Question The "lens" of two NVMe(s) in RAID 0

It's doubtfull that the extra 4 pcie lanes would be making a big difference in most things.
Unless you are running a benchmark that specifically isolates that area.

I believe that whatever your understanding about how "available pcie lanes" are supposed to work is completely wrong. Try reading this (For all practical purposes the Chipset is a giant multiplexer) and see if you can put a more specific example on an actual Motherboard.

I still believe that your understanding is completely wrong because you began asking if available PCIe Lanes makes one Processor slower than other. PCIe Lanes only affect the possible amount of external data movement at the same time (Assuming you are using them), so if you're asking about if more PCIe lanes magically increases Processor processing performance, the answer is no.

For most practical purposes, assume that all data is going to go though the Processor (PCI Express supports P2P (Peer-to-Peer) so one PCI Device may theorically talk directly to another, but that could make the example too complicated...).
Processor has its own PCIe Controller that exposes PCIe Lanes that are wired directly to it, so these are the lowest possible latency connectivity for a PCI Device to get to the Processor, besides than the bandwidth is entirely dedicated because it is not multiplexed at any point with any other device (Well, except most likely internally on the Processor, but we don't care about that).
Some PCIe Lanes (Which Intel calls DMI, don't recall if AMD has any fancy name) also links Processor with Chipset. Chipset acts like a glorified multiplexer because it provides a lot of connectivity (Whenever extra PCIe Lanes, SATA, USB) but you have a limited upstream link bandwidth connecting the Processor and Chipset, so if you were to try to use ALL Chipset connectivity at the same time, you would be bottlenecked by this Processor-Chipset link. For example, most LGA 1700 Chipsets uses 8x PCIe 4.0 lanes between Processor and Chipset, but many Motherboards can be populated with three 4x PCIe 4.0 NVMe SSDs on Chipset lanes, yet you wouldn't get triple performance by putting them on RAID 0 because the uplink bottleneck (If they were PCIe 3.0, you could, because they can't saturate the upstream link). Also, EVEN if you were not bottlenecked, Chipset lanes are slighty worse because they incur extra latency because they're further away from the Processor, having to go though the internal Chipset PCIe Switch to get to the Processor.

Yet, for all practical purposes, I believe that you're most likely overstimating RAID 0 performance. Do you move data in bursts that could require a level of performance beyond what a single PCIe 5.0 SSD on Processor lanes gives?

preWarp said:

The motherboard, can take the max available lanes from the processor and make more lanes for peripherals, but total bandwidth is limited by the sum of the max lanes available from the processor.

Click to expand...

You got it wrong. Is not about total Processor lanes, is about the lanes that are connected between devices. You're failing to view PCIe lanes as what they are: Pipe of a certain size, where at any given time you may have a maximum flow of data going though it. If the devices that are connected to that pipe (Or better said, PCIe link) are actually capable of moving that amount of data is another matter.
If instead of a Video Card you used a 16x PCIe Slot wired to the Processor lanes with a 100G NIC and wanted to download a file to a NVMe SSD plugged to a M.2 slot wired to the Chipset PCIe Lanes, data has to go from NIC -> PCIe link -> Processor internal PCIe Switch -> PCIe link -> Chipset internal PCIe Switch -> PCIe link -> NVMe SSD. The links between the NIC and the Processor and the Chipset and the NVMe SSD are dedicated to usage for these devices, but Processor and Chipset link will be absolutely NOT be used only by the NIC, but data from multiple devices plugged to Chipset would be going though it at the same time. That is the multiplexing part that I mentione before.

I tried to look around if I can get a link that explains it better than me and I found this: https://www.highpoint-tech.com/post...-switch-chipsets-for-pcie-gen3-m-2-nvme-cards
The concept that I try to explain mostly works. You only need to consider than the Chipset is a PCIe Switch PLUS some built in devices (Mainly USB and SATA), but is limited by the upstream bandwidth from its usually 4 or 8 lanes to Processor.
You can also try looking at ThreadRipper Pro / WRX90 platform Block Diagrams if you want to see a 128 PCIe Lanes monster.

preWarp said:

So my question is given the limitation to the max lanes available from the processor, in my case it would be 16 for Skylake and 20 for Rocket Lake, why am I seeing the same performance using desktop applications?

Click to expand...

preWarp said:

Running CPU and Memory benchmarks, you could see speed jumps in your processor bandwidth and for the dynamic RAM but one could not perceive a change in performance of desktop applications. The loading time would make it seem like the computer is running at the same speed as the older machines.

