3

I am getting pretty slow writing performance in comparison with my Z87 board. Does anybody know if these results are to be expected out of the Intel RS3UC080 or what the problem might be if there is one?

Below is the hardware/software i used to conduct the tests. Latest firmware updates for the RS3UC080 and SSDs are already installed.

Controller: Intel RS3UC080(LSI3008)
Mainboard: Asus Hero VI(Z87 Chipset)
CPU: 4770K @ 4,5 GHz
Ram: 16GB @ 2,4GHz
Backplane: HP DL360 G6
SSD: 4x Samsung Evo 840 120GB in Raid-0

Benchmark Tool: AS SSD Benchmark 1.8.5611.39791 and 1.7.4739.38088
OS: Windows XP 8.1 x64

I ran the tests using the same hardware, once with the Intel RS3UC080 by using the Backplane of a DL360 G6 and once without it by directly connecting the drives to the mainboard.

The raid array on the Intel RS3UC080 was created via the setup utility which shows up during booting.

I just noticed that AS SSD was set to german when one screenshot was taken. Below are translations.

lesen = read
schreiben = write
zugriffszeit = access time

RS3UC080 Z87

2 Answers 2

0

I don't know this controller, but have you tried to disable controller cache allowing OS direct access to disk? There was somewhere here similar thread with SSD disks and LSI controller.

Can you provide screenshot of controller configuration options?

12
  • The RS3UC080(LSI3008) is not a real raid controller and has no cache. In the setup utility which shows up during booting are barely any configuration options and i don't quite know yet how this raid web console is supposed to work.
    – Susan
    Commented May 28, 2015 at 15:57
  • According to manual: lsi.com/downloads/Public/Host%20Bus%20Adapters/SAS3_IR_UG.pdf . Disk write caching, which is enabled by default on all Integrated Striping volumes Commented May 28, 2015 at 16:19
  • There is a "disk cache policy" in the raid web console but it was already enabled. I think windows is doing some caching, reading should not be slower than writing in the 4K test on the Z87.
    – Susan
    Commented May 28, 2015 at 16:21
  • Can you disable it and check if it helps? Commented May 28, 2015 at 16:22
  • Just tried it, disabling the disk cache policy does not help either.
    – Susan
    Commented May 28, 2015 at 16:27
0

The exponentially increasing raw bit error rate with 21nm TLC flash is the problem... ECC tries to hide it from the consumer, but the extra heat produced by the SSD CPU causes it to overheat and slowdown. (esp. if it does read - modify - write).

http://techreport.com/review/27727/some-840-evos-still-vulnerable-to-read-speed-slowdowns

Avoid sub 25nm TLC drives if you can (like plague). Use samsung 850 pro or similar with MLC or SLC 3D NAND. Preferably made with 35nm+ process node. (The bigger the charged area, the more electrons it can hold, the longer the retention time).

Make sure the raid driver supports TRIM.

Set the stripe size to flash erase block size.

Use 64K cluster on NTFS.

Make sure the noatime is set and do all recommended SSDs optimization.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .