TX1 PCIE 4x bandwidth

How much PCIE 4x bandwidth can reach?
Is there any test data?
Can It reach 1GB/s?

It’s up to 5Gbps per lane (Gen2).

Hi Trumany

I know it. Is there any test data? Is Transmission efficiency up to 50% or more?

I use TX1 PCIE 4x with FPGA, I use it to transfer 4KP30RGBA(about 950M/s) data.

The efficiency is dependent on your encode format.
Four lanes could reach about 2.5GB/s

Hi Trumany

4KP30RGBA is Raw Data.

2.5GB/s mean WRITE and READ ? is it bilateral?

so WRITE or READ could reach about 1.25GB/s?

Hi watsoncao

As your update - “I use TX1 PCIE 4x with FPGA, I use it to transfer 4KP30RGBA(about 950M/s) data”.
Is there any specific error you met?

You could also set the system to max performance on your use case, please refer to below link:
[url]http://elinux.org/Jetson/TX1_Controlling_Performance[/url]

Thanks

It’s 5Gb/s/lane/direction, the PCIe part of TX1 is standard, you might need to go through PCIe specification for details.

hi kayccc and Trumany
maybe I describe it incorrectly, i want to use TX1 pcie x4 transport big data, but i am not sure it is enough. because I use intel’s CPU test, FPGA read CPU data rate is 1GB/s in fact.
PCIE Gen2 speeds is 5.0 GT/s/lane, so for x4 speeds is 5.0Gbps*8/10(encode)/8(B)*4(lans) = 2GB/s. 2GB/s is Theoretical bandwidth.

https://devtalk.nvidia.com/default/topic/940605/?comment=4899077
this post describe that Drive is detected and working at x4, drive is tested on other systems with >1GB/s speeds.When mounted to the TX1, the read speed I get is good, around 800MB/s, 800MB/s / 2GB/s around 40%. this is not enough, so i mean do you have test data for TX1 PCIE x4 ?

We do have test data where PCIe Gen-2 and x4 speeds touch up to 12Gbps i.e 1.5 GB/s in an environment where end point’s DMA is trying to do read or write to TX1’s system memory.
In case of flash based storage devices, read from flash or write to flash might play some role here.
Coming to this specific device which showed >1GBps on other systems but around 800MBps on TX1, we need to consider clocks at which both systems are running and test environment being same in both cases etc…
Although there will be some gap (little) between perf measured on an x86 system and TX1 because of TX1’s lack of IO-coherency. But achieving around 1.5GBps is very much possible.

Hi vidyas,

Got it, thank you very much.