New Intel SSD firmware (G2) results

[Anvil]

As some of us know there is an update for the old-timer

Link to download

Test performed on a used drive. (X58 Gigabyte UD7, slightly PCIe OC'd)

AS SSD using 02HD (oldfirmware)
as-ssd-bench INTEL SSDSA2M160 1.29.2011 6-01-45 PM.png

using 02M3 (new)
as-ssd-bench INTEL SSDSA2M160 1.29.2011 6-23-37 PM.png


I'll do a test shortly using the Intel P67 6Gb/s ports.

[Anvil]

P67 (ASUS P8P67 Deluxe, no PCIe OC)

Using the 6Gb/s Intel ports.

as-ssd-bench INTEL SSDSA2M160 1.29.2011 6-51-50 PM.png

CDM_X25M_G2_160GB_6GB_P67_02M3.PNG

According to CDM it's pretty close to 50K iops at 4KB random reads.

I'll do a quick iometer test...

[Ao1]

Thanks Anvil. I was thinking of updating my ICH9, but by the looks of it there is no need to rush.

[Anvil]

There is no reason to rush, at least not for the 6Gb/s ports using 3Gb/s drives.
I'll do one more test using the standard 3Gb/s ports.

Part of the difference might be due to my PCIe OC on the X58 platform. (105-107 iirc)

iometer showed 47-48' iops using 1 worker and 49' using 4 workers.

There are of course other reasons to go for SB (mainstream) but for power users I'd say wait for the workstation chipset.
The 2600K is a fun processor and for multimedia stuff it's a great upgrade, especially compared to the 6XXX/9XXX processors.

[thobel]

This is with 3x Intel X25m G2 with the new firmware




I think the new ASSSD has change dthe scoring system?

[zalbard]

This is with 3x Intel X25m G2 with the new firmware

Not sure. Your 4K performance is pretty low, for some reason, though.

[Anvil]

The scoring system hasn't changed.

Your score is low due to the 4K-64 Threads part of AS SSD.

Did you install RST and enable write cache?
It still doesn't explain why the read score is so low.

[thobel]

The scoring system hasn't changed.

Your score is low due to the 4K-64 Threads part of AS SSD.

Did you install RST and enable write cache?
It still doesn't explain why the read score is so low.

All fixed



I turned off Windows write-cache buffer flushing

also ran ccleaner and free space cleaner with FF

Not sure what fixed it. But I'm happy now

[Anvil]

Great score, seems like you got it sorted

2R0 Intel G2 160GB RST 10.0.0.1046

First run

02M3_X58_2R0_X25MG2160GB_SS16KB.png

Pretty good score, the score using 02HD would have been ~730

edit:
833 is more than I ever got using 2R0 Vertex LE, very close though.

[Chickenfeed]

02HD



02M3



Although I've found AS SSD to be inconsistent, after doing about 10 runs I've concluded the newer firmware reduced write performance for me using my raid 0 setup.

[Anvil]

10 runs of AS SSD would reduce performance on most SSDs , each run equals ~3GB of writes.

You can't blame the inconsistent scores solely on AS SSD, it should be fairly consistent with the exception of drives using compression.
The are a lot of factors that leads to inconsistent results while benchmarking SSDs, the SSD itself is not consistent + background tasks,..

It looks like your using 64KB or 128KB strip size?
I've found 16-32KB to be more consistent in raid, sequentials slow down a bit using smaller strips but there aren't that many operations that utilizes the seq. bandwidth of 2-3 SSDs.
(with the exception of copy operations, in the end most operations are random and random performance is great using smaller strips)

Next time youre cleaning the drives you might consider trying a smaller strip size, it might not suit you but it's worth a try.

--

3R0 X25-M G2 160GB 02M3 (secure erased)

02M3_X58_3R0_X25MG2160GB_SS16KB.png

Don't expect getting close to "secure erased" benchmark results while running OS on the drives

[Chickenfeed]

I've tested different strip sizes and all around 128 proved to be most effective for my needs.

