Low 4K read/write numbers in recent reviews

[zetlali]

I saw the write-up today for the Intel 730 and noticed that the ASSSD results seemed a bit off on the 4K reads and writes. This is the review I'm referencing.

http://www.servethehome.com/intel-730-480gb-ssd-quick-benchmarks/

In other reviews I've seen, the 4K read numbers are usually in the 35-38MB/s range while these results show only 21-24 MB/s. The difference in these results should probably be looked into. If I had to guess, C states are enabled in the STH environment or some other form of cpu throttling that is limiting 4K performance. Link to other reviews for reference.

Intel SSD 730 Series SSD Review (2x480GB) | The SSD Review
Intel 730 Series 480GB SSD Review - the Skulltrail of SSDs? - Benchmarks - CrystalDiskMark

 

[Patrick]

I saw the write-up today for the Intel 730 and noticed that the ASSSD results seemed a bit off on the 4K reads and writes. This is the review I'm referencing.

http://www.servethehome.com/intel-730-480gb-ssd-quick-benchmarks/

In other reviews I've seen, the 4K read numbers are usually in the 35-38MB/s range while these results show only 21-24 MB/s. The difference in these results should probably be looked into. If I had to guess, C states are enabled in the STH environment or some other form of cpu throttling that is limiting 4K performance. Link to other reviews for reference.

Intel SSD 730 Series SSD Review (2x480GB) | The SSD Review
Intel 730 Series 480GB SSD Review - the Skulltrail of SSDs? - Benchmarks - CrystalDiskMark

@zetlali - Great insight. I have a December piece queued up discussing LSI v. using onboard Intel controllers with the same drives. Most other sites use base Intel controllers.

 

[zetlali]

@zetlali - Great insight. I have a December piece queued up discussing LSI v. using onboard Intel controllers with the same drives. Most other sites use base Intel controllers.

I hadn't realized the onboard Intel controllers were caching this much. However, I did receive my drives today and I set one of them up on an LSI3008. My results do show a bit higher performance in QD1 4K reads/writes.

 

[Chuntzu]

Wow very interesting info

 

[Jeggs101]

How much pre-conditioning are you doing? 12 hours? 24? More? Are you doing full drive conditioning?

 

[zetlali]

How much pre-conditioning are you doing? 12 hours? 24? More? Are you doing full drive conditioning?

If by conditioning you mean write a bunch of data to the drives, I did zero conditioning. These results were basically hooking up the drive, writing about a GB of data to it, then running a few benchmarks. Determining steady-state performance would certainly be useful information, but I don't believe that was the goal of the quick benchmarks posted. The performance of an SSD can vary drastically when the drive has to start erasing blocks of data on the fly.

edit: Here is what an 8 drive array of 840 pros looks like after 6 months of use.

 

[Patrick]

OK that might be the difference there. Good thought @Jeggs101 . The drives are going through 24 hours of full drive random conditioning.

@zetlali I am also doing IOmeter testing on the drives. Just publishing those results separately. I will try the fresh drive benchmark with the next set of drives I get.

Is the left CDM picture after several months and the right a new array screen grab? Seems order of magnitude what we are seeing.

 

[zetlali]

Screenshot on the left is from July, screenshot on the right is from yesterday. So what are you doing with the drives when conditioning them? Since the LSI controllers don't utilize trim, you're essentially relying on Intel's built-in garbage collection. Given that every manufacturer's clean-up methods will be different, your results across different drives may vary quite a bit from a fresh install. I'd suggest documenting your conditioning method. If you've done this already in a previous review I may have missed it.

If your review is from essentially a completely filled drive, those are actually pretty impressive numbers. The steady-state performance on these drives is suppose to be quite good. It's one of the reasons I bought the drives.

 

[Patrick]

Screenshot on the left is from July, screenshot on the right is from yesterday. So what are you doing with the drives when conditioning them? Since the LSI controllers don't utilize trim, you're essentially relying on Intel's built-in garbage collection. Given that every manufacturer's clean-up methods will be different, your results across different drives may vary quite a bit from a fresh install. I'd suggest documenting your conditioning method. If you've done this already in a previous review I may have missed it.

If your review is from essentially a completely filled drive, those are actually pretty impressive numbers. The steady-state performance on these drives is suppose to be quite good. It's one of the reasons I bought the drives.

I will re-write this up into its own piece and just link it from everything. That is what I am doing. Basically, similar to what Tom's does.