HW: 1xLS2088 CPU (8xA72 Core:1.8GHz, Bus:700MHz) with 2x 4 Lane PCIe 3.0 (2x~31Gbit/s) 2xDDR4 memory block (64bit+ECC, 1.8MT/s=(2x115Gbit/s) There are two parallel scatter/gather DMA data transfer processes: 1. DDR4(1) write from PCIe(1) 2. PCIe(2) write from DDR4(2) What is the achievable native data throughput rate (maximum and average estimated with all overhead)? Where is bottleneck? (Our HW is in production but we have to answer to an important RFP now.)
This case has already been addressed in the case you entered in our internal system.
Please refer to the following answered from the hardware AE team.
The theoretical maximum throughput for Gen3 X4 is roughly 28Gbps. The throughput value measured by the customer is much lower than expected. As you know, many factors come into play when measuring performance, e.g., software that is running on system, measurement equipment, block size, …. For the customer’s system, we are not able to answer the question of where is the bottleneck. The customer would need to perform an analysis of their system to find the bottleneck. We cannot guarantee a maximal data throughput rate as requested. For Gen3 x4, we think the customer should be able to achieve above 20Gbps.
Have a great day,
TIC
-----------------------------------------------------------------------------------------------------------------------
Note: If this post answers your question, please click the Correct Answer button. Thank you!
-----------------------------------------------------------------------------------------------------------------------
Dear Wang,
we are using the LS2088 NXP eval board and NXP FW and a Samsung NVME SSD
card (PCIe 3.0 x4 Lane) which can 2.1Gbit/s write speed in a PC. We arrive
only around 8Gbit/s on the LS2088 platform however I am agree with you, the
speed should be minimum around 16Gbit/s.
Note: Increasing the PCIe block size from 128 Byte to 256 resulted around
16% speed increasing to 9.2Gbit/s but it is still not enough for our
application.
Why can we not arrive the 2.1Gbps card write speed? I don't think that the
HW can't do it.
Thanks and Regards,
ZDG-ZODIAC DATA SYSTEMS GMBH logo
Zodiac Aerosystems | Control Systems
Division
Zodiac On Board & Testing
Csaba RATZ
Senior Development Engineer
Friedrich-Ebert-Straße 75, Haus 51.
- TechnologiePark
51429 Bergisch Gladbach - Germany
Tel: 0049 2204 84 4112 - Fax: 0049
2204 84 4199
Csaba.Ratz@zodiacaerospace.com
http://www.zodiac-data-systems.com
Geschäftsführer / Managing
Directors: Cécile Dubrovin,
Jean-Marie Bétermier
Registergericht / Register Court:
Amtsgericht Köln
Handelsregisternummer / Commercial
register number: HRB 47131
Von: yipingwang <admin@community.nxp.com>
An: Csaba Ratz <csaba.ratz@zodiacaerospace.com>
Datum: 09.08.2018 06:21
Betreff: Re: - Re: LS2088 Plattform speed
NXP Community
Re: LS2088 Plattform speed
reply from Yiping Wang in Layerscape - View the full discussion
This case has already been addressed in the case 00170863 in our internal
system.
Please refer to the following answered from the hardware AE team.
The theoretical maximum throughput for Gen3 X4 is roughly 28Gbps. The
throughput value measured by the customer is much lower than expected. As
you know, many factors come into play when measuring performance, e.g.,
software that is running on system, measurement equipment, block size, ….
For the customer’s system, we are not able to answer the question of where
is the bottleneck. The customer would need to perform an analysis of
their system to find the bottleneck. We cannot guarantee a maximal data
throughput rate as requested. For Gen3 x4, we think the customer should
be able to achieve above 20Gbps.
Have a great day,
TIC