Slashdot Mirror
Intel's New Desktop SSD Is an Overclocked Server Drive
crookedvulture writes "Most of Intel's recent desktop SSDs have followed a familiar formula. Combine off-the-shelf controller with next-gen NAND and firmware tweaks. Rinse. Repeat. The new 730 Series is different, though. It's based on Intel's latest datacenter SSD, which combines a proprietary controller with high-endurance NAND. In the 730 Series, these chips are clocked much higher than their usual speeds. The drive is fully validated to run at the boosted frequencies, and it's rated to endure at least 70GB of writes per day over five years. As one might expect, though, this hot-clocked server SSD is rather pricey for a desktop model. It's slated to sell for around $1/GB, which is close to double the cost of more affordable options. And the 730 Series isn't always faster than its cheaper competition. Although the drive boasts exceptional throughput with random I/O, its sequential transfer rates are nothing special."
Hard for any SATA drive to distinguish itself on sequential transfers, given that SATA is capped around 550MB/s
More data, damnit!
For many (but certainly not all) applications, especially when it comes to UI, what matters is 95% worst performance, not peak throughput. From the Anandtech review, that's where this drive really shines.
Different tradeoffs have to be made for different workloads -- it can't be boiled down to a single (or even a set of) number(s). Some applications are far more tolerant of worst-case performance than others.
Running something at the speed it was designed and verified to run at by the maker isn't overclocking.
That's an excellent point, and a metric I hadn't paid much attention to despite the fact that I run quite a few drives, including one storage pool of 28 drives and growing.
Don't get me wrong, I own five discrete SSDs (all currently in active use), and they're all Intel (one G1, two G2s, and two 330s). However, I've been disappointed with Intel of late. It used to be that they came with a premium price, but also dramatically lower failure rates than the competition, and you could usually find them cheaper than the competition if you waited for the right sale.
These days, however, Samsung's failure rates are lower than Intel's, and their price premium is so large that no sale is going to get their larger SSDs anywhere near as cheap as Samsung's. I was hoping that they might make a comeback with a new consumer model, but the 730 is a disappointment in terms of its extremely poor performance-per-dollar and capacity-per-dollar.
I've bought nothing but Intel in the past, because they were the safe bet, but at this point it looks like my next SSD will be from Samsung.
Yeah, anything less than 1 Farad is just ridiculous. No way I would hook my Blaupunkt up to that thing...
I only wish Intel was offering this in a smaller size, say 100 GB. I think a SSD system drive + slow "green" HDD is a great combo in a desktop, and the price premium on this quality of SSD would be easier to swallow if the drive were $110 instead of $250 even though that would be the same $/GB.
His point was that the general capacitor quality these days is a bit dubious. Of course there are premium brands in all electronic components.
tl;dr: these are storage caps, which don't endure the ripple currents that kill filter caps.
Electrolyte decomposition is usually caused by high ripple current, which is why caps pop mostly (only?) when used as filters, as in motherboard DC-DC converters and gadgets powered by wall-wart adapters. In this particular application, the PSU impedance is quite low and the caps are handled by on-board regulators (V=Q/C and all that), so there's no load ripple and the caps just have to sit pretty and charged with insignificant heat losses until the computer is shut down or outage occurs. Maybe that's why Intel didn't even bother to use the solid (polymer) kind.
If these caps dry out due to age or bad quality they just won't hold as much charge for emergency sync'ing, which is still better than ordinary SSDs/HDDs with no caps.
This post contains no rudeness or derision of any kind. All arguments are friendly. Terms and exclusions may apply.
The caps only need to supply enough juice to sync the RAM buffers to flash to ensure consistency of its internal block-mapping metadata (the filesystem should handle the rest through journaling and whatnot). The caps are rated at 35v but let's assume that they're kept at 12v: E = (12 v)^2 * 47 uF / 2 = 3.4 mJoules. Even at full operating load that should last for half a millisecond counting losses, but when power goes out the drive is going to stop serving requests and all it has to do is write that 1 GB buffer to a few flash blocks. More than enough, methinks.
This post contains no rudeness or derision of any kind. All arguments are friendly. Terms and exclusions may apply.
Except those caps are Nippon Chemi-con. High end high quality capacitors made in Japan. And not the kind involved in the bad caps.
Bad cap syndrome happens to the cheap caps - stuff like CapXon (aka CrapXon) and such.
In fact, a lot of bad caps you're finding are the cheap crap ones by the crap manufacturers. You can easily buy them and they will fail.
That's why you'll find people inspecting caps - and seeing if it's Nippon Chemi-con, Rubycon, Panasonic/Matsushita or other Japanese brand. (You can almost generalize it to those whose brands contain "con" in their name are higher quality - from when they used to be called condensers. The cheap brands all tend to have "cap" in their name).
So no, I don't see the caps being the weak point because Intel went and spec'd top-quality caps.
> Although the drive boasts exceptional throughput with random I/O, its sequential transfer rates are nothing special."
But good random access will give you better overall performance in most cases. You rarely need to deathmarch through the drive.
Oliver's law of assumed responsibility: If you're seen fixing it, you will be blamed for breaking it.
What you discover with SSDs is that for desktop usage pretty much any drive is "fast enough" and that faster doesn't much matter. I went from a SATA-2 SSD that was fairly slow even for that generation (WD Siliconedge) to a SATA-3 SSD that is fairly fast for this generation (Samsung 840 Pro) and I don't notice any difference. I can benchmark a difference, but I don't see any difference in load times and so on. SSDs are fast enough that they are making themselves not the bottleneck.
That's also why there isn't a ton of interest in the PCIe SSDs. You can get way more performance, but it is a somewhat limited set of scenarios (on the desktop at least) where that would matter.
