Talk:PCI Express/Archive 2012

Latest comment: 11 years ago by Diblidabliduu in topic Transfer Speeds


Original research

This appears to be a useful and comprehensive article for the most part. However, some statements, for example:

"PCIe 3.0 or 4.0 can't reasonably displace HTX on AMD boards, and is less likely to reliably support MIPS"

"Accordingly intel is extremely unlikely to invest in any such extensions."

are jarringly speculative and editorial in nature.216.67.48.226 (talk) 02:41, 6 July 2012 (UTC)

Agreed. That entire section is rather poor. There is also the nonsensical portion:

PCI-SIG, the copper wire extension of PCI outside the box, [5], competes more with fiber alternatives but seems also to directly compete with Thunderbolt_(interface).

which seems to conflate PCI-SIG (the standards group), with PCI-SIG's planned OCuLink standard (which is not just copper wire). It would be much better to move that into an "external PCIe" or "OCuLink" section, then scrap much of the "PCIe 3.0 vs. HTX, Thunderbolt, USB 3.0" section, which is poorly written, speculative, and at times self-contradictory (at one point, it implies: that PCIe 3.0 is not widely used; that PCIe is unlikely to compete with HTX (or just HT?); that HTX is not widely used and thus not viable [nulling the second assertion].)
50.92.223.38 (talk) 06:51, 29 July 2012 (UTC)

Fiber centric

The section PCI Express#Extensions and future directions is fiber centric. It is also speculative: PCIe would be good for datacenters, but then it says "as of 2012 there are no major vendors offering solutions in this vein". I'm tempted to delete the entire section. Glrx (talk) 03:15, 22 September 2012 (UTC)

current status

PCI Express dominates the add-in card market. PCI Express was originally only common in disk array controllers, onboard gigabit Ethernet, Wi-Fi and graphics cards. Most sound cards, TV/capture-cards, modems, serial port/USB/FireWire cards, network/Wi-Fi cards that would have used the conventional PCI in the past have moved to PCI Express ×8, ×4, or ×1. While some motherboards have conventional PCI slots, these are primarily for legacy cards and are being phased out.

This seams to be orginal research, PCIe is replacing PCI, but slowly and in many areas, PCI is still dominant. e.g. on this site, which is a very good overview of the european (especiall german and austrian) market, there are 26 PCIe soundcards, 64 PCI soundcards and 4 with an "unknown" interface. The paragraph above also contains no references, so I remove it. --MrBurns (talk)

I agree with the removal since it was unsourced. However I would note that the site doesn't seem to provide much useful info. The number of products in a certain form factor doesn't tell us anything about how many are sold, it could easily be PCI express dominates even if there are fewer products. (And really the sound card market is not a great one to look at, it was dying before PCI express came around and is even more dead now.) Nil Einne (talk) 06:42, 23 November 2012 (UTC)

full-duplex vs. dual-simplex

The article says that PCIe communication is full-duplex. I think this may be technically incorrect. The PCIe specification refers to the link as dual-simplex, not full-duplex. From section 1.2 titled "PCI Express Link": "A Link represents a dual-simplex communications channel between to components." (Please note that this is from "PCI Express 2.0 Base Specification Revision 0.9". Hopefully others can verify that this statement appears in newer versions of the specification.) While functionally equivalent, I believe full-duplex and dual-simplex differ at the physical/electrical level.

Here is an article that describes the difference between full-duplex and dual-simplex: http://www.pcmag.com/article2/0,2817,1152236,00.asp "A dual simplex connection permits data to be transferred in both directions simultaneously, similar to full duplex connections (as in telephones), but with dual simplex, each wire pair has its own ground unlike full duplex, which uses a common ground."

Arakageeta (talk) 18:28, 29 October 2012 (UTC)

I obtained a copy of the latest PCIe 3.0 spec, and section 1.2 titled "PCI Express Link" still states that PCIe is dual-simplex, not full-duplex: "A Link represented a dual-simplex communications channel between two components." (PCI Express Base Specification, Revision 3.0; Nov. 10, 2010).

