PCI Express technology backgrounder now live!

16 replies [Last post]
bcrabtree
Offline
Joined: Mar 7 1999

PCI Express technology backgrounder now live!

Enjoy!

Comments most welcome.

Oh, and it's well worth reading in conjunction with our Serial ATA backgrounder, me thinks.

Bob C

bcrabtree
Offline
Joined: Mar 7 1999

Really is live now.

Bob C

bcrabtree
Offline
Joined: Mar 7 1999

Thanks to Kez for the PM pointing out the typo on page 4!

In the editing I'd changed 2.5Gbit/sec to 2.5GHz/sec.

I've now put that right.

Cheers

Bob

byo
Offline
Joined: May 30 2005
Bus bandwidths

Some of figures on the table on the first page must be wrong as they do not add up. The isa bus for example if it had 8MHz and 16 bit would be (8 * 16) / 8 so 16MB/s

The AGP first implimentation row is not right also as the bandwidth you have given is for 1x clock doubling (ie none). If it were 2x it would have a bandwidth of (66*32*2)/8 ie 533MB/s.

Apart from that, intesting article :p

RayL
Offline
Joined: Mar 31 1999

A good, interesting article.

My only comment would be about the way that the article had to be shoe-horned into the Hexus format. Yes, I understand that Hexus has to make money from advertising and I've got used to the big advertisers box that breaks up the opening heading of each page (my eye just makes the jump, rather like fast-forwarding through TV commercials), but the diagram on the final page, for example, (data flows for HD) is part-obscured by the drop-down column. This would never be allowed in a printed magazine and it just looks clumsy on a web page.

Ray Liffen

johnpr98
Offline
Joined: Aug 20 1999

Ray

Doesn't printer friendly layout make things clearer?

Regards

johnpr98
 
If you have any Forum Suggestions please post them here

RayL
Offline
Joined: Mar 31 1999

John

Thank you. I've not had the need to print anything so far and the significance of 'printer friendly layout' as an aid to simply reading the articles was not obvious.

Ray

bcrabtree
Offline
Joined: Mar 7 1999
RayL wrote:
A good, interesting article.

My only comment would be about the way that the article had to be shoe-horned into the Hexus format. Yes, I understand that Hexus has to make money from advertising and I've got used to the big advertisers box that breaks up the opening heading of each page (my eye just makes the jump, rather like fast-forwarding through TV commercials), but the diagram on the final page, for example, (data flows for HD) is part-obscured by the drop-down column. This would never be allowed in a printed magazine and it just looks clumsy on a web page.

Ray Liffen

Ray,

We are still working on ways to refine the appearance - including avoiding problems with the initial ad box (though that isn't easy, because it's only seen on your first few visits to that page, and so some very clever auto-formatting will need to be implemented.

However, I can't replicate the problem you are describing with the Pinnacle diagram on the final page - and if you are seeing it, it's because I haven't seen it at any stage.

What browser are you using and at what screen resolution?

Do please let me know so I can see the problem myself (an email with a screengrab of the problem would be good, too).

Oh, and you are right that the printer-friendly view of articles that John pointed out (thanks John!) isn't an obvious option to choose to make the formatting more human-friendly - we need to think about ways of making this clearer, on the page and in info about the site.

Bob

bcrabtree
Offline
Joined: Mar 7 1999
Bang to rights (in part, though, I think)
byo wrote:
Some of figures on the table on the first page must be wrong as they do not add up. The isa bus for example if it had 8MHz and 16 bit would be (8 * 16) / 8 so 16MB/s

The AGP first implimentation row is not right also as the bandwidth you have given is for 1x clock doubling (ie none). If it were 2x it would have a bandwidth of (66*32*2)/8 ie 533MB/s.

Apart from that, intesting article :p

Thanks very much for the comments.

I don't pretend to be any kind of expert on low-level PC stuff - which is why I leave it to experts such as James Morris.

You are correct in saying there is something awry with the ISA figures.

Having now looked into this (ain't the internet - especially Google and Wikipedia - wonderful?), I can see that the original version of ISA introduced by IBM in 1981 was 8-bit, not 16-bit - so I'll the change the chart shortly but will leave it until I'm sure in my own mind about your observation of the AGP figures because I don't want to have to change it again.

And, concerning AGP, I'm not sure that what you say is correct - though I am sure that there is room for confusion in the chart.

With the AGP figure, you've assumed that you can calculate the bandwidth by multiply up each column to arrive at the bandwidth in bits - after which, as with ISA, you then divide by 8 to move from bits to bytes; and then by 1024 to move from bytes to megabytes.

And that is an understandable assumption.

But, in this instance, I think it may be wrong because (I believe) the author intended the 2x in column four to be highlighting the fact that that the 66MHz figure of the original AGP 1.0 implementation is 2x that of PCI (ie the base AGP rate - is 2x PCI's 33MHz).

If that's right (and I accept that it might not be), then your calc doubles something that, in effect, was already doubled.

I need to confer with James and will report back here when I've done so.

For the record, I would point out that the second-generation of AGP - AGP 2x - also ran at 66MHz, but had a peak transfer rate of 528MByte per second.

This change was achieved by the way AGP 2x transferred data on the rising and falling edges of the clock - so-called double-pumping.

Again, thanks very much for the comments.

Oh, and for you and others who would like to look further into the background of these technologies, here are some useful links (doubtless, google can produce others) from Wikipedia:

ISA

PCI

AGP

Double-pumping

PCI Express

Bob C

cyberwest
Offline
Joined: Sep 13 2000

Thanks for pointing out those two mistakes - I must have lost concentration whilst writing.

