Talk:PCI Express/Archive 2008

Low Profile

I see no mention in the article of "low profile" PCI Express slots. What does this mean? 12.155.246.10 (talk) 21:43, 8 January 2008 (UTC)

There is no such thing as a low profile slot, only cases that are too small for full-height cards. For those cases, there are low-profile cards. Low-profile AGP, PCI, and PCI-Express cards all exist. — Aluvus t/c 23:23, 8 January 2008 (UTC)

"vast majority of new computers PCI Express capable"

I remove the sentense

The vast majority of new computers from both Intel and AMD are PCI Express capable.

from the article. A visit to the website of my local computer store tells me that this is not yet the case. Thue | talk 8 July 2005 12:59 (UTC)

Um, your local computer store has nowhere near the volume of an OEM such as Dell, HP, etc. (Local shops also tend to lag on newer products. Kind of like how walmart STILL sells nVidia 5200s.) Hence, using it as the basis for PCI Express marketshare is totally baseless. Back on topic, I do believe all the OEMs have pretty much made the switch. So, unless you or anyone else can give a link or show otherwise, that statement should be re-inserted.the1physicist 8 July 2005 17:32 (UTC)

Well, this is rather seriously offtopic for a page about PCI-express, but I don't consider Wal-Mart to be a choice place for parts of any sort, especially for computers; it's more in the league of an office store: emergency parts only. If you're referring to Nvidia FX5200 cards, they're perfectly capable for 99.5% of machines anyway, so nobody cares. Fsiler 03:36, 5 November 2005 (UTC)

My local computer shop has quite a high throughput, so it is not _that_ out-dated. Anyway, I checked out Dell and HP's websites. A survey of 6 Dell PCs showed that all of them had PCI-ex graphic card support. A survey of 6 HP PCs (their HP pavillion line) showed that half of them had PCI graphics card support. NONE of the computers had any other PCI-ex slots apart from the graphics one! Based on this quick survey I will agree that PCI-ex seems to be taking over the graphics slot, but not even starting to replace standard PCI slots yet; I would not call a motherboard unqualified "PCI express capable" with only one PCI-ex slot which is taken by the graphics card. The 75% of motherboard which had a graphics card PCI-ex slot also does not justify the formulation "vast majority", even ignoring that they had no generel-purpose PCI-ex slots. Thue | talk 9 July 2005 06:34 (UTC)

Ok, here's where your argument breaks down. A motherboard does _not_ have to be fully PCI Express to be considered PCI Express capable. That statement is true even if it's mostly just for graphics right now. And actually, most PCI-E motherboards *do* have one or two x1 slots, even if they're not being used. Anyhow, I suppose I retract the word "vast", but the majority of new computers do have some form of PCI Express.the1physicist 9 July 2005 07:37 (UTC)

In my book it is misleading to say that a motherboard is "PCI express capable" when only the graphics slots are PCI express; reading that I would expect them to have at least one free generel-purpose PCIe slot. None of the 12 systems I checked out had any 1x PCIe slots! We can however say in the article that "the majority of systems seems to have a PCIe slot for graphics". Thue | talk 9 July 2005 07:44 (UTC)

Alrighty then. You do know that you can plug a x1 card in to a 16X slot, right? Not to mention that most motherboards with 16X slots usually have 1-2 1x slots. You seem to have gotten unlucky or something.the1physicist 9 July 2005 10:07 (UTC)

Seems like this topic has long since died, but Intel and AMD don't make computers. Elgordon (talk) 18:19, 8 February 2008 (UTC)

No they don't but intel at least make the dominant motherboard chipsets for thier processors. They also make the reference designs for the moetherboards using their chipset. With intel ati and nvidia behind pci express no competitor stands much chance. Plugwash (talk) 18:58, 8 February 2008 (UTC)

Performance hit from using a 1.0/1.1 slot with a 2.0 card

If I use a 2.0 card in a 1.0/1.1 slot what sort of performance degredation will there be compared to the stated benchmarks with it running in a 2.0 slot?

E.g 20% 50% 90% 92.0.43.124 (talk) 02:35, 22 February 2008 (UTC)

• If your card actually uses more bandwidth (per lane) than was available in the 1.x standard: 0 to 50%.
• If your card is a product that currently exists: 0%.

