PCI Express is a fundamental standard in modern computers. A local Input/Output bus and is used for internal connections of motherboard chipsets, communication with the CPU and also for installing important components such as graphics cards, SSDs or dedicated networking or sound solutions.
The history of PCI Express stems from the technology industry’s need to have a unique, high-performance bus that could replace those used in computer architecture a couple of decades ago, from ISA to AGP, and also the original PCI. In this way, PCI-SIG, the group responsible that includes more than 700 major companies, created a standard that has achieved its primary goal of improving performance, replacing legacy interfaces such as SATA for storage, as well as saving cost and complexity in the design of modern computers.
Actually, a good part of the architecture of current PCs is based on a standard that you will see abbreviated as “PCIe” or “PCI-E” and to which today we will dedicate this guide where we will remind you everything you need to know about it, its present and what is to come.
PCI Express versions
The initial version was based on the existing PCI, but with one fundamental change: its structure as point-to-point, full-duplex lanes, working in series. In a practical aspect that is better understood, each individual PCIe port and its installed cards can get the maximum performance out of the bus, as opposed to the much slower PCI that tended to saturate when computers mounted multiple connectors.
With the main foundations laid, from there the main focus was on increasing bandwidth to improve performance and the launch of PCI Express 3.0 in 2010 was a before and after for the standard. A vast improvement over the original PCI 1.0 by quadrupling its transfer speed to 8 GT/s; its total bandwidth to 126 Gbit/s (15.8 GB/s) and its bandwidth per lane to 15.8 Gbit/s (1969.2 MB/s).
As industry needs continue, especially in the graphics market and other applications with high workloads and bandwidth, PCI-SIG has continued to release versions to increase performance, lower latency, offer superior RAS capabilities and improve I/O virtualization. Currently, the most widespread version is PCIe 4.0a standard that increased the number of channels through which the signal passes to double the bandwidth to 16 Gigatransfers per second.
More recently came PCIe 5.0PCIe 5.0, another breakthrough over previous versions of the standard with a substantial performance increase, using a frequency of 32 GHz to reach a bandwidth of 128 GB/s in full duplex, doubling that of PCI Express 4.0 and quadrupling that of PCI Express 3.0. Adoption of this generation is still tepid with only the latest Intel and AMD processor platforms supporting it. In addition, there are still major components, such as graphics cards, that have not yet made the leap to Gen5. Their capabilities are also not being taken advantage of by solid state drive manufacturers.
PCIe 6.0 is on the way
PCI-SIG is always way ahead of the industry, as any revision of the standard takes quite some time until it reaches the end customer and in fact this one is well behind schedule. The main improvement will once again be a substantial performance increase to a bandwidth of 256 GB/s in full duplex, doubling that of PCI Express 5.0 and in turn quadrupling that of PCI Express 4.0. This is achieved by doubling the data rate of a PCIe lane to 8 GB/second in each direction, and much more for multi-lane configurations.
With a frequency of 64 Ghz, it will be able to deliver up to 64 Gigatransfers per second. The new PCIe 6 interface changes the encoding scheme to PAM4 to increase transfer rates. This is what truly enables the specification to achieve such high bandwidth. Technically, it modulates signals at four levels, packing two bits of information into a serial channel in the same amount of time. This PAM4 scheme is widely used in higher-performance networks such as the Enterprise InfiniBand and we have also seen it in GDDR6 graphics memory.
Another of the expected changes in this release will affect the downgrading of the physical size of the bus. A novelty that has been postponed for too long and that should solve the monstrous size of some of the components that are stuck in the PCIe slots, such as graphics cards. This if more efficient cooling systems are achieved, it is understood, because so far it has not been possible.
PCIe 7.0 seeks the speed of light
PCI-SIG announced a major new feature for this version that we will probably not see commercially until the next decade. And it is that the standard will work on optical connectionsthe biggest advance since the standard was created. The group has created a working group that will develop the version independently of the others. Gen7 will support a wide range of optical technologies in the future to further improve the standard’s capabilities in performance, power consumption, range and latency.
There is talk of a bandwidth on one lane (x1) of 128 GT/s, which means that an x16 as used by graphics cards, the theoretical bi-directional throughput will be able to rise to about. stratospheric 512 GB/s. The group will first focus on professional technologies such as quantum computing, cloud and hyperscale data centers, and then move on to high-performance computing at the client level.
Types of PCI Express
PCIe has gone through multiple revisions as you may have read, but they all have a common denominator: they use the same physical connections that you will see in four primary sizes called x1, x4, x8 and x16.. There are also x32 ports, but they are extremely rare and usually not seen on consumer hardware.
The different physical sizes allow moving different amounts of simultaneous connections and data to the backplane.. The larger the port, the greater its maximum capacity. These connections are colloquially known as “lines” or “lanes”, where each PCI-E lane is composed of two signaling pairs, one for sending data and the other for receiving. In practice, the greater number of lanes allows performance and capacity gains and data can flow faster between the peripheral and the rest of the computer system.
Not all devices need the same capacity and although there are no established guidelines on what type of slot to use, we can point out a few practical examples of use. For a common sound card or a Wi-Fi card a PCI-E x1 is sufficient, while a high-end network card, RAID controllers or USB 3 expanders use x4 or x8. Graphics cards often use x16 for maximum transfer capacity. SSDs in M.2 format for PCIe are usually connected to x4 ports, but all indications are that they will be outgrown in future generations.
To be taken into account for practical purposes.
One of the technical sections of PCI-Express that confuses ordinary consumers is that an x16-sized port may not offer the maximum number of lanes allowed by the standard. The explanation is that while PCIe can accommodate individual connections in unlimited numbers, there is a practical limit to the performance of the chipset on the board..
This brings us to a conclusion that I’m sure you know: not all motherboards are the same. Low-end motherboards may have x16 slots but their performance is equivalent to x8 for example. High-end motherboards for gaming PCs or professional workstations, usually have several x16 slots which, in addition to size, take full advantage of the performance and bandwidth allowed by the standard.
If you place a high-end graphics card in a slot that -even though it has an x16 size- does not offer the maximum number of lines, you may have a bottleneck and not get its maximum performance. Another aspect to consider is that many boards with 2 x16 slots only offer the maximum number of lines if you use one of them, dropping to x8 if you use both together. Note that smaller x1 and x4 cards can be installed in the x8 and x16 (obviously not the other way around). Also, some x8s have a different set of pins and cannot be installed in x16 slots.
While the performance improvement that comes with each release is appreciated, it wouldn’t hurt to also review these types of rails that can confuse the consumer. And the same size of slots we talked about above. It’s a very technical topic, but we hope it has become a little clearer to you everything related to PCI Express, the most important (and almost unique) local Input/Output bus of modern computers.
