PCIe 5.0 specification announced, will bring 128GB/s transfer speeds

midian182

Posts: 9,743   +121
Staff member
What just happened? Are you excited about the arrival of PCIe 4.0? Despite the latest standard still not being available to consumers, the PCI-SIG (Peripheral Component Interconnect Special Interest Group) has announced its successor: the PCIe 5.0 specification.

AMD’s new X570 chipset, designed for the upcoming Ryzen 3000 series CPUs, comes with PCIe 4.0 support, enabling hardware such as Gigabyte’s Aorus AIC Gen4—a PCIe 4.0-based SSD card that can reach read speeds of 15 GB/s when running in a RAID 0 configuration. But PCI-SIG isn’t standing still—the consortium says the PCIe 5.0 spec has now been completed and sent out to members.

Taking 18 months to develop, PCIe 5.0 doubles the peak bandwidth of PCIe 4.0 from 64GB/second to 128GB/s via x16 configuration. It also boasts a raw bit rate of 32GT/s, which PCI-SIG says will allow it to target high-performance applications such as machine learning, artificial intelligence, gaming, storage, visual computing and networking.

Other PCIe 5.0 specifications include:

  • Leverages and adds to the PCIe 4.0 specification and its support for higher speeds via extended tags and credits
  • Implements electrical changes to improve signal integrity and mechanical performance of connectors
  • Includes new backwards compatible CEM connector targeted for add-in cards
  • Maintains backwards compatibility with PCIe 4.0, 3.x, 2.x and 1.x

The reason PCIe 5.0 has been ratified before PCIe 4.0’s consumer launch is the latter’s delay; having stuck with a four-year release cycle for PCIe 1.0, 2.0, and 3.0, there were seven years between the arrival of PCIe 3.0 in 2010 and the completion of PCIe 4.0 in 2017.

“AMD congratulates PCI-SIG on the release of the PCI Express 5.0 specification to the industry and the future 2x increase in performance it is expected to deliver. We expect to bring our first PCIe 4.0 specification CPUs to market this year and look forward to meeting the future bandwidth demands of end-users with PCIe 5.0 technology," AMD said in a statement.

With PCIe 4.0 only just rolling out, it will be a couple of years before support for PCIe 5.0 arrives.

Permalink to story.

 
Reshuffling the standard inside out, disrupting the manufacturing in a big way, harassing the end users into upgrading whole computers. Taking 3 years to do this, and all just to double the speed?

I'd wager this PCI Special Interest Group's main interest isn't really that special, it is to stay relevant for as long as they can, and not so much helping anybody except themselves.

To offer anything less than a 4x times performance upgrade is a major waste of everybody's time and money, and a hindrance to computing progress.
 
Last edited:
Is there anything that would benefit from 5.0? Or even 4.0? Or any time in the near future?
 
Hmm...I remeber a time when I heard the same arguments versus a 40MB (yes, MB!) hard drive. I do agree, however, that time will tell and that the vast majority of users will have years to evaluate whether PCIe 5.0 is necessary / valuable or not.
 
With Sony claiming their next gen console will have storage tech significantly faster than what is currently available on pc, wonder whether it will feature pcie 4.0 or 5.0.
 
An SSD that hits 100GB/sec (based on current technology) would need Liquid Cooling

and an external box with a Graphics Card + StupidFast SSD like this would need more than a Thunderbolt connection to make use of it

Time for Optical Connect?
 
Is there anything that would benefit from 5.0? Or even 4.0? Or any time in the near future?

CXL and NVlink that essentially cuts down latency between sub/co- processors, part of the largest issue with SLI and Crossfire was that there was unstable bandwidth and latency that caused issues and required constant tweaking of drivers and gaming software to work, if there is a constant it can be programmed for without variables aka increasing scaling and a system where it would just work without major support on anyone's end where it would see separate cards(I know this is a bad a example but bare with me here) as one card and merging resources as like a raid assembly rather than current methods where VRam is copied rather than doubled, before both GPUs had to access resources from the CPU and RAM and run as separate individual entities where as now GPUs can have access to in a greater amount with hardly any latency each GPU in parallel, meaning it should just operate as a whole.

It's a dually a big deal and although NVidia is doing nothing with it currently Tom Peterson before leaving NVidia for Intel let slip there were big plans down the line as the 2000 series is equipped with NVlink through the new high bandwidth bridge.
 
Is there anything that would benefit from 5.0? Or even 4.0? Or any time in the near future?

The SSDs announced a few days ago certainly benefited. Would be great for professional work.

Professional work is the only area where this will mater.
I doubt there is any consumer work loads that will see a difference.
Buys will most likely be gamers who fall for marketing, Epeen bragging.
 
The SSDs announced a few days ago certainly benefited. Would be great for professional work.
The consumer will likely see instant booting PCs. IMO, that would be impressive.
An SSD that hits 100GB/sec (based on current technology) would need Liquid Cooling

and an external box with a Graphics Card + StupidFast SSD like this would need more than a Thunderbolt connection to make use of it

Time for Optical Connect?
IMO, cooling would only be required if that data transfer rate is sustained over long periods of time. I have to wonder just what applications would need to continuously move 100GB/s. Perhaps large data centers, however, I cannot imagine them sending all their data through just one server.

While the speed will be nice, I cannot imagine that there are very many applications out there at any level, professional or not, that would be able to saturate the bus with that much data for extended periods of time.

To put this into perspective, we are talking moving all the data on a 2-layer blu-ray disk (50GB) in 0.5 s, and all the data on a 3-layer blu-ray disk (100 GB) in 1 s.

To me, this seems more like a solution in search of a problem.
 
To me, this seems more like a solution in search of a problem.
I know 'pure research' can give unusual benefits. I'm wondering if we are looking at 'pure application' and what that may imply. What if there is an unidentified problem that this solves. Neat to watch for.
 
So with 10 gb lan or 100gb lan you can get a 100gb file in 1-4 sec if the speed from ssd m2 allows it. so getting a game at 1-8tb with 1-16k textures you could play games that are enormous into the future. even CAD could bi big presentation over 100gb. lets say the pak file to games are about 1tb eatch you would have enormus speed to load those levels with pcie 4.0-5.0 gpu speed. if we watch 8k movies and gaming in 8k textures we gonna needing that speed. overheating may be a problem but making ssd gpus bigger smaller would go past that problem. I dont talk about pcie 6.0 yet. (greetings FUTURE games CAD s)
 
Back