Question about SSD capacity speeds

[jelome1989]

Benchmark results from Anandtech's 'The Destroyer' and IOmeter show that 250gb ones are generally slower than their higher capacity counterparts. Why are smaller capacity SSDs slower? Say I have a 500gb SSD and I partitioned it into two partitions, 250gb each, will the drive perform like the 250gb ones?

Thanks

 

[Valantar]

Second question: No*

First question: Parallelism. Higher capacity SSDs (generally) have more flash dies (the actual flash "chips") inside. These are accessed in parallel (files are divided into tiny chunks and spread out around all available dies), which means that the more dies the SSD controller can read from/write to at the same time, the faster said operation will be.


*This depends on how your OS/SSD controller treats partitions. Does it limit each partition to certain physical regions of the drive? Are these analogous to physical chips (or, say, half of each chip)? If the partitions are set to specific dies, this will limit performance, as the available dies for parallelism will be limited. If the partitions are "virtual" (i.e. simply restrictions in how large a percentage of the total drive capacity can be taken up by data in a given partition), this won't affect performance. As for which of these is the case, I have no idea, nor do I have any idea what actually determines this (or if what I'm saying here is purely a philosophical argument with myself). Should be pretty easy to test out, though.

 

[Billy Tallis]

Second question: No*

First question: Parallelism. Higher capacity SSDs (generally) have more flash dies (the actual flash "chips") inside. These are accessed in parallel (files are divided into tiny chunks and spread out around all available dies), which means that the more dies the SSD controller can read from/write to at the same time, the faster said operation will be.


*This depends on how your OS/SSD controller treats partitions. Does it limit each partition to certain physical regions of the drive? Are these analogous to physical chips (or, say, half of each chip)? If the partitions are set to specific dies, this will limit performance, as the available dies for parallelism will be limited. If the partitions are "virtual" (i.e. simply restrictions in how large a percentage of the total drive capacity can be taken up by data in a given partition), this won't affect performance. As for which of these is the case, I have no idea, nor do I have any idea what actually determines this (or if what I'm saying here is purely a philosophical argument with myself). Should be pretty easy to test out, though.

The flash translation layer running in the SSD controller is not aware of partitions. It only tracks logical blocks. By convention, PCs use the first few LBAs to store partition information, but this is only interpreted by the host system, not the SSD. To the SSD, there's nothing special about the LBAs that might contain an MBR or GPT partition table.

NVMe introduces the concept of namespaces, which are in some ways partitions that the drive is aware of. Support for multiple namespaces is optional, but on drives that support multiple namespaces, they can be used to create separated pools of the available storage. They can also support some advanced enterprise-oriented features like thin provisioning and having different block formats per namespace (eg. 512B or 4kB LBAs, with or without extra metadata/protection information).

 

[BonzaiDuck]

Not hi-jacking, I hope. But the OP raised the point that smaller means slower and larger means faster. This is one facet in my decision to make a planned M.2 NVMe purchase in the next two to four months.

Here is a link to an offer of a Samsung 960 EVO M.2 NVMe:

250 GB 960 Pro M.2 NVMe

After posting here, I'm going to verify for myself whether the seq read and seq write spec in MB/s is correct for this particular sub-model. But if you know -- you can volunteer it.

Go to LATER if TLTR.
===
Second, what is a likely guess as to the longevity or reliability of the EVO compared to a Pro? Somehow I had vaguely concluded for myself that the EVO has a shorter lifespan.

If the spec for the 250GB unit is really less than stated, or the real-world tests show that it is less than stated, what might be the correct expectations in MB/s?

The longevity factor would carry less weight or more depending on a pattern of usage, and I will say that it is likely more than for simply containing the boot-system volumes.

. . . . LATER -- This is what I found, and it answers all those questions:

Samsung 960 EVO

Should I put in a pre-order, or wait? How much would I save waiting, if the equivalent SATA SSD was maybe $50 less? Make that the only question I posed. Maybe nobody knows the answer . . .

 

[Valantar]

Not hi-jacking, I hope. But the OP raised the point that smaller means slower and larger means faster. This is one facet in my decision to make a planned M.2 NVMe purchase in the next two to four months.