— Aluvus t/c 03:01, 22 February 2008 (UTC)

"If your card is a product that currently exists: 0%" Do you mean by this that current PCIe 2.0 cards are still running at the same bandwidth as 1.x?

Seems wierd to make a 2.0 then not take advantage of it —Preceding unsigned comment added by 92.0.43.124 (talk) 20:57, 22 February 2008 (UTC)

It's not quite that simple, lets say I make a card that requires 2 GB/s to get full performance. If I make it an 1.0 slot card then to get full performance the user must put it in a slot with at least 8 lanes wired. If I make it a 2.0 card then the user can get full performance by putting it in a 1.0 slot with at least 8 lanes wired OR by putting it in a 2.0 slot with at least 4 lanes wired.

Also it's nice to have some headroom so that when I manage the increase the clock speed of the core of my card I don't have to reengineer the PCIe interface as well.

Finally where the bottlenecks are depends a lot on how the software is using the card. You may get a card that performs better on 2.0 in some extreme cases but performs the same on 1.0 and 2.0 in most benchmarks. Plugwash (talk) 21:21, 22 February 2008 (UTC)

Oh dear my brain has exploded --- 92.0.43.124 (talk) 21:36, 22 February 2008 (UTC)

So using a 3850 in a 1.1 slot for instance leads to little/ no performace loss then?

PIPE != PCIe PHY

"The PCIe physical Layer interface is known by the acronym PIPE which stands for Physical Interface for PCI Express."

This statement is false, I think. PIPE is a specification from Intel, intended to ensure compatibility between different core logic providers. A such it really only specifies an interface between a MAC layer and a PCS layer. The PCIe Base Specification only defines PHY logical and PHY electrical layers; the electrical sublayer encompasses the serdes and the logic sublayer encompasses the MAC and PCS. The MAC is named after, and analogous to, the Media Access Control layer of Ethernet, while the PCS is named after, and analogous to, the Physical Coding Sublayer of Ethernet.

PIPE is nowhere defined in the PCIe Base Specification and is purely an implementation option. Rsmoore (talk) 20:39, 17 March 2008 (UTC)

External PCIe

I'd love to see more information on this. There are a few professional video products using external PCIe. The Blackmagic Multibridge (http://blackmagic-design.com/products/multibridge/) and the CalDigit HDPro (http://www.caldigit.com/HDPro.asp). I'd prefer someone else write it up.76.91.63.151 (talk) 07:04, 25 March 2008 (UTC)

I don't belive there are any official standards for it yet, Afaict all the external PCIe products on the market use thier own propietry connection schmes for the external cable and use an adaptor card to connect to a PCI express slot. Plugwash (talk) 09:46, 25 March 2008 (UTC)

Speed Negotiation

I heard that a given slot is only required to support its own number of logical lanes, and 1 Lane-- but nothing in between. For instance, if you have a 16x slot and put a 8x card into it, there is no guarantee it will link at 8x, it very likely will link at 1x.

I'm looking to confirm/deny this and find some way to mention it in the article as it can be a big issue for AV and throughput intensive apps. —Preceding unsigned comment added by 12.106.16.98 (talk • contribs) 16:08, January 2, 2007 (UTC)

There's two category of terms that are used - short-wiring/short-plugging/down-plugging/up-plugging, which is physically putting a different card into another size slot, and down-shifting, which is using a larger slot with less lanes hooked up (mainly a x8 port with four lanes).

The spec requires that all slots support x1 and native, and are allowed to support intermediate speeds/lanes. E.g., the x16 card has: x16 (required), x8 (allowed), x4 (allowed), x1 (required). However, the spec does say that all up-plugging is "fully allowed" so I wouldn't say x1 speed is "very likely." Also, x8 cards must be able to support x4 operation. There are pins in the spec to negotiate speed after lanes 0, 3, 7, and 15 (x1, x4, x8, and x16).

