"SSDs are not the magic bullet that some would have you believe"

Regardless of any tech specifics a 5-6 year old 5400rpm Spinpoint is by no means faster then a newer 7200rpm WD Black. So the slower and much older drive gets a 5.9 and the faster drive should by contrast score a .1 or greater at the very least. But MS wants us to believe in fairy tales and help push new shiny tech we don't really need. Because they know people will pay for higher index scores as they do with system benchmark scores.

No doubt SSD has a place in the market but it's not the most cost effective storage medium for every application. The average joe has no need for SSD imho and as long as the masses keep throwing away money the prices will remain at a premium. Lest also not forget that not all SSD are created equal and performance will in fact degrade over time. Meanwhile mechanical drives tend to be at or near the same speed until they give up the ghost. So in many cases the added money for SSD might be better spent on more memory, faster processor, better motherboard and or much greater storage capacity.

I had the SSD in my MacBook Air which was solely used for Windows on bootcamp and my own work software which is pretty hard drive intensive. I found no speed improvement over a 7200 rpm hd. I do find a huge difference between a 5400 and 7200 rpm hd though. One time, years back, I equipped a Dell tower with two 10,000 rpm Raptors. They were screaming fast but burned out fairly quickly.

Any thoughts on the 3tb fusion drive Apple says is the same as having an ssd?

Also, figured I'd ask, I know it is very easy, and much cheaper, to upgrade ram on the new iMacs than to buy it from Apple. But can the hard drive be easily upgraded? I saw a Youtube video showing it done on the older, fatter model, and it required removing the front glass.

isoMorphic wrote in post #15802233
Any SSD even if it was the same speed as a standard 7200rpm would raise the index rating simply because that index rating is a gimmick. Microsoft wants people to upgrade systems just about yearly which in turn nets them more sales of Windows. So it basically detects certain hardware flags and prevents specific types of hardware from getting high marks on purpose.

My mother has a first gen i7 and gets a 7.9 on her CPU and 5.9 on her 7200rpm drive. I have a third gen system that nets the same 7.9 score cap which a $50k supercomputer would get. Yet my WD 64mb cache 7200rpm drive scores 5.9 which is the same exact score given for my slower Spinpoint 5400rpm 32mb cache drive.

There are laws of physics regarding a spinning platter...That is not a gimmick.
1) SSDs --no spin-up is required
2) SSDs offer fast random access for reading, as there is no read/write head to move
3)Extremely low read latency times, as SSD seek-times are orders of magnitude lower than the best current hard disk drives. Look at a desktop HDD and you will see a number that may be ~11ms seek time

There are drawbacks to a SSD. In my opinion the main ones are:
1)flash-memory cells will often wear out(All manufacturers use some sort of wear leveling algorithm)
2)SSD prices are still considerably higher per-gigabyte than traditional HDDs
3)Lower storage density when compared to HDD

Regards,

110yd

110yd wrote in post #15804949
There are laws of physics regarding a spinning platter...That is not a gimmick.

I think you are missing the point as faster hardware is still faster and should in fact score higher. You can't tell me a 5400rpm with half the cache should be given the same score as a newer and superior 7200rpm as it does in my case.

I understand SSD is faster but no SSD is going to be given a measly 6.0 score either which makes the gap in Windows performance scoring pretty misleading as a performance gauge. They purposely hold back a 10k Raptor at 5.9 the same as a 7200rpm and the fastest 5400rpm which makes the drive score itself almost meaningless.

http://community.wdc.c​om …low-sub-score/td-p/249288

If someone wants to upgrade hardware based on what Microsoft thinks is suitable that's great for hardware makers. I like to see factual numbers from benchmarks and what Microsoft tells us is obviously not something that is entirely true. What I know is that I don't need SSD as there is nothing my drives do that would be done any better at three to four times the cost.

That's not to say next generation SSD's or Hybrids wont be a total game changer and blow mechanical drives out of the water in every way possible including write endurance. But current gen SSD drives are not designed to constantly be filled near their total capacity as with mechanical drives which are designed for bulk storage. In fact all data on SSD is shifted periodically for wear leveling which in turn uses write cycles and lessens the overall life span just by storing data.

It doesn't matter if you've got a 10,000 RPM raptor or a 5400 RPM storage drive. Your seek times are going to be piss poor. In the current consumer storage industry, the statistical performance difference between handful of drives with 7ms seek time (I think maybe 6,7?) and 11ms seek times is completely skewed by the .2ms seek times of the plethora of SSDs on the market.

The item that is being overlooked is the SEEK times. On a spinning platter you are going to have delays that regardless of how fast the platter spins, there will be milliseconds added. The WD Velociraptor has seek times that are half of a normal HDD. This amounts to a delay of ~5Ms. The SSD is a clear winner speed wise if you look at the specs.

Regards,

110yd

As I understand it, the RPM of a drive does not affect the seek times much, only the rate which the data is pulled off the track. Once it arrives at a track, the head has to wait for the start of sector to come around, and this is where a higher disk RPM helps a bit. Several other variables come into play, such as the platter diameter, and number of platters. Data organization on the disk also matters, as seeking across more tracks increases track-to-track time of the head/arm. Defragging a disk can help somewhat in this regard.

