Slashdot Mirror

Given a choice between malware outbreak and killer multi-patterning issues I would take the malware outbreak every time. But I would also instantly ban Windows inside the corporate perimeter, it's a simple cost benefit thing. Lord help them if they store primary engineering assets on Windows machines.

It only took a single full day meltdown for the London Stock Exchange to learn this lesson.

https://techreport.com/news/33...

It's not possible recall all the processors that ever existed. Society doesn't have the resources even to think about such a thing.

A recall actually wouldn't be that expensive for Intel. Most CPUs and SoCs with the same die size as Intel's CPUs sell for only around $20. The $100 to several $1000 Intel is able to command for their CPUs is due to marketing and them being top dog. The material cost of the Intel* CPUs themselves is only a tiny fraction of their retail price. (Which raises the possibility of Intel doing a recall, taking a charge for it on their books based on the retail price, and getting a tax deduction for the "loss" which exceeds the manufacturing cost of the recall.)

The bigger expense would be in the time to manufacture the CPUs (it would tie up their fabs for years) and the labor of actually replacing them. Desktops are pretty easy, but laptops with soldered BGA CPUs would be difficult to impossible. Another problem is that there simply is no performance-equivalent replacement for the highest-end CPUs. Any fix severely curtails performance, so a fixed CPU which has the same performance as a high-end pre-fix CPU may not exist for for several more years.

* (We're pretty much only talking about Meltdown and Intel. From what I've been reading, the fixes for the other bugs affecting non-Intel CPUs result in minor to negligible performance hits. Which also makes me think I haven't been giving AMD the respect they deserve. I'd assumed they hadn't been able to match Intel's performance because they simply weren't as good at designing processors. Now it appears a large part of the reason was because they refused to follow Intel in doing something that compromised security to generate more performance.)

Results varied. Some workflows experienced a very significant performance loss however, e.g. Firefox became twice as slow, WinRAR four times as slow.

Older AMD CPUs (read: Phenom 2 and earlier) do not have any kind of management processor. I don't know about the desktop versions of the earthmover cores (the FX-series), but a fair few of the mobile chips (the A-series) have it (https://hothardware.com/reviews/amd-beema-and-mullins-mainstream-and-lowpower-2014-apus-tested?page=2 and http://www.tomshardware.com/reviews/amd-tablet-processor,3813-2.html). Ryzen most definitely has this, and makes heavy use of it (http://techreport.com/review/32125/amd-epyc-7000-series-cpus-revealed).

The code for the Intel management processors is stored on the mainboard's flash chip. Intel's version is surprisingly modular and it's possible to remove at least some of the components (https://github.com/corna/me_cleaner). Note that said management processor has some rather strong self-preservation instincts and won't allow anything to write to its region of flash memory. Since it (not your x86 chip) is the true master of "your" computer, this means that you need to yank the power cable and program the flash chip directly using a Beaglebone or Raspberry Pi and a SOIC clip (https://libreboot.org/docs/install/rpi_setup.html). Annoying, but doable.

I do not know how AMD CPUs store the code for their management processor, but I'd guess that it's done in a similar manner to the Intel CPUs - in a region of the motherboard's flash memory. I don't know of any investigations into it yet, but one advantage you have there is that it's an ARM processor and as such there are a lot of very mature debugging and disassembly tools which can be used to investigate the code. Additionally, AMD uses the Trustonic codebase for their management processor (https://www.trustonic.com/news/company/amd-licences-trustonic-trusted-execution-environment/), which I've seen before in phones and was very modular with each "trustlet" (separate tasks dealing with things like kernel integrity monitoring, OAUTH tokens, or Widevine DRM) being a separate file on the filesystem - if this is the case on Ryzen, it might be possible to remove some of the more offensive components with minimal effort.

With this new technology announced less than two weeks ago?

Above 1 TB only geeks and IT companies care. Just like it was case for CCD above 10 Megapixel. At 1 TB disk is "solved" problem for general audience and things that matter now are performance and durability.

Not for HDDs, the "high performance" market has all but disappeared except for a few hybrid SHDDs for laptops. Even failure rate is for consumers "it can fail", they don't have enough in a RAID setup or whatever to care about the statistics, at most they have a SSD for performance and a HDD for bulk storage. And the enterprise typically has this in some kind of SAN or storage server to handle failures, unless it's so bad that the failure rate works out to a $/TB difference they don't care. So the HDD market has almost been reduced to a singular metric, $/TB.

