AMD SB850/890GX with SSD RAID-0 first look

[tdbone1]

anyone?
could use some help here?

[tdbone1]

Bump!

[tdbone1]

is this forum dead or is it just this thread?
i been asking for help like a week now
not one reply in a thread that had 37,000 views
come on

[F@32]

is this forum dead or is it just this thread?
i been asking for help like a week now
not one reply in a thread that had 37,000 views
come on

Sorry no one got back to you. And yes, this thread is dead. Folks moved on to new shiny things including me
Any how. Back on topic. First of all please run CDM with 1Gb test size for both scenarios. In off chance that your Windows RAID is still faster - do not despair. It is not unusual what you are observing. On some hardware Soft RAID-0 will be faster when it runs on host CPU. When you are running SB850 it is in essence Hybrid RAID. Someone will have better explanation as to why this is the case. This is even more pronounced on PCIe HBA like 9240 and 9211.

Nothing to worry mate.

[tdbone1]

Sorry no one got back to you. And yes, this thread is dead. Folks moved on to new shiny things including me
Any how. Back on topic. First of all please run CDM with 1Gb test size for both scenarios. In off chance that your Windows RAID is still faster - do not despair. It is not unusual what you are observing. On some hardware Soft RAID-0 will be faster when it runs on host CPU. When you are running SB850 it is in essence Hybrid RAID. Someone will have better explanation as to why this is the case. This is even more pronounced on PCIe HBA like 9240 and 9211.

Nothing to worry mate.

i already posted a pic of the ATTO using bios-raid 0 with 64KB stripe but here it is again along with CDM



By nunyerb at 2011-09-22

i dont actually have my 80GB back until tomorrow but when i get it back i will run more tests with the (2) 60GB back in windows soft raid

i already did the ATTO and it was almost 1100MBs with win7 soft raid0
it was awesome

btw i dont know what you mean about shiny new hardware
i have the best micro mboard out for amd (even though it is 1 year old)
my SSD are sata3 and rock
so dont know what you mean about shiny better....

[mattkosem]

btw i dont know what you mean about shiny new hardware
i have the best micro mboard out for amd (even though it is 1 year old)
my SSD are sata3 and rock
so dont know what you mean about shiny better....

Llano chipset has better/faster raid. If you run AS SSD benchmark does it indicate any alignment issues?

--Matt

[tdbone1]

windows 7 x64 Dynamic Disk = two 60GB sata III (OCZ Solid 3`s)

hmm interesting results

[tdbone1]

no one interested in compairing my pics in post # 80 and 82?

[mattkosem]

The CDM benches are within a margin of error of each other. The drives had quite a bit less data on them in the second benches too. ATTO really sucks as a benchmark though tbqh, and is not to be trusted if you're looking for a realistic representation of the performance of the drive because of the highly compressible nature of the data that it uses. Do you have CnQ enabled? Have you tried AS SSD? Were the drives secure erased before either of the tests? Are the partitions on the drives aligned? Are you sure sata3 is enabled? If you pick 0 fill in CDM, do the numbers look closer to ATTO? Try Anvil's bench or straight Iometer.

You have to remember that these are SSDs and, in raid especially, are not going to show the same bench results on every bench run. Writing a lot to the drives in a given day can easily explain a 200MB/sec drop in seq reads. I've seen it time and time again, on 4+ drive setups at least. Heck, if you left your system completely idle overnight and ran the benches again you'd probably see better numbers too. These are also sandforce based drives and will respond much differently with compressible vs incompressible data.

--Matt

[tdbone1]

The CDM benches are within a margin of error of each other. The drives had quite a bit less data on them in the second benches too. ATTO really sucks as a benchmark though tbqh, and is not to be trusted if you're looking for a realistic representation of the performance of the drive because of the highly compressible nature of the data that it uses. Do you have CnQ enabled? Have you tried AS SSD? Were the drives secure erased before either of the tests? Are the partitions on the drives aligned? Are you sure sata3 is enabled? If you pick 0 fill in CDM, do the numbers look closer to ATTO? Try Anvil's bench or straight Iometer.