This install is less than a week old and was securely erased. As far as 10 runs tangiblely effecting performance I haven't seen that myself. I've noticed my sequential and random write numbers vary with as sdd by 5-15 MBs run to run. Cdm seems more consistent but I still appreciate that they are both useful tests.

[felix_w]

Well, for sure my eyes can see that 4K-64thrd have been improved....let them rest for a while and bench in 3-4 days....to get more results....

[mezcal]

Thanks for the heads up, Anvil.

For those of us running the Kingston-branded X25-V drives it appears that Telnet @ overclock.net has already hacked this release to work with our drives.

Not working on my sig rig unfortunately.
I presume this is because the drives are in an array however I do not know this for a fact.
Not inclined to tackle it tonight since the OS lives there.

The G2 in my lappy flashed fine.

The G1 firmware version included with this release is unchanged (I'm still running a G1 in a cruncher.)

[Mitchb]

I saw virtually no change:

Mitch

[Anvil]

The main change is 4K at high queue depths, nothing spectacular/noticeable has happened wrt sequentials.

It does look like max latency might have improved, during iometer testing I didn't find the usual spikes.

A few P67 results.

3R0
02M3_3R0_X25M_G2_2X6GB_1X3GB_PORT_SS16KB.png 02M3_3R0_X25M_G2_2X6GB_1X3GB_PORT_SS16KB_CDM_1GB.png


4R0 - 16KB STRIP SIZE (LEFT) VS 64KB STRIP SIZE (RIGHT)
02M3_4R0_X25M_G2_2X6GB_2X3GB_PORT_SS16KB.png 02M3_4R0_X25M_G2_2X6GB_2X3GB_PORT_SS64KB.png

CDM - 16KB STRIP SIZE VS 64KB
02M3_4R0_X25M_G2_2X6GB_1X3GB_PORT_SS16KB_CDM_1GB.PNG 02M3_4R0_X25M_G2_2X6GB_1X3GB_PORT_SS64KB_CDM_1GB.PNG

[Computurd]

Cdm seems more consistent but I still appreciate that they are both useful tests.

doesnt CBM write like 15gb per run?

[Johnny87au]

Anvil can you run 4R0 again with 128k stripe?

[bot@xs]

thanks for the insight anvil

[Anvil]

@mezcal,

I've got a few Kingstons as well, maybe I'll give it a go.
I checked one of the drives and it had only 64GB host writes
Time to put it into use before it gets out-of-date.

doesnt CBM write like 15gb per run?

Depends on the SSD, it does write a lot.
The SF drives are hit the hardest, as well as the C300 top models.

Anvil can you run 4R0 again with 128k stripe?

I do concentrate on small strips, 8KB-32KB, but I'll see what I can do

[Johnny87au]

@Anvil may i ask why u prefer smaller stripe sizes?

[Anvil]

Most IO is random and random IO performs better using smaller strips. (ymmv)

So, I seldom look at the sequential figures, they are far above whats needed anyway.

[cmanser]

My numbers seem lower than most here.

I am running an ASROCK X58 Extreme & i7 950 stock while using the onboard intel ICH10 Raid controller.

All the drives have the newest firmware and were secure erased and fresh load of Win 7 64bit SP1 was loaded right before the test. Stripe size is 128k and performance is the same if I have Write Cache Buffer Flushing on or off.

Any ideas?

thanks

[Anvil]

Your score looks pretty good to me.
The 80GB drives are somewhat limited in write vs the 160GB and that should largely cover for the lower score.

Lowering the strip size to 16KB might bring a few more points, not sure if it's worth it though.

edit:

Can you verify that you have the Write Back Cache Enabled.
(The WBC is enabled using the Intel Rapid Storage Technology or "IRST or RST" for short)

[cmanser]

WOO HOO.

Thanks Anvil - that did the trick - I knew there was some setting I was missing.

New scores