Down-shifting was removed from the spec and is now up to the OEM's implementation.

some useful URLs: [1] [2] — Dsm (talk) 00:17, 10 April 2008 (UTC)

PCI-E Throughput

I'm getting conflicting information. The wikipedia article states pci-e's 1 bit serial connection, or 1 lane, transfers at a rate of 250 MB/s bidirectional. However, the actual pci-e specification by the PCI-SIG state that the speed for pci-e 1.1 is 2.5 GT/s (Giga-transfers). Which is transfers of 2.5 gb/s*1024 (to bring it to mega bits) which = 2560 Mb/s divided by 8 (to bring to mega Bytes) and so = 320 MB/s. However, since pci-e in reality transfers 10 bits per 8 bits of actual data we then need to figure 80% of that 320 MB/s is actually the transfer speed we should be quoting, which ends up being 320*0.80 = 256 MB/s. Thus the throughput for pci-e 1.1 is 256 MB/s, not 250 MB/s currently written in the article. And since pci-e 2.0 boost the throughput for each lane, we get 512 MB/s in the pci-e standard. Or at least from what i understand. Can anyone verify this? I'd like someone else to verify it to make sure before its added to the main article. Jarrod1937 (talk) 23:14, 8 April 2008 (UTC)

You're right. It's 2.5 GHz with 1 bit/cycle and 10-bit bytes (half duplex). Total bandwidth for one lane is 5GHz, or 5Gbps, or .5GBps (=512MBps). However, the numbers are often quoted using SI units, so .5GBps = 500MBps (250MBps half-duplex).

As an example, see [3]. — Dsm (talk) 00:26, 10 April 2008 (UTC)

We're talking about telecommunication/data rates here, not memory sizes - so the prefix G means decimal billion (10^9). 2.5 Gb/s /10*8 /8 = 250 GB/s. Zac67 (talk) 11:30, 10 April 2008 (UTC)

In this article I put the figures after low level fixed overhead but before high level variable overheads. This gives the figure that is most comparable to the ones we quote for PCI and similar standards. As Zac67 says datarates traditionally use SI units not binary units (yes this is confusing). To get the figure in the article multiply the gigatransfers per second by 1000 to convert to megatransfers per second and divide by 10 to get megabytes per second allowing for 8b10b overhead. Plugwash (talk) 17:26, 10 April 2008 (UTC)

Ok, thanks for the clarification. So, to make sure, only data storage is measured using binary units, everything else including data tranfers use SI? Jarrod1937 (talk) 21:15, 10 April 2008 (UTC)

Unfortunately it's not that easy. RAM, Flash etc. sizes are usually in binary units (due to the rectangle layout of the chips, side length of powers of 2); mass storage (ie. HDD) varies: MS Windows reports binary units, manufacturers usually use SI units (larger figures). See binary prefix for details. Zac67 (talk) 09:31, 12 April 2008 (UTC)

AER

This article does not mention AER whatever that is. -- Frap (talk) 19:56, 25 April 2008 (UTC)

Advanced Error Reporting? — Aluvus t/c 16:37, 26 April 2008 (UTC)

PCIe is supported primarily by Intel,

I'm running an AMD Opteron Server board right now that has two PCIe slots, the other MBs I was looking at had them also. I'm not sure how the intel support line is accurate anymore. —Preceding unsigned comment added by 76.185.6.18 (talk) 07:01, 27 April 2008 (UTC)

This article does not mention PCIe x12 with a speed of 3000 MB/s (3 GB/s)

Source: Page 13 of Chapter 1. Book: "CompTIA A+ Complete Study Guide" - Sybex - Authors: Quentin Docter, Emmett Dulaney and Toby Skandier. —Preceding unsigned comment added by 206.116.26.91 (talk) 23:13, 27 May 2008 (UTC)

8b/10b derating

The article gives lane speed of 250MB/s, which already includes overhead from 8b/10b coding, and then says "However the data rates cited must be derated because 8b/10b coding". —Preceding unsigned comment added by 195.237.242.93 (talk) 17:18, 4 June 2008 (UTC)

graphic card

sir, i hve a computer of 3gbram,core 2 quad,s975xbx2, and i work on m0delling softwares which is the best graphic card do you suggest —Preceding unsigned comment added by 122.167.108.95 (talk) 15:21, 9 June 2008 (UTC)

Would it be possible also to use a few real world examples. For me I am still unsure but i believe that a new graphics card like an ATI HD4850, HD3870 or Nvidia 9800GTX or 8800GT or 8800GTS will NOT be bottlenecked if it is put into an Intel P35 motherboard for example my GA-P35-DS3R which is not PCI-E 2.0, Is that correct? My understanding is that PCI-E 1.0/1.1 still has a great deal of head room. Am I wrong about that? And will the newest upcoming lines like the HD4870 & Nvidia GTX280 come close to filling the bandwidth and bottlenecking? Thankyou. —Preceding unsigned comment added by 97.90.235.129 (talk) 13:27, 21 June 2008 (UTC)

Okaay...

