SSD performance and consistency test with overall performance

these guys at TT are very knowledgeable in SSD field I urge everyone to head over and check out their latest test method and SSD performance. The information they provided are explained and easy to understand. SSD Consistency Testing - 18 Drives Tested for Real-World Performance | TweakTown

 

Wow, game loading times are all over the place from SSD to SSD which I find quite surprising. I mean 100MB/sec a proper defragmented hard drive can manage that, as long as the file sizes are not real small and more or less sequentially stored on the drive.

 

yea it's pretty amazing, and weird.

I went through their Toshiba Q pro series SSD review and it doesn't have great numbers and IOPS, but the result for their windows 8 application seems to be off the chart and doing way better than many other SSDs.

I'd like to think there something much more out there than just faster flash and controller. Maybe a good firmware and garbage collection algorithm also affect SSD's performance in multiple directions.

 

Too bad this wasn't a real, real world test - still, it does show that SSD performance is variable and gives some weight to my comments from up to three years ago that 'any SSD' is not the same as a 'good SSD'.

In many of these tests; the 840 Pro, considered the top drive by others; still sucks and we now have the data to prove it.

Too bad this test didn't do a power/heat comparison of the drives too so that all this relevant data would be in one place (and tested identically).


The SanDisk Extreme II is king. Especially in the 240GB capacity point (I know that too).

The M500 is not only too warm for my mobile use - but the performance is not up to today's standards by a long shot (I never got that far (testing performance) in my own testing because of the heat issues). In either of its 240/480GB capacities.

The EVO 500GB may be okay if it is not pushed (what other reason do we need an SSD for? To push it...) - but TLC nand still leaves a lot to be desired imo - especially when the SanDisk Extreme II 480GB model regularly goes on sale for $299 (since boxing day).


The Toshiba Q Series Pro is interesting (although too small for my needs today) - but it is also a relatively obscure model at this time.


I have been talking about real world performance of storage subsystems for over 4 1/2 years now on this forum alone. The findings in this article are very closely aligned with what I have personally seen (I haven't tested all the SSD's they have though) in my own use, testing and or otherwise playing with other systems with SSD's installed.

I can't wait to see the next 6 SSD's they're talking about (does anyone else know what's being introduced in the next 45 days or so) - it will be interesting to see if the consistency has improved over the current crop of SSD's we are able to use.


In the end; a good SSD experience is again a matter of balance.

Power use (the lower the better - especially under load)
Performance (in all aspects, not just a freshly SE SSD without any data on it and tested in Windows Safe mode, sigh).
Heat output (not just a number in Celsius; but a real world use of the drive inside a notebook's sealed drive cage).


It is interesting to note that the time to run these tests is almost 24 hours.

See:
Real World SSD Performance - Why Time Matters When Testing | TweakTown

The interesting part is almost exactly how long it takes me to 'judge' how good a new SSD is in my own testing - although the 24 hours are spread over the course of a week (roughly) instead of a continuous session.


I hope they run this test with a standard HDD at some point soon (a TravelStar 7K500 would be perfect) to show the relative difference better between HDD's and SSD's.


Good find unityole!

 

yes, it saddens me whenever I talk to someone about SSD they all point to samsung 840 pro. imo it is still a very fast drive if you don't write it heavily but with application and games nowadays, those are definitely heavy writes. installing, uninstalling it, video encoding etc. Evo was done very well with SLC boosting it's performance and I think it is the only reason it does well on that chart, rapid mode doesn't do a thing, yet people still look at only numbers off a CDM or HDT results.

the 6 drives they talked about interest me too, I think they are under NDA and can't talk about it just yet. I can't wait for them to release test results from their SLC S301 128gb SSD, although it is useless for it's size but I have two in raid so it works out.

the results aren't real life world test for sure but thumbs up to them trying to replicate something similar to it, and to tell the truth it is much MUCH more real than Tomshardware and Anandtech results. as for why I won't talk about it but queue depth makes up most of it.

their battery and thermal tests are covered in their SSD review.

 

Hello wingnut, just so you know if I run a hdd with the same test it will be lucky to hit 10MB/s. The way you are thinking of bandwidth is not the way the test measures it.

 

You have the updated PCMark 8 testing suite they used?

Still, I wouldn't care about the 'score' so much as how long it would take to run the test (i.e. would the ~24 hours become 48 hours?).

 

I am a storage editor at TweakTown.com a HDD would take day's to run a good ssd like the q series pro etc takes 13 to 15 hours to run. The slower ones like 840 Pro's take close to 20 hours to run.

 

Jon, thanks for the info.

No way no how you could run a HDD through the suite for an old techie? Please?

 

Okay it will be interesting I will do it in a week or two
I will be suprised if a hdd can exceed 5 MB/s

 

My point was that the game load times should represent actual game load times, or storage drive performance (as measured in MB/sec) in doing so. If it doesn't, it's a useless test. What would be more beneficial is level load times. I know from running off SSD or HDD, that loading off HDD can take about twice as long as from SSD, but it's still only about 20sec vs 40sec, not 20sec vs 200sec as you would suggest.