Arakageeta (talk) 18:09, 2 November 2012 (UTC)

"Simplex" and "dual-simplex" are just ITU jargon for what is commonly known as "half duplex" and "full duplex". And I thought the links do share a common ground... Zac67 (talk) 19:57, 2 November 2012 (UTC)

Power options

I made these changes [1]. As I mentioned in the hidden comment, the claim that PCI express 2.x adds support for 150W per slot seems to be common in some sources, particularly those from around the time of release. However a number of more recent sources either directly dispute this, or don't mention any such change. This includes some which refer to presentations from PCI-sig. The source used there was fairly poor. What it basically said was the card probably uses more then 75W and it doesn't have a 6 pin connector so it must be relying on the (alleged) ability of the PCI express 2.x slot to provide 150W. Except that they didn't actual try to estimate power consumption of the card and when they did measure power consumption under load it was with Furmark, and evidence suggests both GPU and card manufacturers consider Furmark may impose power usage requirements beyond what they spec for. (Estimates and official specs for the 6850 suggest power consumption was high enough it's unlikely the card is under 75W but you can never know without testing.) More importantly, there's no actual evidence to suggest the card is compliant with the spec. Perhaps the manufacturer felt enough motherboards would work without issue even if they were exceeding the spec or perhaps the manufacturer was similarly confused by the rumours. Either way the reference didn't come across as particularly reliable.

As I mentioned, there are quite a few references including those using presentation or other info from PCI-sig suggesting that the 75W per slot limitation remains in 2.x and even the 3.0 spec. While ORry, I've had a look at this info myself and even had a quick check at the PCI express spec (doesn't ask how I got it) and couldn't find any info of a 150W slot standard. I'm not sure where the 150W per slot claim came from, I've seen some suggestion there were plans for 150W per slot initially but this was abandoned (and perhaps considered again for 3 but again abandoned.) Alternatively, it's also notable that occasionally PCI-sig refer to power per slot when they actually mean the maximum power a single card can draw from all sources incluing the extra connectors and not solely from the PCI express motherboard connectors. I don't believe the claim 150W per slot was introduced with PCI express 2.1 even makes sense since from what I can tell the 2.0 spec was introduced together with the 2.0 Card Electromechanical Specification. There's no such thing as a 2.1 CEM spec. And while I'm not an expert, I don't see how the you could add support for 150W from the slot without updating the CEM (well unless it was already speced in the CEM but not in the main spec for some reason although both the 2.0 CEM and 225/300W high power CEM only seem to mention 75W from the slot).

What does seem clear is that initially PCI-express only allowed 150W total (75W from the slot and 75W from the 6 pin) but with the 2.0 standard, the option for 225W and 300W were introduced with the addition of the 8 pin connector. According to PCI-sig, 225W can either come from an 8pin plus 75W from the slot or two 6 pin with 75W each and 75W from the slot or one 6 pin and one 8 pin but still 75W from each and 75W from the slot. 300W uses 75W from the slot, a 8 pin connector for 150W and a 6 pin connector for 75W (the option for 3 6 pin is also mentioned but it's stated as a unpreferred option). I have come across any mention of getting 150W fron the slot from PCI-sig themselves. Note that while some cards with 375W exist using two 8 pin connectors and 75W from the slot, these aren't specified in the spec. This increaee in power options with PCI express 2 may be one source of the confusion. From my research, it seems to me in recent times we're another source of confusion hence my demands for high quality referencing. Really I think we need references from an expert (perhaps a hardware engineer) who has referred to the PCI-sig specs themselves not simple hardware review sites.

Nil Einne (talk) 07:21, 23 November 2012 (UTC)

Transfer Speeds

Is this right? Does the B->T conversion really scale like that? What is that unit anyway?:

Per lane (each direction):
  • v1.x: 250 MB/s (2.5 GT/s)
  • v2.x: 500 MB/s (5 GT/s)
  • v3.0: 1 GB/s (8 GT/s)
  • v4.0: 2 GB/s (16 GT/s)
16 lane slot (each direction):
  • v1.x: 4 GB/s (40 GT/s)
  • v2.x: 8 GB/s (80 GT/s)
  • v3.0: 16 GB/s (128 GT/s)

--Diblidabliduu (talk) 21:22, 9 December 2012 (UTC)

It's explained in the article. GT/s = gigatransfers/s.
PCIe 1 & 2 use 8b/10b encoding, thus 2.5 GT/s = 250 MB/s, 5 GT/s = 500 MB/s. PCIe switched to 128b/130b encoding, so it's 8 GT/s ≈ 985 MB/s. Zac67 (talk) 21:51, 9 December 2012 (UTC)
Yeah, ok. Thanks. --Diblidabliduu (talk) 21:57, 9 December 2012 (UTC)