SSD Thread (Benchmarks, Brands, News, and Advice)

[eYe-I-aïe...]

The problem with nand cells is that as soon as there's something written on a specific cell, which is part of a particular page, itself part of a said block, any other write to occur on any cell part of the block will be slower than it should because the controler has to read the already written data, delete it, write it in RAM, and then re-write it ALONG with the new data that was to be written...

Therefore, until or unless SSD's manufacturers find a way either to overwrite data on nand cells, or to write to a cell on a page in a block without having to go through the 4 steps described above, then, there will be slowdown coming from writing to partially written blocks. This is because, TRIM or not, there will always be partially written blocks, therefore, some slowdown.

This being said, and while any SSD is WAY faster than any HDD, I can myself feel the difference with my two Intels in RAID (so no TRIM); they slowly become slower, as I write to them. It's not the end of the world, but over a 6 months period say, windows boot time goes from 15 seconds to 25 ~ 30 seconds; launching an app instantly takes now 1 or 2 seconds, you know, not a so big deal, however, I do indeed feel the difference. I' ve been playing with 2 SSDs in RAID for more than a year and a half now, so again never got TRIM on my side, and while my SSDs are still way faster than my previous 7200RPM RAIDed HDDs, I do notice the speed decrease over time.

Fortunately for me, I just have to reinstall windows every six months or so to recover the speed. This is being done by quick formating my drives while installing windows (one of the first screens when you install). While overprovisionning an SSD with more nand cells than advertized (and user-usable) do indeed help, it's not a panacea, so let's hope the technology will evolve and solve this issue.

[chimpanzee]

The problem with nand cells is that as soon as there's something written on a specific cell, which is part of a particular page, itself part of a said block, any other write to occur on any cell part of the block will be slower than it should because the controler has to read the already written data, delete it, write it in RAM, and then re-write it ALONG with the new data that was to be written...

Whad do you mean by block ? erasable block or writable block ?

[eYe-I-aïe...]

Data is written to cells, which are grouped in pages, which are grouped in blocks. The problem is while the controller can erase an entire block in one shot, it cannot however write to a cell if any other cell within the same block is holding data...

Therefore, the controller has to first delete the already written data, so it can write something onto a cell within the said block. If the controller just deletes that data, you loose it; so, it has to first read it, then delete it, then write it to RAM, and then re-write it along with the data that was to be written.

[tilleroftheearth]

When SSD's get into multi-TB capacities, the write/erase page/blocks will be identically sized - there will be so much nand available that it will be 'affordable' to do that.

Once there, this read/modify/erase/write cycle should disappear once and for all (unless people start filling TB sized SSD's to capacity...).

Just like DSLR camera sensors, more (smaller) photo sites (=more nand capacity) is not necessarily a bad thing (primarily; noise in the final image). Increased photo sites not only offer better resolution and acuity, but they also offer increased colour graduations (colour fidelity) and maybe more importantly increased dynamic range too.

With SSD's, an order of magnitude increase in capacity (compared to today's puny offerings) will mask many more of these physical limits that nand has (glaringly) today.

Not only will it make how many write cycles each nand has less important - it will also enable the use of less sophisticated (read: cheaper) controllers too. In addition, with less 'GC' needed, the promised power saving features of SSD's might finally be (universally) be realized as well.

When SSD's finally get out of their diapers and past their teens - that will be the time when 2TB SSD's are not only affordable - but available too.

Can't wait to meet a (grown up) SSD.

(Hope it happens while I'm still 'young at heart').

[chimpanzee]

Data is written to cells, which are grouped in pages, which are grouped in blocks. The problem is while the controller can erase an entire block in one shot, it cannot however write to a cell if any other cell within the same block is holding data...

Um, I am not sure that is the case. I would suggest double check.

[Phil]

Source:
Q&A: Tom's Hardware And Kingston On SSD Technology : Kingston Up Close

[tilleroftheearth]

Nice info Phil.

Just pointing out though that this was early 2010 when nand was 50nm and 34nm based.

Not to mention the much lower (Kingston's) 4K R r/w's that greatly affect write endurance. With a strictly random 4K r/w load (the 'harshest' nand can be subjected to); the write limits of all SSD's are nowhere close to that chart.

But it does illustrate nicely how capacity quickly masks all normal (and even my 'un-normal') write needs.

Even though it is 'only' a linear function; it quickly grows past the point of 'consumer' usage patterns and begins to handle professional use quite well.

If manufacturers were clear about the tradeoffs made to get their product to market, we could be making a solid choice of which SSD to purchase for our use.

Intel is the only one who consistently communicates and delivers products that focus on reliabilty just as much as performance. All others are simply chasing the latest 'benchmark' fad to push their stuff (when they're not outright lying).

With all the parameters that can be tweaked, adjusted and manipulated (via the controller/firmware), the raw capacity of SSD's will be a clear indication of the reliablity of SSD's going forward.

[eYe-I-aïe...]

Um, I am not sure that is the case. I would suggest double check.

I learned that before I actually bought my first SSDs, so it's a no brainer for me, it's already checked and double-checked.
I'll however get back to you on this. → See here

[Rachel]

I'm looking for a 60GB-64GB SSD with TRIM, at least a balanced performer, low power consumption and it must be under £100.

Kingston i've noticed have come out with some new drives. Well, new to me since i last checked lol. I like the V+ but that is just over £100. I'm unsure how the V100 drives are power consumption wise.

Here are some drives i've seen.

Corsair CSSD-V64GB2-BRKT Nova
Corsair CSSD-V64GB2-BRKT Nova: Amazon.co.uk: Computers & Accessories

Integral 2.5 inch 64GB SATA II MLC
Integral 2.5 inch 64GB SATA II MLC Internal Solid State Drive: Amazon.co.uk: Computers & Accessories

Kingston SSD S100 new V series
Kingston SSD V100 64GB SATA2 2.5inch Hard Drive: Amazon.co.uk: Computers & Accessories

OCZ OCZSSD2-1VTX60G
OCZ OCZSSD2-1VTX60G 60 GB Vertex Series SATA II 2.5-Inch SSD Drive: Amazon.co.uk: Computers & Accessories

Which drive do you think fits the bill the most? Integral is unknown brand but it's a cheap drive so i won't overlook it. The Vertex is also very reasonably priced but i'm unsure if it fits all of my criteria.

[Cape Consultant]

Dave p, what do you think of an Intel 320 160GB for my next drive? My NOVA crapped out badly Thanks.