Just to clarify, the original IBM PC ISA bus was 8-bit running at 4.77MHz, so had a bandwidth of 4.77Mbytes/sec. But the IBM PC-AT raised this to the 8MHz 16-bit bus I mention, which does of course operate at 16Mbytes/sec. Doh!

The AGP mistake is a bit different. I stupidly shortened two different AGPs into one there - the original spec was for 266Mbytes/sec but the 2x mode was released at the same time I seem to remember, so that mode was always available on all AGP motherboards. So there should be two rows, one for the base spec, and one for the base spec with its 2x clock doubling mode.

Anyone with a long memory can correct me if I'm wrong about 2x being there at the same time as the basic version. I actually broke my ankle in the rush to get to the launch (May 7th 1997). I still hobbled along, but I might have been too delirious to get my facts right!

James Morris

bcrabtree
Offline
Joined: Mar 7 1999
At sea and drowning

Okay, I can follow all this but am concerned why the figures I'm working out are less well rounded than those in the original chart - for reasons unknown to me.

The MByte/sec figs I get are:

ISA, 4.66

ISA Post IBM AT, 15.63 (not 16)

PCI V2.1, 128.9 (not 133)

PCI-X 66MHz, 515.6 (not 533)

PCI-X 100MHz, 781.3 (not 800)

and so it goes on with the rest of the non-PCI Express figures.

Worse, I think I've lost the plot totally when it comes to the figures for PCI Express.

When I check them they, come out wrong by a factor approximating to 10.

Let's start with PCI Express single lane.

This has a frequency of 2.5GHz (2500MHz), and an 8-bit bus.

So, the figures are:

(2500x1000) to get the figure in Hz ie the number per second (which is 2,500,000); then x8 (for the number of bits) which comes out to 20,000,000 bits/second.

Now, we need to divide by 8 to get it in bytes/second (2,500,000) and then by 1024 to get it into megabytes - and that comes out at 2441.4MByte/sec - NOT 250MByte/sec as the chart shows.

The other PCI Express bandwidth numbers seem to me to be out by a factor of 10-ish, too, but I can't figure out why.

James? Anyone?

Bob C

cyberwest
Offline
Joined: Sep 13 2000

Well, it's all a bit confusing, isn't it? PCI isn't really running at 33MHz but 33.3333333MHz. There's also the fact that the 8MHz ISA bus was really running at 8.333333MHz, not 8MHz, so the true figure is about 15.9Mbytes/sec, although most people just round it to 16Mbytes/sec. PCI-X 66MHz runs at 508.6Mbytes/sec if you do the math, but some literature uses the 1Mbyte=1,000,000 bits simplification.

On the PCI Express front, one bit is actually delivered per clock, so 2.5GHz=2.5Gbits/sec, in each direction. Just to really wind you up, PCI Express operates with a 10-bit signal, even though the PC interface is 8-bit. So you can't just multiply the megahertz by 8 to get the true throughput. 250Mbytes/sec is the published spec, but 200Mbytes is the true throughput, because of 10b/8b encoding. Even more confused now?

James Morris

RayL
Offline
Joined: Mar 31 1999
bcrabtree wrote:
Ray,

What browser are you using and at what screen resolution?

Do please let me know so I can see the problem myself (an email with a screengrab of the problem would be good, too).

Bob

Bob,

Mozilla Firefox. 1024 x 768.

I'll send a screen grab by email.

Ray

bcrabtree
Offline
Joined: Mar 7 1999
RayL wrote:
Bob,

Mozilla Firefox. 1024 x 768.

I'll send a screen grab by email.

Ray

Ray,

Thanks.

The site is optimised for 1024x768 and that page can be viewed in full at that resolution, as the screengrab I've emailed you shows.

However, if you have the History sidebar open, and if there are images of 600pixels width on the page, then the problem that you see (images sliding out under the page's RH column) does arise.

Two solutions - close the history sidebar or use the printer-friendly mode.

Bob

DVdoctor
Offline
Joined: Apr 1 1999
Link for Additional AGP History INFORMATION

Hi James
Here is a link for additional AGP information, you are correct 1x an 2x were all part of the version 1 spec
http://www.interfacebus.com/Design_Connector_AGP.html

John

RayL
Offline
Joined: Mar 31 1999
bcrabtree wrote:
Two solutions - close the history sidebar or use the printer-friendly mode.

Bob

Bob

Surely there is a third option - have a design that doesn't have parts of the page sliding over other parts?

Am I not right that most websites resize themselvesto fit the 'active area' that they are allocated? The original reason I mentioned the 'masking' effect was because it is not normal with other websites.

My 'history bar' is actually a list of bookmarks - not something that I would want to keep collapsing - I use it all the time and in all of that list only the Hexus site has this 'menu lists sliding over pictures' effect.

Am I the only one with this problem or am I the small boy who says aloud that the emperor has no clothes?

Ray

PD
Offline
Joined: Aug 6 1999

hi Ray - and to all - :-

RayL wrote:
Surely there is a third option - have a design that doesn't have parts of the page sliding over other parts?

please note that what you see now, is NOT the new HEXUS/DVdoctor site(s), just a very short-term measure to have somewhere more appropriate to present DVdoctor content.

cheers,

PD

HEXUS.swankyDynamicSignature - Give it a click!