Even the SSD market is seeing some form of saturation, for consumer workloads with low queue depth the differences from one SSD to the other is very modest, if you flip the graphs from MB/s to seconds/task it's like a really fast SSD boots Windows loaded up with software in 30 seconds and a "slow" SSD takes 35. Some games have ridiculously long load times regardless if you put them on HDD or SSD, whatever it's waiting for it's not the disk. It's a bit annoying that games that take 50GB+ can't split their assets up over two disks for fast/slow access but it's not going to change so whatever, if you play it a lot make room on your SSD and put the more rarely played games on HDD if you run out of space.

Yet the hardware decoding adoption for consumers for VP9 stands precisely at 0%

Intel has been shipping VP9 hardware decoding for years. By default Microsoft Edge enables VP9 when hardware decoding is present (though you can override that to enable VP9 in software). VP9 is a standard video format on Android and many Android phones have VP9 hardware decoding.

Hardware VP9 decoding is rare enough of a feature already.

It's not that rare. Intel's been shipping VP9 decode acceleration for two years now. Android has supported VP9 decoding since Android 4.4, which was released in late 2013. If you have an Android phone, you probably have VP9 hardware acceleration. Plenty of AV1 hardware will be released in late 2018.

But also don't underestimate today's mobile devices. I have an iPhone 7 and I can play VP9 video in software in VLC for iOS without issue. A future VLC update will add AV1 support.

AMD actually beats intel in AES both in performance-per-watt, in performance-per-dollar and with most CPUs also in raw performance. See for example http://www.anandtech.com/show/... and http://techreport.com/review/3.... Note also that the only Intel offering in these beating the Ryzen 1800x in performance-per-CPU is the i7-6950x, which is more than three times the cost and almost 50% increase in TDP. That said, most of the benchmarks don't have the Intel server offerings included, so perhaps you can prove me wrong?

Some games really do benefit from having faster memory. Arma 3 for example, really benefits from higher memory speed; in The Techreport's test, for example, moving from DDR4 2133 RAM to DDR4 3866 RAM increases the fps from 46 to 56. http://techreport.com/review/3... Of course, the DDR4 3000 kit they tested got 54 fps; that is probably about where diminishing returns kick in for most games. Fallout 4 is another game where having faster memory can make a 10-15% difference in certain areas.

I can't see an Apple only processor wining over Intel, either. At minimum, Intel's process advantage would have to be nullified and I can't see that happening until scaling comes to a full stop.

Well Intel has been very tight lipped about die sizes lately but the 14nm Broadwell-U has 1.3 billion transistors in 82 mm^2 and 1.9 billion transistors in 133 mm^2 so 15-16 million transistors/mm^2, same with the Xeon E5-2600 v4 it's 7.2 billion in 456 mm^2 so 16 million/mm^2 too and they haven't had a die shrink since. Apple's 16nm A10 that's in the latest iPhones have 3.3 billion transistors in 125 mm^2 die size so 26 million transistors/mm^2, it might not be an apples to apples comparison but seems to me like Apple is already ahead in density. It's probably easier on a low-power chip, but really I don't think Intel is ahead anymore.

Unfortunately, I can't google "ipad fire" as it collides with "kindle fire".

Here you go.

Couple of choice ones:

Fire that killed four dogs in rescue home started by an iPad charger iPad fire

So to show us "Apple battery fires", you link us to stories about "(chargers (supposedly) from Apple fire (or in one case more of a smolder)"

"iPad Fire" (read the post)

iPad charger fire - understand the post. There was no fire damage on he iPad, only a cracked screen from the idiot dropping it on the floor. The only other damage was a molten connector/cable (quote: " It's almost on FIRE!") How the hell was that an "iPad" fire? Heck, how was that a fire at all? Like I said, a smolder

Anyway, even if either actually were an iPad fire, both fires required a charger to be used - can you show any mention of a charger being onboard flight MS804? Yeah, thought so.

Unfortunately, I can't google "ipad fire" as it collides with "kindle fire".

Here you go.