Click to expand...

Because SSDs may be faster in throughput but latency has remained roughly the same since the introduction of the NVMe protocol, so you feel no difference in desktop applications, and you will not feel them either in a RAID 0 with PCIe 5 SSDs unless you daily routine involves moving terabytes of data from one place to another.
The only technology that was faster than standard Flash in a SSD was what Intel used in its Optane SSD and it was discontinued. This is what you want:

You don't want more bandwidth/throughput, you want lower latency.

preWarp said:

How much faster are desktop applications on this machine versus a z590 based motherboard with a Intel core i5-11600 and only one NVMe drive?

If the z890 chipset machine described only had one nVME drive would not this new PC seem the same slow speed of its predecessor because I only have one NVMe and not two?

Click to expand...

I don't know how many times I have to state this. YOUR UNDERSTANDING OF HOW PCI EXPRESS IMPACTS PERFORMANCE IS COMPLETELY WRONG. A PCIe link is a highway with X amount of lanes and Y max car speed (Given by Generation, like PCIe 3/4/5) which gives a potential maximum of Z bandwidth if it was used to its full theorical capacity. If you are NOT actively using the bandwidth because there is no device connected, having more lanes does nothing, and even if you have a device connected, it doesn't mean that it will use the maximum bandwidth, so it doesn't matters how much lanes a platform has unless you plan to actively use them by having either more or wider, faster devices plugged in.
Having two SSDs doesn't even mean that you are accessing them simultaneously, because if you had a common non-RAID configuration with one drive for Windows and the other exclusively for games, the other SSD 4 lanes are sitting near idle unless you're actually loading a game.

Your dekstop application performance and how fast and responsive a system feels is 100% limited by latency, NOT throughput. Two SSDs in RAID 0 would have twice the maximum throughput, but latency would still remain the same OR worse, cause you need some processing power to strip what bits goes to which SSD, and that extra step will increase latency, even if it is minimum. As I said in my previous post, the only thing with lower latency than possible with a high end NAND Flash based SSD (Which is every SSD in existence that is NOT 3D XPoint/Optane) is an Optane SSD itself (If somehow you need more than that you're in fancy PCIe RAMDisk territory). All other performance differences are minimal or incremental unless you're actually moving or processing data. I don't know how else to explain this before you understand that PCIe lanes is the least of your problems.

Stop wasting your time with PCIe related questions because NOTHING related to that will make your system feel faster. Spend your time researching what makes Optane faster.

zir_blazer said:

I don't know how many times I have to state this. YOUR UNDERSTANDING OF HOW PCI EXPRESS IMPACTS PERFORMANCE IS COMPLETELY WRONG. A PCIe link is a highway with X amount of lanes and Y max car speed (Given by Generation, like PCIe 3/4/5) which gives a potential maximum of Z bandwidth if it was used to its full theorical capacity. If you are NOT actively using the bandwidth because there is no device connected, having more lanes does nothing, and even if you have a device connected, it doesn't mean that it will use the maximum bandwidth, so it doesn't matters how much lanes a platform has unless you plan to actively use them by having either more or wider, faster devices plugged in.
Having two SSDs doesn't even mean that you are accessing them simultaneously, because if you had a common non-RAID configuration with one drive for Windows and the other exclusively for games, the other SSD 4 lanes are sitting near idle unless you're actually loading a game.

Your dekstop application performance and how fast and responsive a system feels is 100% limited by latency, NOT throughput. Two SSDs in RAID 0 would have twice the maximum throughput, but latency would still remain the same OR worse, cause you need some processing power to strip what bits goes to which SSD, and that extra step will increase latency, even if it is minimum. As I said in my previous post, the only thing with lower latency than possible with a high end NAND Flash based SSD (Which is every SSD in existence that is NOT 3D XPoint/Optane) is an Optane SSD itself (If somehow you need more than that you're in fancy PCIe RAMDisk territory). All other performance differences are minimal or incremental unless you're actually moving or processing data. I don't know how else to explain this before you understand that PCIe lanes is the least of your problems.

Stop wasting your time with PCIe related questions because NOTHING related to that will make your system feel faster. Spend your time researching what makes Optane faster.

Click to expand...

Question, 2x4TB m.2 drives in raid to get around dual drive letters, would that be worth doing? 2x4TB was at the time cheaper than 1 8TB. I just don't want some that may access my PC forget to change install location from C to D.

Don't care about the speed increase from RAID0, or JBOD.