into a new upgrade cycle?

Damned if you do, damned if you don't.

Re:can't please everybody I guess

[aborchers]

Or is this just yet another way to force us into a new upgrade cycle?

I suspect this will be a long attrition as it was with phasing ISA out of motherboards.

From the Anandtech article:

With the advent of RAID arrays, Gigabit Ethernet and other high bandwidth devices on consumer class systems, PCI's 133MB/s available bandwidth is clearly insufficient to handle these demands.

So, for many users PCI-Express will not be a necessity because the unwashed masses are by and large not on the cutting edge of the sort of technology that demands it. The early adapters will drive the market and the rest will follow along when there is a pressing need or when the industry drives us there, e.g. when we can't easily get motherboards or cards without it. I am a fairly cutting edge user and it has only been in the last couple years I switched off the last machine at my house with an ISA slot.

If, as stated elsewhere, this will outperform AGP, then hardcore gamers certainly will go there, but who needs to force an upgrade on them anyway? :-)

Does PCI Express solve the shared IRQ problem?

[WeiszNet]

More than bandwidth, what I need would be a bus
that doesn't have a problem with too many extensions
because of a limited number of IRQs.

Today most mainboard come with many onboard PCI componentes. If you really are going to put in 3-5 extra PCI components in a stock PC, you usually end up in a nice game of 'let's see what order works best', or cannot use all cards together at all.

Is this what the consumers want or...

[joshwa1]

Is it what the manufacturers think we want? The traditional Hard Drive is still the main componant in the PC slowing everything down, yet the manufacturers still keep increasing CPU, and BUS speeds and increase noise and heat levels.

Re:Is this what the consumers want or...

[confused one]

you can go out and get a fast drive array (I think 2 scsi u160 drives with 10k spindle speeds in a raid 0 array is enough), which will actually saturate the 32bit pci bus. If your looking for a fast disk, the technology is available. It's not cheap; but, it's available.

So, at some point, even if you develop faster drives you're going to need a faster bus to support it.

Re:Is this what the consumers want or...

[dissy]

> Is it what the manufacturers think we want? The traditional Hard Drive is still
> the main componant in the PC slowing everything down

That statement is false.
It is indeed the bus that is the slowest bottleneck at the moment.

Go out and get a firewire controller.
Right there you have almost 3 times the bandwidth of a SCSI controller, and 4 times the bandwidth of an ata100 IDE controller.

If you want full disk IO bandwidth, you can only use ONE ide disk per bus. For scsi its actually the same. Technically you can fit 1.6 disks on a bus, but since we are talking about not sharing bandwidth, you can only put one disk on it, because there is not enough bandwidth left over to speak to a 2nd disk at full speed. Yet people commonly throw 14 disks on a scsi chain.
Firewire can support 4 disks at full speed, so using the same math one can throw 52 disks on the bus and be equal with a 160mbit scsi bus.

I have 4 such firewire buses on my system.

PCI is what is limiting me from using ONE of them to its fullest, let alone all 4.

And this is firewire 400.. 800 is out as well, which even ONE card can saturate a PCI bus even with no other devices on the PCI bus at the same time.

Standard PCI also cant fully use Gigabit ethernet to its potential, let alone multiple cards like we do with 100mbit ethernet today.

All of the above apps will use over 100% of the PCI bus *with just one of them in use!*

So, now lets say you want 4 firewire 800 bus's, 2 IDE buses, 2 gigabit ethernet cards.. wow you have just exceeded your PCI bus's capabilitys by over 6 times!

Yes, it is safe to say PCI is our main problem right now.

Also as a nit pick
> and increase noise and heat levels.

manufacturers do this? Concidering the only parts in my systemsthat make ANY noise (outside of sound hardware of course) are fans.
Nothing else in the computer really makes alot of noise.
Disks make some sound yes, but very little. And they are definatly drowned out by the sound of the fans. I cant think of any other devices in my computers that DO make noise except for fans and drives with moving parts now that i think about it.

But comparing a fan design flaw with 'CPU RAM and BUS' technology is like bitching about the color of paint on a car and claiming that effects performance and thus is the car manufacturers problem.

Granted its a pain that parts keep getting faster at the expence of generating more heat, but thats a law of physics so really hard to work with there :)
Ideally we need a new cooling system for PCs in general to replace the fan.

About time

[zensonic]

Given that the PCI interface was introduced to the world by intel in 1992 and that we since have increased the cpu processing powers by a hundred fold (give or take a little) it is really about time that the bus catches up.

What upgrade cycle?

[simong_oz]

Or is this just another way to force an upgrade cycle?

It may well be one of the intentions of it, but one thing I don't get is that with CPU speeds and hard disk capacities where they are now, the average computer buyer (which probably is not very well represented on slashdot) no longer really needs to upgrade their computer, so changing interface/slot shape/etc won't really matter to them.

I know I'm generalising, but the only applications that really push today's computers are games (and high end scientific programs, but they're a fairly minor special case) and I would guess that most computers are not used primarily for games (ie. "serious gamers" - think families). Serious gamers will always be upgrading their computer to the latest and greatest anyway - they don't need to be forced into an upgrade cycle.

It's getting to the point now where by the time the average family decides they need to upgrade their computer, it is easier (and maybe even cheaper) to just buy the latest middle-of-the-line computer package.