 

imo this is one of the biggest problem with benchmark software and I believe companies target and to learn how these program works and release their SSD firmware and build off of it. marketing is important for them but does nothing for consumers, samsung rapid mode was such a thing and I honestly believe it had no effect. This made me believe a realistic test would prove everything, which is a good thing. Anandtech hasn't change their test for a long time can be a target and also why thumbs up to Tweaktown for making a new different test.

 

I was in the market for an SSD, like most, i was looking at a crucial or samsung, but now sandisk seems like it is worth a look

 

I'm also considering the SanDisk Extreme II but curious to see what will come out of the woodworks in the next 45 days.

Why the slight performance difference between the 240 GB and 480 GB models?

 

The performance difference between the 240GB and 480GB models is a factor of:

The optimized use of the channels available. (Both are equal here at these capacities is my guess, but there still may be subtle variations of how those channels are populated too).

The nand interleaving between each channel. (The 240GB model has the superior or optimum interleaving ratios with regards to how it physically is constructed and to how efficiently/directly it can interface with the controller/firmware).

Most importantly; the firmware has been tweaked the most for the 240GB model (it has simply been around longer than the larger capacities).


I am still hoping for a 10 or 12 channel (better; 16 or 20 channel) based SSD controller that will be more efficient and with higher performance at the larger capacities as SSD manufacturers drop the useless smaller capacities in favor of a capacity that makes sense for 2014 and on.

Even though these smaller capacities were already too small half a decade ago.

I would also like to buy a 240/250GB SSD with ~80GB additional nand dedicated just for OP'ing (and a 480/512GB SSD with 160GB OP additional nand). With no option of using the whole capacity. Or, at least the option of a 320GB/640GB/1280GB SSD so that the capacity for the user can be comparable to a HDD when OP'ed properly. I'm just tired of having to explain to clients that the SSD capacity they're buying is not the capacity they can actually use (at least not long term; at anywhere close to the performance advertised).

While manufacturers have figured out that 7%, 12% and even 4% OP'ing is enough to keep the drive fast (at least long enough for benchmarks to 'prove' it is fast; see!), I still have not seen a drive that hasn't benefitted from OP'ing by a large amount (25% or greater in addition to what is already built in).


The new Intel 730 Series (desktop only...) SSD's have some amazing consistency - and become phenomenal when OP'ed by 25% (writes). This tells me that how they're shipping them, they're setup wrong.


See:
AnandTech | Intel SSD 730 (480GB) Review: Bringing Enterprise to the Consumers

(See the last graph for how the 730 Series just destroys (with 25% OP'ing) anything we've had previously).


I would be willing to pay for the additional capacity as built in ~30% OP'ing - it is a tiny, one time cost (~$40/$80/$160 for the 240/480/960GB capacities). The benefits over the lifecycle of the drive far outweigh this additional cost.

Right now; the cost is capacity to get those benefits. I want to simply trade dollars for the capacities that today's workloads (mine) demand.


My default setups:

O/S: ~10GB (Win8.1x64 Pro no pagefile, no hibernation file, no System Restore and no error reporting junk).

Programs: ~60GB (Office 2013 Pro, CS6 MS, Acrobat 11 Pro, accounting and misc programs).

Free space: ~25GB minimum (35GB is a better 'working' number, but Windows complains too much when it is less than ~25GB).

Scratch space: ~50-100GB (depending on the PS projects being run).

Data: ~xxGB's - this is the reason to have a computer!


If we add up the above and estimate the active, WIP (work in progress) or current data to be (conservatively) ~50GB; we have a drive capacity right around 250GB.


Today, I have to buy a 480GB SSD or higher to accommodate my needs at the performance level I demand (with a usable capacity of ~313GB; sure some extra capacity (over 240/250GB) but the drive just cost me more than double too).

I'd rather spend only ~$40 more than a $240GB drive costs to have the same performance and at a usable capacity I need instead.

And; probably be faster than the 480GB drive too (especially if the controller and firmware were further optimized for built in 30% OP'ing, right out of the box).


TLC nand is not the way to an enthusiast's heart (promising less cost but no additional benefits): building a product that addresses real world concerns is (consistency and actual usable capacity at an absolute, truly lower price point).


Hope I answered your question?

 

@tilleroftheearth, SSD expert lol.

I always wondered whats the importance of consistency test on Anandtech reviews. all these consistency tests runs at 32 QD which we'll never reach, perhaps one day we'll see a consistency test for QD 1-4?

 

Nah; far from expert. Just had to do my own digging to get to the bottom of all the SSD myths (see my first few posts circa 2009/2010 if you want to laugh/cry at the responses to my questions).


I agree that QD 1-4 will be a better scenario to test with for workstation workloads. But it is the difference between SSD's that I'm interested in though; not the absolute 'scores' they post.

And before we saw even the QD32 benchmarks; I was saying how slow/laggy the 840 Pro can feel (while everyone else was calling me @$%@$%$@$, ah; names).


Balance baby! In life, in love and in computers and storage subsystems too!

 

This test is similar to storage bench 2013 not the consistency test. It is running at low queue depths for the most part during the traces.