How important are 4K random reads?

[perdomot]

The Kingston Hyperx 3K 90GB SSD I ordered performs better on the sequential and 512K tests in CDM benchmark but the 4K reads are about 19MB/sec on the new drive while my old Crucial C300 64GB SSD scores about 28MB/sec. I use my SSD strictly for a boot drive for the OS and programs and am thinking that the new SSD is actually slower than my old one for how I use it. Am I wrong in this? Not even sure what I should be expecting in results as nobody has reviewed the 90GB version of the drive although I expect the 120GB scores to be similar.

 

[Nizzzlle]

Here's a pretty good review

Seems pretty solid.

 

[BrightCandle]

Depends on the workload. Standard windows and application stuff is mostly dominated by read, and the mostly dominated by 4k reads. In these workloads they make up about 60% of the total workload for the drive. Its this particular performance advantage in SSDs to HDDs that makes them feel so different, not the doubling of sustained transfer which most drives are advertised on.

However when it comes to game loading its dominated by sustained reads because the files are packed together and loaded in a batch (normally). So the workload makes a big difference as to which type of optimisation makes the most difference. But I would argue you'll notice 10MB/s on 4k more than you will 100MB/s on sustained.

 

[perdomot]

Here's a pretty good review

Seems pretty solid.

That's the 240GB SSD which I've seen reviewed elsewhere along with the 120GB one but nobody seems to have done anything with the 90GB version. Thanks though.

 

[perdomot]

Depends on the workload. Standard windows and application stuff is mostly dominated by read, and the mostly dominated by 4k reads. In these workloads they make up about 60% of the total workload for the drive. Its this particular performance advantage in SSDs to HDDs that makes them feel so different, not the doubling of sustained transfer which most drives are advertised on.

However when it comes to game loading its dominated by sustained reads because the files are packed together and loaded in a batch (normally). So the workload makes a big difference as to which type of optimisation makes the most difference. But I would argue you'll notice 10MB/s on 4k more than you will 100MB/s on sustained.

This is kind of what I figured which is why I was so disappointed in the 4K performance. How can a newer and larger SSD have 30% slower 4K reads than something a couple of years old?? I'm not a gamer so the sequential stuff is secondary to me. To make matters worse, the 120GB version is scoring nearly double in 4K reads at just under 40 MB/s. I feel like I just got robbed by Newegg.

 

[npaladin-2000]

This is kind of what I figured which is why I was so disappointed in the 4K performance. How can a newer and larger SSD have 30% slower 4K reads than something a couple of years old??

Probably they re-did the firmware to give back higher sequential read speeds during testing, because they benchmark better an provide a marketing point. The cost is generally to random read/write performance, which is what happens in real life.

 

[Makaveli]

4k performance is very important and is why the intel drives were killing everything when introduced.

 

[BrightCandle]

The marketing and marking of SSD's is pretty poor. They are all list the sustained transfer speeds, but those really are not what make the SSDs special. Intel continues to have good IO performance and personally I think people should pay more attention to the 4k read than any other figure, because its the one that actually makes the difference.

 

[Cerb]

4K random reads aren't too important, just for 4K accesses. 4K represents an aligned worst-case, and most drives natively write at either 4k or 8k granularity, IIRC. A drive with excellent 4K performance will also have excellent 32K performance, FI, and 32-128K are going to be fairly common, reading/writing config files, cached web files, page file(s), search indices, etc..

For SD cards, I wish they would include 32K tests (FAT32), but that's getting OT.

Sadly, you can't always find what drives use all the channels, which is usually a good indicator of performance, relative to the review model, which almost always uses them all. Even the 'good' brands are known to do this, too, for smaller models.

Max. sustained random 4k read/write:
90GB – 20,000/50,000 IOPS
120GB – 20,000/60,000 IOPS
240GB – 40,000/57,000 IOPS
480GB – 60,000/45,000 IOPS

Taken from SR's review. Note the disparity in sustained IOPS. In the future, pay close attention to such specs. Peak IOPS are generally irrelevant.

I feel like I just got robbed by Newegg.

By Kingston. If you just got it, wipe it and RMA it. You'll get stuck with a restocking fee, I'm sure, but you'll probably be happier with a 128GB M4 or 830.

P.S. No refund/swap from Newegg? Yeah, that does suck. You might be able to sell it for less loss than the restocking fee would be, though. It's a nice drive, from what I can see, but favors writes, and high-QD. Really, it would probably be OK for you, as well, except that you'd constantly have in the back of your mind that it wasn't what you were expecting (read: intending to be paying for), and that's always irritating.

 

[perdomot]

4K random reads aren't too important, just for 4K accesses. 4K represents an aligned worst-case, and most drives natively write at either 4k or 8k granularity, IIRC. A drive with excellent 4K performance will also have excellent 32K performance, FI, and 32-128K are going to be fairly common, reading/writing config files, cached web files, page file(s), search indices, etc..

For SD cards, I wish they would include 32K tests (FAT32), but that's getting OT.

Sadly, you can't always find what drives use all the channels, which is usually a good indicator of performance, relative to the review model, which almost always uses them all. Even the 'good' brands are known to do this, too, for smaller models.
Taken from SR's review. Note the disparity in sustained IOPS. In the future, pay close attention to such specs. Peak IOPS are generally irrelevant.

By Kingston. If you just got it, wipe it and RMA it. You'll get stuck with a restocking fee, I'm sure, but you'll probably be happier with a 128GB M4 or 830.

That chart is the reason I got the 90GB model since the reads which are my main focus, match the 120GB model. Sadly, the reviews on the 120 show considerably better performance on 4K random reads. As for RMA, I went to do that and Newegg only offers replacement, not refunds. Spoke to customer service and they said the same thing. Wouldn't let me change to either the M4 or 830 which is what I wanted to do.