Compatibility issue slows PCI Express
Compatibility issue slows PCI Express
By Vikram Karvat, Director of Marketing, ServerWorks Corp., Santa Clara, Calif., EE Times
October 28, 2002 (10:14 a.m. EST)
URL: http://www.eetimes.com/story/OEG20021024S0014
PCI Express' innovative approach to I/O is fascinating and, when faster versions become available, it may have interesting potential for future applications. However, the current generation of PCI Express doesn't have a strong value proposition.
PCI and PCI-X can support the bandwidth requirements at a lower cost while also offering backward compatibility that PCI Express can't offer. Because there are hundreds of available PCI adapter cards in existence, products that are compatible with traditional PCI are significantly more useful than those that lack it. Therefore, PCI best fits all the needs of the desktop space, while PCI-X 2.0 is the best fit for server, workstation and communications applications.
PCI-X 2.0 uses the same protocols, signals and connector as traditional PCI, but increases data rates to 266 MHz and 533 MHz. It maintains hardware and software backward compatibility to even the lowest-performance version of PCI. Beyon d PCI-X 533, there are already plans to extend PCI-X to 1,066 MHz. And beyond that, there are also possibilities for PCI-X 2133.
PCI Express, on the other hand, is a radical new approach to local I/O, building upon the traditional PCI software model but using an entirely new hardware interface. Instead of the traditional PCI signaling, PCI Express uses new, high-speed differential signaling. A dedicated differential pair is used to transmit data, and a second differential pair is used to receive data. In addition Express embeds the clocks within the signals. These changes make it possible to achieve high bandwidths with fewer pins. However, because the signaling is new, its connector is unique and can't be shared with traditional PCI adapters.
All versions of PCI Express run at the same 2.5-Gbit/second signaling rate. To enable multiple performance points, different data widths have been defined. The x1 version is enabled for the low end of the market, x4 and x8 for the middle, and the x16 ve rsion for the high end of the market. At the high end, using the x16 version, 4 Gbytes/s can be achieved with a 164-pin connector.
With PCI-X 2.0, two data widths are available, at 32 and 64 bits. PCI-X 2.0 frequencies can range from 33 MHz to support traditional PCI and up to 533 MHz for the fastest version of PCI-X 2.0. At 533 MHz, 4.3 Gbytes/s can be achieved with the traditional 184-pin PCI connector.
The PCI Express designers successfully created a standard that could transmit data with fewer pins. However, PCI Express does not lower costs when all cost factors are considered. For example, the PCI Express protocol engine has turned out to be very large. It will require several hundred thousand gates, which will make it several times larger than the protocol engines for traditional PCI architectures.
In addition, the PCI Express physical interfac e will require significantly more silicon area than traditional PCI I/O buffers. PCI Express also requires a new, more expensive connector. Overall, PCI Express is a more expensive I/O interface than PCI-X 2.0 for similar bandwidth.
PCI-X 2.0 has a similar cost model to traditional PCI. The buffers are similar in size and cost, the logic is similar and the pin counts remain the same. In addition, PCI technologies offer the capability to support multiple loads on a bus. This significantly reduces cost over point-to-point technologies such as PCI Express that require independent buses for each load.
PCI and PCI-X 2.0 have significant utility because there are so many systems and cards that are compatible with this common standard. In time, PCI Express can become equally useful, but it will require a few years to build up a base of compatible cards and systems.
PCI-X is geared for server, workstation and communications manufacturers who have requirements for high bandwidths, low cost an d backward compatibility. High-performance applications such as 10-Gbit Ethernet, 10-Gbit Fibre Channel and Infiniband need at least 2 Gbytes/s of bandwidth. Because PCI-X 2.0 is capable of 4.3 Gbytes/s, it has sufficient bandwidth to fully enable these applications, for both single-port and dual-port adapter card designs.
Sticking with traditional PCI is the best approach in the desktop computer space, where low cost and backward compatibility are typically the highest priorities. Modems, sound cards, Ethernet, game cards and other desktop adapter-card applications don't require bandwidth beyond what the current PCI can offer. In the future, if additional bandwidth is needed, PCs will be able to migrate to higher-performance local I/Os such as PCI-X.
Current versions of PCI Express use 2.5-Gbit/s signaling. At these signaling rates, the cost of PCI Express is high relative to the performance offered. For the future, 5- and 10-Gbit/s versions of PCI Express are being planned. When these ver sions come to fruition, the cost and performance of PCI Express may make the transition from other I/O technologies to PCI Express worthwhile. The 5-Gbit/s version of PCI Express is slated to arrive about 2006. When it becomes available there may be a viable reason to migrate to PCI Express.
Related Semiconductor IP
- AES GCM IP Core
- High Speed Ethernet Quad 10G to 100G PCS
- High Speed Ethernet Gen-2 Quad 100G PCS IP
- High Speed Ethernet 4/2/1-Lane 100G PCS
- High Speed Ethernet 2/4/8-Lane 200G/400G PCS
Related White Papers
- How HyperTransport and PCI Express complement each other
- Advanced switching boosts PCI Express
- With StarFabric as an on-ramp, the PCI Express Advanced Switching is ready
- Verifying PCI Express design IP
Latest White Papers
- New Realities Demand a New Approach to System Verification and Validation
- How silicon and circuit optimizations help FPGAs offer lower size, power and cost in video bridging applications
- Sustainable Hardware Specialization
- PCIe IP With Enhanced Security For The Automotive Market
- Top 5 Reasons why CPU is the Best Processor for AI Inference