Look at this section

PCI-SIG announced the availability of the PCI Express Base 2.0 specification on 15 January 2007.[8] PCIe 2.0 doubles the bus standard's bandwidth from 0.25 GByte/s to 0.5 GByte/s, meaning a x32 connector can transfer data at up to 16 GByte/s in each direction.

In June 2007 Intel released the specification of the P35 chipset which does not support PCIe 2.0 only PCIe 1.1.[10] Some people may be confused by the P35 block diagram[11] which states the Intel P35 has a PCIe x16 graphics link (8 GB/s)

So...your saying that 1.1 is x16 w/ 8gb and 2.0 is x32 w/ 16 gb ? lol, I never even heard of a x32 connector before, correct me if im wrong please. —Preceding unsigned comment added by 74.57.98.234 (talk) 04:33, 24 July 2008 (UTC)

The important words you are missing are in each direction. Afaict many manufacturers like to inflate thier specs by quoting the aggregate data rate of both directions rather than the data rate in each direction (just like in the early days of full duplex fast ethernet you often saw full duplex fast ethernet products advertised as 200 megabit per second).

As for x32 I'm pretty sure it is allowed by the specs for all versions but i've never heared of one being seen in the wild. Plugwash (talk) 22:29, 27 July 2008 (UTC)

PCI Express 2.0 Specification

Does anyone have a soild source regarding the upcoming PCI Express 2.0 specification? Most notably the expected final release date and the added hardware support for virtualization in this new standard?

I'm missing information on the PCIE 1.0a spec. What is different for 1.0a?

What is newer, 1.0 or 1.0a? What does it mean for a graphics card to be 1.0a compliant? —Preceding unsigned comment added by 83.67.94.68 (talk • contribs) 10:06 May 9, 2006 (UTC)

I would like to know to.... 1.0, 1.0a, 1.1 ...... No information are given about the difference of them. And I would like to know, witch chipset and/or VGA do use witch, becouse:

“ In some cases it is possible that a PCI-E 2.0 card will not work correctly on a PCI-E 1.0a slot. This is only limited to certain video cards. ”

Placi1982 (talk) 14:57, 23 August 2008 (UTC)

xx

xxx70.30.24.159 (talk) 03:17, 2 October 2008 (UTC)

Ok, I've corrected some past time things written in future tense, such release dates, and "will be" parts that by this date are "was" or "where". 201.247.28.25 (talk) 05:09, 3 November 2008 (UTC)

Lead section

until march of this year this article had an oversized intro (6 paragraphs though some were very short paragraphs). Kellycook then reduced this down to a single paragraph which is far below the guidelines and WP:LEAD and IMO is nowhere near enough to do justice to an article of this size. I have taken the old intro and tried to cut it down to what is most important which is mainly

• What it is (a PC expansion interface)
• What that x means and how it impacts compatibility of cards and slots
• How it compares to it's predecessors

-- Plugwash (talk) 13:27, 12 November 2008 (UTC)

summary and overview

I have merged theese two sections, they both seem to be saying pretty much the same things in different wordings. -- Plugwash (talk) 13:28, 12 November 2008 (UTC)

Compatibility between x16 (card) and x8 (slot)

Is it possible to plug a (long) PCI-E x16 card into a (short) x8 slot? The socket is open-ended so the card physically fits, but overhangs. As this appears to work (yes, did it last night!) then presumably this is some deliberately supported "fall-back" position where it simply drops back to using x8 lanes rather than 16.

However I note that the article says, ""A PCIe card will fit into a slot of its physical size or bigger, but not into a smaller PCIe slot." This is clearly untrue in one aspect (it will definitely fit, as it's an open-ended socket), but should such a plugging be made? Will it "work" (at reduced bandwidth), will it "bodge" (at terrible bandwidth and error prone) or am I likely to let the smoke out of the motherboard altogether? Thanks for any clarifications that can be offered. Andy Dingley (talk) 11:52, 28 November 2008 (UTC)

Strike that, there's a clarification buried in the body text of the article. I'll re-word the Overview to clarify. Andy Dingley (talk) 11:56, 28 November 2008 (UTC)

You've made me curious... could you please point it out?

Em27 (talk) 07:01, 24 December 2008 (UTC)