You have to remember that these are SSDs and, in raid especially, are not going to show the same bench results on every bench run. Writing a lot to the drives in a given day can easily explain a 200MB/sec drop in seq reads. I've seen it time and time again, on 4+ drive setups at least. Heck, if you left your system completely idle overnight and ran the benches again you'd probably see better numbers too. These are also sandforce based drives and will respond much differently with compressible vs incompressible data.

--Matt

everything is setup as per post #1

800MBs seq read = my onboard raid controller vs 1050MBs = windows 7 stripe volume (dynamic disk)

i think my onboard raid controller is capping me dont you?

[mattkosem]

You didn't really answer many of my questions. I'm trying to help you diagnose this, but can't really help if you won't try these things. As I explained, not all of these benches produce apples to apples comparable numbers (especially on SF drives) and what you are seeing is probably correct/normal for the hardware you are using.

That said, I doubt that your controller is crapping out on you. The "onboard raid controller" is the same hardware as the non-raid onboard controller and, as F@32 mentioned, software raid is sometimes faster than hardware raid (hardware being used loosely in this case since its just an alternate implentation of software raid in this case).

--Matt

[tdbone1]

i posted benchmarks from different programs
both programs show the dynamic raid is faster then the fake-raid/bios-raid.
i have read even in this thread that the 890gx raid controller caps out at 800MBs wasnt it and the intel capped out at 600MB`s?
if that is true it shows that is what i am getting.

the problem is i can not put back the dynamic raid to bios-raid.
my computer is now setup with the OS on a single ssd x25-m 80GB with the two ocz solid 3 60GB in a dynamic raid
its going to be hard to benchmark the solid 3 on the bios-raid as i have a lot of data on the dynamic raid set.

with the data i have already given....2 benchmark apps that were run on the bios-raid and on the dynamic raid we should be able to make some kind of conclusion shouldnt we?

[mattkosem]

i posted benchmarks from different programs
both programs show the dynamic raid is faster then the fake-raid/bios-raid.
i have read even in this thread that the 890gx raid controller caps out at 800MBs wasnt it and the intel capped out at 600MB`s?
if that is true it shows that is what i am getting.

the problem is i can not put back the dynamic raid to bios-raid.
my computer is now setup with the OS on a single ssd x25-m 80GB with the two ocz solid 3 60GB in a dynamic raid
its going to be hard to benchmark the solid 3 on the bios-raid as i have a lot of data on the dynamic raid set.

with the data i have already given....2 benchmark apps that were run on the bios-raid and on the dynamic raid we should be able to make some kind of conclusion shouldnt we?

Just because the answers that F@32 and I are giving you aren't the ones you want, doesn't mean they aren't correct or aren't answering your questions. If you look back even one page, you will see that SB850 with bios raid produces numbers of ~1GB/sec in both AS SSD and CDM with the right drives. If you were to search a bit more in this forum, you would find that SB850 can even produce numbers of nearly 1100MB/sec.

I'm just going to summarize/paraphrase what we have already told you again here:
1) Not all raid implementations (software, hybrid, or otherwise) perform the same.
2) Your sandforce-based drives will only produce the advertised max throughput numbers with incompressible data (as is used by ATTO).
3) CDM, AS SSD, and Anvil's utilities use almost completely incompressible data by default and, as a result, will show you numbers that are lower than ATTO. These numbers should give you a realistic representation of worst-case scenario performance on your drives and are generally very consistent from test run to test run.
4) If you were to run CDM with the 0-fill option enabled, it would likely produce numbers similar to what you see in ATTO because the 0-fill data can be compressed.
5) If you run the same benchmark on the same SSDs, you will not always see the same numbers.
6) Completely wiping the drives prior to benching will generally produce the highest bench results.
7) Having more data on your SSDs will slow them down.

The answers to your questions lie within those pieces of information. All signs point to "what you're seeing is probably normal".

--Matt

[tdbone1]

3) CDM, AS SSD, and Anvil's utilities use almost completely incompressible data by default and, as a result, will show you numbers that are lower than ATTO. These numbers should give you a realistic representation of worst-case scenario performance on your drives and are generally very consistent from test run to test run.

is that supposed to say "compressible"?
otherwise im not understanding

[mattkosem]

is that supposed to say "compressible"?
otherwise im not understanding

No, it's definitely meaning to say incompressible (as in, cannot be compressed). ATTO, as well as CDM if using 0 fill, uses very compressible data. Since the SandForce controllers derive much of their performance edge from compressing data, the nature of the data makes a big difference when benchmarking those drives.