I'd almost question whether the idea whole idea of upgrading is itself becoming obsolete for an average computer user?

Re:Remember USB?

[Anonymous Coward]

Yup. I just went to look. They have the PDFs avialable, but a password is required. You can sign up for an account, but only if you are an employee of a company which is part of the consortium.

Oh, I'm sure at least a couple of Linux companies will get access to the specs. That isn't a whole lot of help to those of us who are not working on Linux. We either have to wait for the code to be completed and available from Linux (From which we then have to reverse engineer the exact process from) or hunt around until someone leaks the specs and we can snatch them on the quiet.

Don't even bother to flame me and claim that these standards bodies have a right to make money from these; its a damn specification They are funded by the hardware companies who are going to make millions of dollars in upgrades and new hardware alone. Charging a few bucks for the specs is a little on the cheap side.

Re:PCI doesn't need to be replaced

[Anonymous Coward]

I believe that onboard devices are still part of the PCI bus, even though they're not plugged into a PCI slot. So onboard Gigabit ethernet will still eat into your available PCI bandwith.

Also, you don't seem to be looking at individual PCI devices rather than the total bandwith for all devices. Right now if you want more than 2 IDE drives and have them not affect each other, you need multiple IDE controllers. Individually they may fit into the available bandwith fine, but combine several and you can be in trouble.

The same can be said with multiple ethernet cards. One gigabit ethernet card may work fine, but if you want to have multiple, you may have issues.

You're thinking max theoretical

[Demon-Xanth]

The real PCI bandwidth is usually something like 75-90MB/sec. Depending on the chipset.

Now, add in IDE RAID cards, and SCSI cards and those along can saturate the bus. Consider that a single SCSI HD can now pump out about 70MB/sec when used in an STR intensive application.

Re:Say goodbye to legacy crap

[Anonymous Coward]

As long as PCI-Express is completely backwards compatible with all my current PCI cards I have no problems if they change. If it requires a different form factor or is incompatible then don't call it PCI anything or you'll just confuse the customers. I don't know what the big complaint is anyway, computers are more than fast enough for anything we really need to do on a desktop these days. Until Office2005 comes out my 2 GHZ Athlon should be plenty fast enough.

No use. The well has been poisoned.

[JCCyC]

Never again will any announcement of new hardware technology be received by us geeks with the glee it once was. The only thing that comes to our minds now is "great, another opportunity for them to add DRM and phase out hardware that allows copying"

Re:No use. The well has been poisoned.

[Jahf]

Adding DRM to the PCI pipeline would drag down the benchmarks so badly it wouldn't get adopted. DRM needs to go into a layer of the mix where you know that the data is intended to be.

It would be like adding SPAM filtering to the transport layer of a network (ie, encapsulated data that you have no way of knowing what the data is intended for) rather than a higher level where you know it is targetted to be email.

Not saying no one will try it, but the horsepower improvement would have to be immense to cover up the downside. DRM isn't going into the backplane for QUITE some time.

Re:Why?

[dissy]

> What PCI device are you using that is bandwidth limited & will benefit
> from a faster PCI bus?

Gigabit ethernet, soon 10gbit ethernet..
multiple firewire buses, or even one firewire 800 bus..
Multiple high speed graphics cards..
Multiple SCSI or fiberchannel buses..

> I don't have anything.

> I really have nothing that will gain any benefit.

Well thank you for deciding that what you need is exactly what everyone else needs and they should be happy with that :P

Re:PCI doesn't need to be replaced

[cait56]

And 640 KB is more than enough memory for any desktop too,

Try thinking new applications. What if your "desktop" machine is capturing one TV show, downloading a major update to a software application, and your viewing two versions of a video in parallel in order to determine how to further edit the thrid copy that you have open in another window.

And oh yes, you just received 73 wonderful opportunities to engage in financial transactions with a former Nigerian minister.

It seems to me that a "desktop" machine could end up needing quite a bit of internal bandwidth.

Re:cards consumers can install w/o a screwdriver?

[Smallpond]

SCSI and fibre channel drives have used SCA (single connector attach) for years. Telephone and realtime control industries are big on cPCI also. But the reason isn't to make it easy to install hardware, its because the servers run an OS that supports hot plug. The goal is zero downtime, even for maintenance.

Standard PCI is laughable for high availability.

Parity is for farmers -- Seymour Cray

Re:Parallel faster than Serial

[akuma(x86)]

I'm afraid this might add to the confusion about serial interfaces being 'faster' than parallel. While it is true that you don't have to worry about data/clock skew when using serial interfaces, enabling you to clock them faster, a parallel interface running at the same clock speed as a serial interface will always be faster in terms of data throughput. The reason for this is simple: serial == 1 bit per clock, parallel = > 1 bit per clock.

That's the whole point isn't it? Wide and slow or narrow and fast, you still get the same throughput. You can't clock a parallel interface as fast as a serial one, so you shouldn't compare them with the same clock speed.

I am reminded of the whole P4 vs Athlon debates. It's stupid to compare P4 and Athlon at the same frequency or use the stupid "but Athlon does more per clock" arguments. The P4 is designed to run at a higher clock, but can do less in parallel (IPC), but makes up for that with the higher clock.