Couple of choice ones:

Fire that killed four dogs in rescue home started by an iPad charger
iPad fire

So to show us "Apple battery fires", you link us to stories about "(chargers (supposedly) from Apple fire (or in one case more of a smolder)"

"iPad Fire" (read the post)

Unfortunately, I can't google "ipad fire" as it collides with "kindle fire".

Here you go.

Couple of choice ones:

Fire that killed four dogs in rescue home started by an iPad charger iPad fire

So to show us "Apple battery fires", you link us to stories about "(chargers (supposedly) from Apple fire (or in one case more of a smolder)"

Unfortunately, I can't google "ipad fire" as it collides with "kindle fire".

Here you go.

Couple of choice ones:

Fire that killed four dogs in rescue home started by an iPad charger
iPad fire

Wow, Apple censor team is active today.

Unfortunately, I can't google "ipad fire" as it collides with "kindle fire".

Here you go.

Couple of choice ones:

Fire that killed four dogs in rescue home started by an iPad charger
iPad fire

You mean, make their newest product available to people so they can sabotage their launch by "benchmarking" it with synthetic software which doesn't really support it and is specifically tuned to run well on Intel hardware and compiled with a compiler which specifically sabotages AMD CPU's? Why would they want to do that?

Intel does not have exclusivity on intelligence, though for a few years they have had exclusivity on marketing. So, AMD is leap-frogging over Intel. In the future, ARM systems will leapfrog over the two. Technology moves on. Time to buy AMD shares.

You mean, make their newest product available to people so they can sabotage their launch by "benchmarking" it with synthetic software which doesn't really support it and is specifically tuned to run well on Intel hardware and compiled with a compiler which specifically sabotages AMD CPU's? Why would they want to do that?

http://techreport.com/review/3... Conclusion: "If time is money for your work, and your work can take advantages of lots of threads, the i7-6950X is the fastest high-end desktop CPU we've ever tested, full stop. If you don't need all of its cores and threads, however, the Core i7-7700K arguably delivers the best gaming performance on the market for about a fifth of the price. Intel's Extreme Edition CPUs have never been good values, but the i7-6950X takes the definition of "halo product" to eye-watering new heights. If the return-on-investment calculations work out for you, though, the i7-6950X is an amazing chip."

CR is far more trustworthy than Apple.

Thus implying that Apple is untrusthworthy. So when was Apple busted for cheating, ATI style? Or is this just one of those Hatorade-based tautologies, like holding it wrong or Bendghazi?

Many phones, tablets, and PCs have hardware acceleration for VP9. AMD, Nvidia, and Intel support it. By default, Microsoft Edge turns on VP9 support when hardware acceleration is available (you can override it to turn VP9 on all the time).

The "no hardware acceleration" argument is tired. Things have moved on.

Hot hardware? Seriously, if it the posting is from them (and it is) always link in Anandtech and Techreport if available. http://www.anandtech.com/show/... http://techreport.com/review/3...

So, your single point of data is more important than tests that have actually been run?

http://techreport.com/review/2...

I would love to see the spinning disk that can handle 2.1 petabytes being written to it. But of course you have run the thousands of tests for years on end needed to identify the unreliability of SSDs vs HDD.

http://www.tomshardware.com/re...

SSD lives in an actual experiment are in this article:

http://techreport.com/review/2...

The drive that did the worst failed at the 728TB written mark. These were 250 GB drives, so I would expect 1 TB drives to be able to sustain approximately four times the write volume. The means we should expect failure at about the 3.5 Petabyte mark. Two video streams should pretty much never exceed 10GB/hour. 3.5 PB/10GB =350,000 hours. That's about 40 years.

  Yeah, I think SSDs are OK for DVRs.

That was a lot of work - easier to just link here.

Let's work it out. A few years ago, TechReport ran an SSD endurance experiment to figure out how much punishment current-gen SSDs could take before failing. Their test setup essentially involved writing random data at maximum speed for 18 months straight. The results indicated that the worst SSD in their bunch, a Intel's 335 Series, wrote about 700 TB before dying, and the best SSD, a Samsung 840 Pro SSD, went on to 2.4 PB.