--Matt

[tdbone1]

No, it's definitely meaning to say incompressible (as in, cannot be compressed). ATTO, as well as CDM if using 0 fill, uses very compressible data. Since the SandForce controllers derive much of their performance edge from compressing data, the nature of the data makes a big difference when benchmarking those drives.

--Matt

in "as ssd" it shows my dynamic drive is showing as 2 seperate drives.
how do i test when it is showing them as seperate drives?

[mattkosem]

in "as ssd" it shows my dynamic drive is showing as 2 seperate drives.
how do i test when it is showing them as seperate drives?

Ah, yeah, forgot it's not volume aware. You'll have to try anvil's bench if you want to run a third bench.

--Maty

[tdbone1]

ok here is Anvil with default settings on my (2) sata 3 60GB drives in a windows 7 x64 dynamic volume stripe.

[F@32]

OK, let's back up for a second. What are you trying to figure out or accomplish? Is your goal to test Windows Soft RAID vs AMD Hybrid RAID? Or is your goal to ensure your SB850 performs at spec?

To ensure it performs up to the spec - see the 1st post tips. For apples to apples comparison you will need OS installed on independent drive and perform CDM or Anvil tests after drives were secure erased. Make sure to set everything up as posted in the 1st post of this thread. With 2 of my X-25M Soft and Hybrid 2R0 performed the same. Let me see if I can dig out bench screen...

Here you go:
SB850 Hybrid RAID (click to enlarge)


SB850 Soft RAID (click to enlarge)


P.S. Regarding new shiny things. Mattkosem is partially right. You could blame Intel's PCH for stealing SB850 short lived throughput crown. But mostly because thanks to Ao1 with his and others research on this forum, you will soon realize that all that you need to remove disk bottleneck is single SSD capable of pushing at QD1 200MB/s sequentially and about 20MB/s randomly...

[tdbone1]

Is your goal to test Windows Soft RAID vs AMD Hybrid RAID? Or is your goal to ensure your SB850 performs at spec?

i did a simple test when i bought both the sata3 drives (i made it a dynamic volume) and seen them do 1050MB`s
originally my goal was to use the drives on the bios-raid (thinking this would be as fast as they could go) little to my dismay did i discover that they were only doing 850MB`s (so i thought)
guess i might have been using the wrong benchmark tool but i really dont understand how that matters because i use that same tool on both tests
it clearly shows a difference from bios-raid to dynamic volume stripe
actually did two tests (CDM and ATTO) and CDM didnt show to much of a difference but ATTO showed 300MB`s read difference
so i decided that was to much to lose so i put them back to dynamic volume stripe

but now i have my os installed on a signle 80GB ssd (x25-m) and i have the dynamic volume setup as a stripe (with 100GB of data on it)
so i really can not go back.

now i am moved on to the 2nd part of your question and just making sure they are performing as they should

i am configured as per post #1 except i never did the secure format but i did do an acronis disk director format
i think 1050MB`s is doing pretty good isnt it?

[TassadarDavid]

Hi all!!

I'm back looking for advices

I've just install 2xC4 128mb raid0 on a Asus M5A88-M and the results doesn't seems to me as good as they should be.

The configuration is:

-Phenom II x6 1075 (still stock, I'll overclock)
-2x4Gb RAM 1600mhz
-Asus M5A88-M
-SSD's in SATA1 and SATA2 ports

Important info:

-Sata3 enabled on bios
-C1E disabled on bios (I've also try to disable C&Q, result is the same)
-Catalyst 11.10
-Write cache enabled (by default) on Win7 device manager.

In RaidXpert th NCQ is enabled on both disks (it's the accurate, isn't it?):



But, in the other hand I've noticed this configuration that seems to indicate that both caches (write and read) are disabled, and I can't enable any of them!! (No matter I click on the other options, checks can't be activated):



The results in this moment are those:



Are them ok??? Do I have to change some configuration??

Thank you very much

[davidm71]

Hi guys,

Can anyone answer if AMD has Raid Trim support for SSDs? I have an Asus CHIV. Think thats an 890 chipset?

Thanks.