PCIe 7.0 draft specs become available, detail 4x bandwidth gains over PCIe 5.0

Daniel Sims

Posts: 1,375   +43
Staff
What just happened? Most consumer PCs currently use PCIe 3.0 or 4.0 components, while 5.0-supported systems are considered cutting-edge, but the organization overseeing PCIe development is already working on 7.0. The latest milestone represents a significant step on the technology's future roadmap.

This week, the PCI Special Interest Group (PCI-SIG) consortium released the first review draft specification for PCIe 7.0 to members. Details on the in-development technology are still early, but the announcement confirms that the evolution of PCIe is proceeding on schedule.

The consortium released the 0.3 draft specification during the 2023 PCI-SIG Developers Conference in Santa Clara. While there hasn't been much new publicly-available information on PCIe 7.0 since its initial unveiling last summer, the milestone allows hardware developers to begin early work on products that will eventually support the standard.

Like PCIe 6.0, 7.0 is designed to double its predecessor's bandwidth, effectively achieving four times the bandwidth of PCIe 5.0. It takes the data rate from 64 to 128 gigatransfers per second for a bidirectional 512GB/s of throughput via an x16 configuration.

The new specification also continues the previous generation's use of PAM4 signaling. The designers are striving to offer low latency and high reliability while increasing power efficiency. Unsurprisingly, PCIe 7.0 maintains backward compatibility with its predecessors.

During the conference, PCI-SIG will also provide an update on PCIe 5.0 and 6.0 cabling specifications – mostly for data center usage – which it plans to release in Q4 2023. Both standards will use the same specification, sharing the same signaling frequency.

Also read: The Inner Workings of PCI Express – How Hardware Works

Most consumers won't see devices utilizing PCIe 6.0 for some time, much less 7.0. The latest PC components started the upgrade to PCIe 5.0 connections last year, and most users haven't adopted them yet. The consortium released the final PCIe 6.0 specification last January, but no available products yet employ it. The final specs for 7.0 are planned to arrive in 2025 while supporting hardware could appear in 2027 or 2028.

The two upcoming generations will first target enterprise markets. PCIe 6.0 will likely debut in data centers or other industrial, automotive, military, and aerospace settings. PCI-SIG is aiming 7.0 toward 800G ethernet, AI, machine learning, cloud computing, quantum computing, hyperscale data centers, and high-performance computing.

Permalink to story.

 
Getting to the point where a cache module could become a third level cache for graphics card on the mobo or in a adjacent 16x pci-e to store textures and LLMs and not even touch the hard drive
 
Getting to the point where a cache module could become a third level cache for graphics card on the mobo or in a adjacent 16x pci-e to store textures and LLMs and not even touch the hard drive
Any memory off-die isn't ideal, regardless of PCIe 7.0's transfer rate. For example, the Radeon RX 7900 XTX actually has longer L3 cache and DRAM latencies than the RX 7600, even though the bandwidth between each individual MCD and the GCD is over 800 GB/s. Far better to just increase the on-die cache and amount of VRAM.
 
Not interesting considering GPU prices. What the point of PCIe speed when speed costs so crazy?
 
Yeah but PCI-E 8.0 is way better. We don't even have PCI-E 6.0 yet, so who cares even more so when you see how PCI-E 5.0 has made almost no difference.
 
Any memory off-die isn't ideal, regardless of PCIe 7.0's transfer rate. For example, the Radeon RX 7900 XTX actually has longer L3 cache and DRAM latencies than the RX 7600, even though the bandwidth between each individual MCD and the GCD is over 800 GB/s. Far better to just increase the on-die cache and amount of VRAM.
Completely agree, but I also think any motherboard should come out with a standard fast 16GB onboard linked directly to the CPU, why they never did it makes no sense
 
That's a stupid clickbait title. Yes, it's 4 times more than PCI-E 5.0. And 8 times more than 4.0. And 16 times more than 3.0. Because literally every new version of the standard doubles the bandwidth, it's nothing new, and it's utterly pointless to compare it to an arbitrary older revision, because "four times faster" sounds more exciting than "twice as fast" or something.
 
That's a stupid clickbait title. Yes, it's 4 times more than PCI-E 5.0. And 8 times more than 4.0. And 16 times more than 3.0. Because literally every new version of the standard doubles the bandwidth, it's nothing new, and it's utterly pointless to compare it to an arbitrary older revision, because "four times faster" sounds more exciting than "twice as fast" or something.
Completely disagree that it's a clickbait title. Nobody is using PCIe 6.0 as there are no devices that utilize it, so the performance aspect isn't common knowledge. On the other hand, although PCIe 5.0 is still relatively new, enough people have a sense of how much bandwidth it offers over 4.0 and 3.0, ergo it's a more logical point of reference.
 
Back