In regard to SSD's, can someone clarify the advantages/disadvantag​es of SLC versus MLC structure? I notice on Newegg that some of the latest Samsung drives allege to be MLC, but not sure this is true.

Another large advantage of SSD's is the durability, particularily in a portable device. Working for a school district, we have a high rate of HDD failures due to the rough treatment by students. Sure they have to pay for damage, but it is nearly impossible to prove shock damage to a hard drive.

The SEEK time (plus settling) isn't change AT ALL by the rotational speed. The rotational latency is.

Now here's another gotcha -- a 5400 RPM 3TB drive will OFTEN be faster than 500GB 7200 RPM drive, sometimes by a LOT.

Guess why? Areal density. That is, transfer rate (once you have seeked) is controlled by the number of bits that pass under the head over a unit of time. The more dense the data on the platter, the more bits pass. Therefore, a FASTER spinning drive can actually transfer SLOWER!

I have older 250, 320, 500 and 750GB WD RE-series drives here in service, all 7200 RPM, and the cheap 5400 RPM 2TB Seagates blow them out of the water on transfer speed. They're comparable in seek times too. These are facts "as measured", not theory. So no, it is NOT a foregone conclusion that MIcrosoft is "cheating." I have very high-end adapters that can sustain ridiculous I/O rates if the drives can keep up; they're the limiting factor in my testing (but just barely with a whole bunch of SSDs connected to them.)

SLCs are more durable than MLC drives and also can erase smaller blocks, which can matter on rewrite performance, but for consumer use it is not particularly material. I have OCZ Vertex drives that are several years old and yet show nearly new life statistics. The controller knows how much spare space it has and when a write fails to "take" it allocates from that pool and uses it. The wear pattern is known and pretty predictable -- SSDs actually will wear out in a way that you can detect BEFORE they fail most of the time where a spinning drive usually fails with little or no warning first.

SSDs are almost-indestructible. A spinning drive will be destroyed by ~100G shocks (which is a drop of a surprisingly small distance -- a couple of INCHES -- onto a hard surface) if running and the heads are unparked, although if parked they will usually (but not always!) survive MUCH more abuse.

jemanner wrote in post #15805569
In regard to SSD's, can someone clarify the advantages/disadvantag​es of SLC versus MLC structure? I notice on Newegg that some of the latest Samsung drives allege to be MLC, but not sure this is true.

http://www.computerwee​kly.com …lash-SSD-is-right-for-you
http://tech.slashdot.o​rg …ok-at-ssd-write-endurance

All I know is that my batch processing time was hugely improved by working off of SSD. Again, once the photo is loaded in Photoshop, I don't feel any improvement at all, since the work is being done in memory.

I've got a 128GB SSD for the OS, a 256GB "work drive" for my current photo projects, and a 256GB "cache" drive. I also have a few Win7 VMs running on the "cache" drive when not doing photography work.

Another note...CS6 seems quicker than CS5. Perhaps it's because it has better support for Open GL video acceleration. (And OpenCL...which my card doesn't support.)

Speed is one thing.

Reliability is another one. IMHO much of consumer SATA SSD is still outright crap. And when they fail they fail brutally with the controller down. Regular HDs don't only often allow you to retrieve much of the data after a fail, they also tend to have occasional read errors or other things that you can pick up in S.M.A.R.T. before going down.

Raid over SSDs is an entirely matter still.

Given the speed at which SSDs developed I am very pessimistic about lurking firmware bugs.

tickerguy wrote in post #15805740
The SEEK time (plus settling) isn't change AT ALL by the rotational speed. The rotational latency is.

My mistake, I was looking at seek time as the time it takes from reading one record to the next, but you are technically correct, seek time is only cylinder-to-cylinder. So, can I presume seek time + rotational latency = total time between accessed sectors/records?

Well, yes and no Jemanner.

The usual strategy is to lay out tracks so that sequential one-track seeks (in or out) suffer as little rotational latency as possible. But even so with a 7200 RPM drive you have 120 rotations/sec, or about 8.3ms/rotation.

The maximum transfer rate is set by the areal density of the bits that pass under the head during that time, and the maximum sequential transfer rate across tracks is controlled by the fact that once you have to move the head you take the latency of doing so plus whatever the gap is to the next sector you wish to read.

If the track-to-track latency on a seek is 8ms then laying out tracks with no gap may prove to be close to optimal. If it takes 4ms then you might want to lay out tracks to be offset 180 degrees for their starting sector, and so on. The other governing factor is that drives today all have cache memory and their firmware will attempt to use it for the purpose of optimizing out as much of the latency as possible.

Back in the 1980s and early 90s the best way to get relatively "screaming" performance out of inexpensive Winchester drives was a secret that few PC builders understood. We built them for customers with track-buffering controllers where pretty-much nobody else was and in addition drives were PHYSICALLY low-level formatted in the field. By knowing the performance characteristics of our drives and controllers I was able to optimize the physical format to match the disk controller and as a result we had monstrous performance gains over the other common systems sold into business markets. Our sequential (e.g. program load time) I/O was often 3-5x faster than our competition as a consequence. I made a LOT of money by understanding how these things actually work and how to make a difference the actual user would see rather than playing "spec games."

