Why does my Intel based SSD have such slow random read/write performance?

[jimhsu]

How much free space is on the drive? Are you using as a OS drive?

I bought my SSD because I had $250 spare cash (ebay sales) and had the option of
1. A graphics card (to replace my 8800GTS). Useful, but... not really necessary. (I game on a single monitor at 1050p). This kind of speaks to the longevity of Nvidia's 8 series.
2. More memory - I can't fit any, I have 8GB.
3. A Core i7/i5 build - too expensive, and I have a E8400.
4. One of those "newfangled SSD things".

The option I chose should be obvious. It's difficult to pass up a several hundredfold theoretical increase in performance (yes, it's theoretical, but still.) That decision was helped by the launch of the G2 series along with the massive price cut that took place (I was set on a G1 model on ebay when it was released).

 

[jimhsu]

About the disk space issue:

I'm trying to gather some statistics so that I can plug it into something like eureqa and see what's going on.

Here's a public spreadsheet for any AS SSD benchmarks for Intel G2 SSDs:
http://spreadsheets.google.com/ccc?key=0AvbXytCFRyFCdEdKdHJIVk9FYjZadndidnQ0OTEybVE&hl=en

Benchmark protocol:
1. Close all running programs.
2. Run 1 pass of the SSD toolbox optimizer on the drive.
3. Note the amount of free space on the SSD (My computer).
4. Run 1 pass of AS SSD.
5. Fill in the spreadsheet.

I know this is not the most scientific way of looking at this, but it'll have to suffice.

With 4 data points only, my observations are inconclusive. Though it is fairly apparent that performance drops off as a function of less free space.

 

[Beautiful_Life]

About the disk space issue:

I'm trying to gather some statistics so that I can plug it into something like eureqa and see what's going on.

Here's a public spreadsheet for any AS SSD benchmarks for Intel G2 SSDs:
http://spreadsheets.google.com/ccc?key=0AvbXytCFRyFCdEdKdHJIVk9FYjZadndidnQ0OTEybVE&hl=en

Benchmark protocol:
1. Close all running programs.
2. Run 1 pass of the SSD toolbox optimizer on the drive.
3. Note the amount of free space on the SSD (My computer).
4. Run 1 pass of AS SSD.
5. Fill in the spreadsheet.

I know this is not the most scientific way of looking at this, but it'll have to suffice.

With 4 data points only, my observations are inconclusive. Though it is fairly apparent that performance drops off as a function of less free space.

Did you successfully update your firmware? It seems like TRIM may be an issue here. The only thing decreasing significantly is your writes. Nevertheless, those are rather interesting and problematic number. I too would like to know the reason behind it.

 

[jimhsu]

TRIM is updated successfully (the only reason I can run SSD toolbox). fsutil behavior query DisableDeleteNotify returns 0. Newest (Dec 2009) chipset drivers were just installed. Not using Intel's IMSM (removed it). AHCI is on (obviously).

Should I try an update from 02HA to 02HD?

Sorry, I don't see anything else.

 

[Beautiful_Life]

TRIM is updated successfully (the only reason I can run SSD toolbox). fsutil behavior query DisableDeleteNotify returns 0. Newest (Dec 2009) chipset drivers were just installed. Not using Intel's IMSM (removed it). AHCI is on (obviously).

Should I try an update from 02HA to 02HD?

Sorry, I don't see anything else.

I think you should update your firmware. I've seen people reporting 40%+ increase in speed.

If I'm not mistaken there were a few firmware updates
- One sometime in October, still buggy and caused some of the drives to die
- One on the 30th of November, had the full TRIM support, less buggy but caused some drives to die (This was 02HD). Was pulled out quickly
- One released on the 8th of December. Finally everything worked well, best speed and drives didn't die. (A never revision of 02HD)

 

[jimhsu]

The 30th November one was, I recall, 02HA which did actually complete successfully on my computer. Updating to 02HD made no difference.

Other things I tried in the past few hours:
- Moving the drive from SATA port 5 to port 1 - no difference
- Installing Intel Rapid Storage Technology 9.5.6 - this actually seemed to make some difference. I'll try to get benchmarks.

Well, that's interesting. My 4K64T write score plunged nearly 30%, but system boots consistently 2 sec faster (timed with Reboot-Time) and subjectively is "faster". DPC latency has gone down from an average of 110 us to under 100 us.

 

[EarthwormJim]

Have you tried secure erasing the drive and starting over?

 

[Beautiful_Life]

The 30th November one was, I recall, 02HA which did actually complete successfully on my computer. Updating to 02HD made no difference.

Other things I tried in the past few hours:
- Moving the drive from SATA port 5 to port 1 - no difference
- Installing Intel Rapid Storage Technology 9.5.6 - this actually seemed to make some difference. I'll try to get benchmarks.

Well, that's interesting. My 4K64T write score plunged nearly 30%, but system boots consistently 2 sec faster (timed with Reboot-Time) and subjectively is "faster". DPC latency has gone down from an average of 110 us to under 100 us.

You should be very happy then. 2 sec reduction on boot times really shows that your random reads are getting faster. In this case I'd say just screw your benchmarks.

 

[jimhsu]

And Crystaldiskmark shows that my random writes are much faster:

I'm beginning to trust theoretical SSD benchmarks less and less. http://www.storagesearch.com/ already mentioned the limitations of traditional benchmarks (as opposed to application-centric benchmarks), and at least this is being more clear.

 

[jimhsu]

Then comes researchers and businessmen...

Ah, yes, grant money (I know because I'm involved with academia at the periphery). Research projects that justify investments into $20000 32-core monsters with 64GB of ram, that get upgraded every year because "we can't sparsify the data enough to fit into "just" 32GB". As far as SSDs are concerned, they seem to be completely ignoring SATA models and are investigating the PCI-express monsters.

 

[darkrunner25]

Thought this might be relevant.

Intel X25-M G2 160GB.

Fresh out of the box, no optimization, IDE:

Updated Firmware, ran Optimizer, still IDE:

Changed to AHCI:

-Chad.

 

[Beautiful_Life]

Thanks Chad. It's extremely relevant and useful. Could you please let us know what's your motherboard and your I/O controller (or southbridge)?

 

[sub.mesa]

IDE emulation would not allow usage of NCQ. Intel SSDs use NCQ to buffer commands, to process them in parallel. That you can clearly see on the posted benchmarks; the 4K benchmark with 64 queued I/O's is skyroof while in IDE its virtually the same as single queue.

The benchmarks above show up to 40.000 IOps when reading randomly, a very good score! So when doing random I/O, you will be doing between 22-152 MB/s depending on queue size (depends on application). But you need AHCI interface to go higher than 22MB/s, because it needs NCQ to allow the SSD to process multiple I/O at once. Note that HDDs score very low on strong random I/O.

 

[darkrunner25]

Thanks Chad. It's extremely relevant and useful. Could you please let us know what's your motherboard and your I/O controller (or southbridge)?

Motherboard: http://www.newegg.com/Product/Produc...82E16813188054

CPU: http://www.newegg.com/Product/Produc...82E16819115214

I believe the controller is integrated into the i7 chips.

-Chad.

 

[EarthwormJim]

Motherboard: http://www.newegg.com/Product/Produc...82E16813188054

CPU: http://www.newegg.com/Product/Produc...82E16819115214

I believe the controller is integrated into the i7 chips.

-Chad.

ICH10R still in i7 (P55) motherboards. The memory controller is integrated, not the southbridge (yet).