To begin i would like to inform you and apologize about my “bad” english (i’m a french man). Well i joined this forum because i like level1techs videos and i know their explanation qualities.
My tiny question but mostly difficult for me (because i do not have any new pc, mine his from 2010 so i can’t do test on it) : Can i expect with the new pcie standard pcie 4.0 to fully exploited my futur gpu on pci 4.0 x16 reduce in the bios by a x4 one since we know pcie 4.0 double bandwith.
Thanks for your futur answers
Nothing really for the next 5 years. It takes 2 Titan cards using the Vulkan API in Ashes of the Singularity to come close to maxing out the PCI-E 3.0 spec. With PCI-E 5 on the horizon we won’t be seeing any GPU’s coming close to maxing out that bandwidth for at least 5 years. The only potential place to exploitation is RAID0 NVMe SSDs. Even then they are very pricey.
TL;DR: Don’t worry about it.
Thanks for your answer. But i haven’t been enough clear. I just whant to know if for exemple a quadro rtx 4000 sell with a pci-e 3.0 x16 interface (well it’s stupid like you said because pci-e 3.0 x8 would have been enough but bigger are numbers bigger are profit) will work on a pci-e 4.0 x4 at 100%.
I don’t think so, a 3.0 card can only run at 3.0 speeds. So even though the x4 slot will have the bandwidth the card will only be running at 3.0 x4 speed.
Thanks for your attention
Even if an adapter for exemple pci-e 4.0 x4 to pci-e 3.0 x8 will exist ?
Most likely not.
PCIe is only as fast as the oldest generation in the chain, so PCIe 3.0 in your example. Lanes can’t be multiplied out to previous generations as far as I know either, so a PCIe 3.0 card in a PCIe 4.0 x4 slot will only be as fast as PCIe 3.0 x4.
Okay well i’m disapointed in this technologie i have now a problem of not enough pci-e lanes on the futur ryzen 3950x.
Thanks to three of you.
One exception do this is that you can run two PCI 3.0 devices through the x570 chipset on AMD devices at full speed because the chipset will handle routing the traffic from separate pci3.0 busses to pci 4.0 back to the CPU. Without bottlenecking (basically that is the full bandwidth though).
Unless you have devices through a plx bridge, this will always be so. Because technically with a bridge you are servicing requests at pci 4.0. Will someone build an add in card that would take 16 pcie 3.0 lanes and give you 8 pci 4 lanes? unlikely. Will someone take a card that gives you 4x pcie 3.0 m.2 lanes and gives you a pci3 4.0 x8 interface? Possibly.
Okay i see your point. So if i want to run on x570 taichi with a 3950x : 1 m2 nvme gen4 ssd + 2 m2 nvme gen3 + 1 10GbE card + 1 quadro gpu + 8 drives sata, All of them at full speed it’s possible ?
the sata lanes might be shared and it’ll be cutting it close because of 2 gen3 ssd… if its < 3gb/sec pcie3 ssd then no bottlenecking but practically all those devices arent going to be busy at once, so it should be fine. the quadro rull run at pcie3.
the sata will probably bottleneck somewhat, that’s a good test, if they are used simultaneously (e.g. big raid array) assuming its not shared with pcie somehow, which some of the sata might be depending on how the mobo vendor has hooked it up
The 2 ssd will not be working as much (and as the same time) as the 4gen ssd wich will be writing things as l2arc for freenas and the 8 hdd will be however working with it simultaneously as raidz2. I hope to achieve 10GbE without l2arc.
Thank you for your advice
With the aorus taichi x570 mb i have understood that 2 m2 port were directly took in charge by the chipset and only on was directly connected to the cpu. I Hope the 8 hdd will not bottleneck. But it’s only a project that i have been rolling since 1-2 years and money isn’t there yet so when it will maybe we’ll see ryzen 4 gen.
The 3950X isnt there yet.
But looking at the things you need for pci-e connectivity,
i wonder if Threadripper wouldn’t be a better platform for you.
pci-e4.0 does not really tell me that much yet.
Simply because the mainstream platform is limmited,
in the amount of available lanes anyways.
My previous mind configuration was using a threadripper but the goal of this is to run fedora as the Host kvm and two guest (Windows 10, freebsd(freenas)) a plex server will run on the windows 10 (and some cad stuff) and i need as much power as possible. Threadripper cpu been erased by the 3900x in passemark so… But yes the power of the 3900x ( or 3950x) with the number of pci-e lanes of the threadripper is the perfect cpu.
But like i said previously i did not have the money yet i am still a student in engenieer so money should flow in few years but not with my summer job Uber eats .
But still, i like sharing my project and improving my skills with you guys