Nowdays rotating media all have buffers on the drive that take care of this. It matters a LOT which is why the disk manufacturers all use buffer cache on their disks.

I've never had an SSD go down without warning. I have had a couple post up warnings that (since I pay attention to them) resulted in me junking them before they failed. I've been using them now as boot + OS + scratch drives for several years, along with using them as volatile cache drives attached to RAID 1 and 10 arrays (you have to know what you're doing with this or you WILL lose data!) in certain circumstances, particularly for database servers. When used properly they provide ridiculous performance improvements, particularly for the price.

Conventional HDDs fail all the time and 90+% of the time when they do so you get ZERO effective warning before data is irretrievably lost -- your FIRST warning is an unrecoverable read error. S.M.A.R.T. provides effective warning of failure rarely, but when it does you better pay attention to it because failure is likely hours or days away, not weeks or months. I expect that rotating media will run for about 2 years before no-warning failure in 24x7 service. Sometimes I'm disappointed and sometimes they run for what appears to be "forever" with multiple instances of the same unit in the same case being fed by the same power and environmental conditions having wildly different outcomes.

In my opinion and experience SSDs from quality manufacturers (e.g. Intel, OCZ and a few others) are more reliable than HDDs, in that absent firmware bugs (which incidentally strike HDD firmware code too, and when they do they cost you everything on the disk as well -- there was a bad run of WD drives a few years ago that had a 100% failure rate within a few months that ultimately was traced to the firmware!) they are FAR more likely to give you clear warning before they fail and data is lost. They also have dramatically superior environmental (e.g. shock and temperature) resistance.

In terms of performance they have an effective zero seek + settling + latency time. They make a HUGE difference in program load time (e.g. system boot and program startup) along with any operation that is read-heavy, especially random reads (compared to a HDD.) They are less-effective in write-heavy workloads as once their internal cache fills their actual effective I/O rate for writes is quite slow.

Their Achilles heel is that due to how flash memory works writes are inherently slower than reads as a write to other than an erased block requires three operations (read/erase/write.) With a HDD writes and reads take the same amount of time. For this reason using a SSD in nearly-all-write environments, especially when the writes are sequential (e.g. a DBMS write-intent log) they're inappropriate since once the cache fills they slow down a LOT and in some instances of this sort of workload can be materially slower than a HDD.

For consumer PCs my advice is that a SSD as a boot + OS + program + cache drive is one of the most cost-effective improvements you can make to your computer, second only to having enough RAM that you do not page. If you have a read-cache-heavy workload adding TWO SSDs, one for OS + program and a SECOND for that read cache will add even more user-perceptible performance, but the benefit there is smaller than for the first one. For best performance SSDs need to be kept under 80% full and the OS must support TRIM to tell the drive about released space as this allows their firmware to manage and prepare for writing all unused space on the volume in the background. OS's without TRIM support or systems where you run a SSD beyond 80% capacity will suffer material and progressive performance degradation to the point that IMHO you should not use SSDs without being able to meet those two requirements.

Thanks again tickerguy, a LOT of good information to digest. I never even thought about laying out tracks 180 offset, but it makes good sense. When I build my next PC (soon!) will for sure have an SSD as the OS/app volume. Haven't decided regarding data (the bulk of which is photos, some 270 GB and growing fast). If I can afford it will probably go RAID 5 with conventional HDD's, and with a periodic off-site backup.

uOpt wrote in post #15805872
Speed is one thing.

IMHO much of consumer SATA SSD is still outright crap. And when they fail they fail brutally with the controller down. Regular HDs don't only often allow you to retrieve much of the data after a fail, they also tend to have occasional read errors or other things that you can pick up in S.M.A.R.T. before going down.

You're halfway there. Not only are most SSDs crap, so are most disk drives. The drives available as "desktop drives" are mass produced commodity items, with one simple engineering design criterion: to outlast the operating system installed on it. The aren't particularly high tech, and are certainly not designed for reliability, simply because the desktop market is, as Western Digital puts it (when they explain why you shouldn't use desktop drives and expect long term reliability), "a price-sensitive market." This is simply a description of a market that values reliability, but doesn't want to, or simply wont, pay the price necessary for actual reliable disk drives. If you check the prices/types of the disk drives used in commercial server environments, you'll find they are much more expensive than WD Blacks - as for reliable SSDs, HP makes a commercial 800 GB SSD - it sells fro $6400 bucks. There's a reason that enterprise level components cost more than "desktop" drives, and it isn't necessarily speed.

These days, if a drive outlasts Moore's Law, the odds are excellent that you,as a user, won't use it in your next machine because it's too old, or too small, but not because it's too slow.

Here are the two primary rules of disk drive usage:

1. All disk drives fail.

2. The larger capacity the drive, the more data you will lose.

Full Stop.