Here is a link to an offer of a Samsung 960 EVO M.2 NVMe:

250 GB 960 Pro M.2 NVMe

After posting here, I'm going to verify for myself whether the seq read and seq write spec in MB/s is correct for this particular sub-model. But if you know -- you can volunteer it.

Go to LATER if TLTR.
===
Second, what is a likely guess as to the longevity or reliability of the EVO compared to a Pro? Somehow I had vaguely concluded for myself that the EVO has a shorter lifespan.

If the spec for the 250GB unit is really less than stated, or the real-world tests show that it is less than stated, what might be the correct expectations in MB/s?

The longevity factor would carry less weight or more depending on a pattern of usage, and I will say that it is likely more than for simply containing the boot-system volumes.

. . . . LATER -- This is what I found, and it answers all those questions:

Samsung 960 EVO

Should I put in a pre-order, or wait? How much would I save waiting, if the equivalent SATA SSD was maybe $50 less? Make that the only question I posed. Maybe nobody knows the answer . . .

Different capacities have different specs. See here, scroll down to the 960 Evo spec comparison table.

As for longevity, I wouldn't worry. History has shown that SSDs have - at all times - been seriously underspecced with regard to longevity. Sudden, unexpected failures do happen, but they do with all hardware. The 250GB 960 Evo i specced for 100TB written, with a three year warranty. For regular consumer usage, you'll never see even close to 100TB of writes before that drives is woefully out of date. My three-year-old 256GB 840 Pro system drive has just passed 9TBW. My 1 1/2-year-old 850 Evo 500GB game drive is sitting pretty at 1,45TBW. Don't worry.

 

[bononos]

Not hi-jacking, I hope. But the OP raised the point that smaller means slower and larger means faster. This is one facet in my decision to make a planned M.2 NVMe purchase in the next two to four months.
Here is a link to an offer of a Samsung 960 EVO M.2 NVMe:
250 GB 960 Pro M.2 NVMe
After posting here, I'm going to verify for myself whether the seq read and seq write spec in MB/s is correct for this particular sub-model. But if you know -- you can volunteer it.
.......

Check out userbenchmark for ssd test results.
http://ssd.userbenchmark.com/SpeedTest/200373/Samsung-SSD-960-EVO-250GB

The results on that site can vary quite abit because of real world conditions like bios settings. Disabling power saving options like c-states (in bios) is proven to affect random disk performance. The typical user don't feel sequential performance as much as random disk performance.
http://www.thessdreview.com/forums/...d-performance-enhancement-pros-and-cons.2763/
http://www.overclock.net/t/1333701/intel-c-states-off-better-ssd-performance

 

[BonzaiDuck]

Check out userbenchmark for ssd test results.
http://ssd.userbenchmark.com/SpeedTest/200373/Samsung-SSD-960-EVO-250GB

The results on that site can vary quite abit because of real world conditions like bios settings. Disabling power saving options like c-states (in bios) is proven to affect random disk performance. The typical user don't feel sequential performance as much as random disk performance.
http://www.thessdreview.com/forums/...d-performance-enhancement-pros-and-cons.2763/
http://www.overclock.net/t/1333701/intel-c-states-off-better-ssd-performance

True, but the idea I have for a 256GB 960 EVO M.2 with two equally allocated volumes on it isn't so much typical use. If the drive still boosts random scores by a factor of 2 or 3, then it will work it's magic.

The community seems equally split on the topic of tiered caching, and notably one veteran observed that the payoff would be bigger or worthwhile for a server situation.

I expect to deploy three or four storage devices from my motherboard: the 960 at the top of a storage pyramid; a boot-system SATA SSD; a 2TB 2.5" and 1TB 2.5" spinner. If it works right, the wear and tear on the spinners will be considerably less, and the pattern-of-usage access to those drives and the SSD as well should show benchmark scores somewhere between expected SATA SSD and the M.2.

But for an outlay of maybe $130, performance won't in any way decline, and it will be a worthwhile experiment for me. And I can save my money for a bigger M.2 much later -- when there will be more to choose from at lower prices.