I think maybe because it is something that can be easily shown off, or because it can be done cheaper, or because they have a misguided belief that it makes everything fast.
Personally if I can't afford an SSD big enough to stick all the apps I normally want on there, I don't bother with an SSD in a system.
His point was that the general capacitor quality these days is a bit dubious. Of course there are premium brands in all electronic components.
Well, I don't know.... his point sounded more like, "Oh, remember those awful Firestone tires from 2000? I'm never using tires again - they all just blow out. Tank treads all the way for me."
Nothing wrong with solid caps, but the premise for the argument is a little weak.
There are two types of people in the world; those who believe there are two types of people, and those who don't.
There is a new standard which will increase SATA speed ( http://en.wikipedia.org/wiki/S... )
Currently, Apple computers use PCIe SSD disks, which increases their performance:
http://www.anandtech.com/show/...
"I'm very pleased with Apple's PCIe SSD, at least based on Samsung's new PCIe controller. Sequential performance is up considerably over last year's 6Gbps SATA drive. Go back any further and the difference will be like night and day, especially if you were one of the unfortunate few with an older Toshiba drive. Internal transfers are quicker, but to actually use the new SSD to its potential you'll really need a very fast external Thunderbolt array - even USB 3.0 can't completely tax it. There's still a lot more investigating that I want to do on Samsung's new controller, but my early results look very promising. It's sort of crazy that Apple now ships a mainstream consumer notebook with a PCIe SSD capable of almost 800MB/s. Now that Apple is off SATA, scaling storage performance should be much easier to do going forward. "
A quality electrolytic capacitor will last a long time.
The ones used here look like Nippon Chemi-Con, rated at 105 C. They'll most likely last forever.
To be fair, an HDD can use its platters as a flywheel to quickly flush its (relatively tiny) buffer. I never did see proof that that was ever done, though.
Plus they are one of the few components left that anyone with a soldering iron can diagnose and replace, even in this age of surface-mount stacked-chip fiddleyness.
At 90% health too. intel ssd drives are worth the premium
Seek help for that.
Intel (and everybody else) does this for good reason .. high endurance components (Milspec, server, whatever) are usually designed for tolerances far beyond the actual spec, because manufacturing issues can cause the tolerances of the finished product to deviate somewhat.
.. and the failure rate there doesn't really matter as much because you just print a disclaimer about "your data may go poof" and RMA the broken ones. As long as the defect rate is low enough to remain profitable, yay again.
If they design a [gizmo] to operate at 1.5ghz and sell it as a 1ghz chip knowing full well there is plenty of overhead but chances of failure running it at 65% of design are pretty much nil, yay for them for meeting the rejection rate.
Then along comes marketing and says "hey, we can sell the rest of them at 1.5ghz as consumer units"
* This is effectively regenerative braking, which I'm not sure you can do with a stepper motor.
* The arm servo needs extra energy; not sure the platter+rotor have enough.
* What if it's stopped, heads unloaded, when the power fails?
This post contains no rudeness or derision of any kind. All arguments are friendly. Terms and exclusions may apply.
The regulators are there - you can see the 'big' coils in TFA. The capacitance for a simple "drain till you drop" scheme would have to be a lot here - very roughly 2*Energy/efficiency/(Vddmax^2 - Vddmin^2). So, step the voltage up optionally, keep the caps charged as high as practical, squeeze them dry when needed through a step-down converter.
TFA also says that the drive periodically monitors the "status" of the caps; I'm not sure if that means charge level or charge-holding capacity but it could test-discharge one cap at a time.
This post contains no rudeness or derision of any kind. All arguments are friendly. Terms and exclusions may apply.
If the heads are unloaded, there shouldn't be any operations going on, so no harm done (unless some genius decided that it wasn't worth it to immediately get the heads in place the moment data comes in). Note that few hard drives actually spin down, but again, if it's not spinning, there's no data flowing (unless you're really unlucky and it just started).
None used it to flush the cache because it is too risky - the platters are not maintaining a fixed speed (they're slowing down to generate electricity) so writes to platters become tricky as the timing is off which means you can overwrite more than you expect.
Far better to just dump the buffers.
In fact, the electricity generated by the spinning platters slowing down is used to park the heads - it's called an emergency head park because it basically dumps the electricity into the voice coil that flings the heads to the mechanical stops in the park area. It's fairly violent and most hard drives have much less emergency head park life than standard power down (where the drive moves the heads to the parking area in a controlled fashion) life - a drive may have 50,000+ head load/unload cycles, but under 10,000 emergency park cycles.
You can tell because a soft-park makes only the smallest of clicking sounds on a drive when it spins down. But emergency park it and it's a much louder clunk.
On a side note, it has been YEARS since I've witnessed a laptop that actually had the chance to spin down it's drives. Probably since win3.x days. Software being what it is today (McAfee at my place of employment) seems to have this bug of reading/writing to disk every few seconds, defeating any power management setups.
Karnal
On a laptop, it may make sense to enable Windows' write cache, since there's a battery backup available. That may help you save some power.
Eh. Except that a blue screen or if you have to do a forced power-off will lose data and require a chkdsk run. I ran my desktop that way for a little while (with a UPS) but it had problems.
Mostly right. Any data in write cache is discarded (which is fine, if the OS wanted it written that badly it would have issued a flush/fua/... and that won't get acknowledged as done until the data is actually on platter). A currently active sector write *is* finished using mechanical energy of the platter (and on newer drives also with the help of electrical energy from capacitors ... that was one of the challenges in moving to 4k sectors).
Think about it, if this wasn't done a mid-write power loss would cause a unreadable sector, as you'd have a mess of part new and old data and ecc on platter.
Many non-sas drives will falsely return that the flush is complete. Unless you look at the firmware on your drive, or do some statistical analysis of your write latencies, you can't really know.