Various estimates say you can put between 60-75 hours of HD content on a 500GB drive, so, assuming the largest possible size, that works out to about 8.3 GB/hour. Since you're writing two streams, that's 16.6 GB/hour, or 145 TB per year. For the worst drive in the bunch, that's about 4.8 years of service (right at the upper end of your HDD's service life); for the best drive, it's over 16 years.

Keep in mind that these tests were all run on 250GB drives. Smaller drives have less flash to work with, and have to write over the same flash cells more often. Therefore, if you bought a 1TB drive, you can expect the lifetime to be easily 4x better (more if you're using a more recent drive, such as the Samsung 850 Pro) - 64 years of DVR recording should be more than sufficient.

The largest recording I've ever seen off of cable TV is about 8GB/hr. I know OTA broadcasts can be slightly bigger, so lets say 10GB/hr. To record that 24/7 requires about 87 TB/year.

There was a long term test of SSDs done here:
http://techreport.com/review/2...

Many of the drives ended up getting close to 1 PB of writes, and the best even got over 2PB. Thats enough for you to run 2 tuners 24/7 for a decade. And note, their tests were with 250GB drives. As you increase SSD capacity, longevity increases almost linearly. If you were building a DVR, you'd probably want something like a 1TB drive.

As far as the original question of whether the SSD can outlive HDD in the most extreme application....probably at the most extreme, no. But for the vast majority of cases, including a DVR, most likely yes.

Samsung Evo 840 is rated for a 28-year life span at 10GB of data write per day. That's about 100TB written. According to some tests, the 840 starts experiencing sector relocations (bad NAND) around 100TB; somewhere about 9 times that, it suddenly fails without warning.

If you're constantly buffering HD video at 11GB/hr, that should give you 378 days to 100TB and maybe 9 years to sudden catastrophic failure.

Once you have a decent SSD, adding storage bandwidth doesn't dramatically improve load times for most games and applications. Intel's old X25-M G2, which is tied to a 3Gbps interface that has limited bandwidth even by SATA standards, loads games and application data just as fast as recent PCIe drives that achieve much higher performance in targeted synthetic tests.

There is, but it was patented and nobody sells them.

https://techreport.com/news/27...

http://www.extremetech.com/ext...

SSD endurance is fine. All good brands last a long time and even the supposedly cheap and short lifespan Samsung TLC drives last well beyond what most people will ever get out of them.

SSDs don't see to be any worse than HDDs, and for laptops that get moved around a lot are probably even more durable.

I'm aware of at least two endurance tests of SSDs that showed write endurance greatly exceeding expectations:

http://techreport.com/review/2...

This one was widely reported on Slashdot and other sites.

http://packet.company/blog/?ca...

This one was more recent and I haven't seen it show up here or elsewhere, but it's moderately more interesting because it's a newer Samsung 850 Pro 1 TB and it didn't fail until after 7 PB of writes.

Now, there's all kinds of problems with these tests as being not exactly definitive or especially scientific, but they do show that MLC SSDs are more durable than generally believed.

IMHO, a lot of the complaints about SSDs are mostly regurgitated info from early adopters and often early adopters who bought cheaper drives. I've put about a dozen 840s in random desktop PCs with no failures, an even older Samsung 120GB model in my old desktop never failed before I retired the drive after 4+ years.

Personally, I'd like to see someone step up and put 850 Pros in an external enclosure in a hardware RAID set and run the kind of normal production workloads a 12 or 24 drive shelf might see in a normal business -- neither too low utilization or too high utilization -- to get an idea what 3-5 year lifespans would be. I'd like to guess that failures might exceed normal HDD failure rates (although, what's normal? I'm staring at a 5 year old EQL PS4000E with 6 of 16 512GB disks replaced) -- but, the drives themselves are cheap enough that the high performance would be worth a slightly elevated replacement rate. Plus with sensible RAID policies like double parity and hot spares, the risk of catastrophic array failure wouldn't be any worse than a HDD array, especially when you consider rebuild times would be an order of magnitude faster.

My general suspicion is that other than for extreme write workloads, better MLC SSDs are probably good enough now and only inertia and SAN vendor profit margins on SLC SSDs are keeping us from seeing cheap, flash-based arrays now.

You missed the USB drive in the middle... 3 strikes... http://techreport.com/news/130...

And don't forget their special USB drives. http://techreport.com/news/130...

Bleh. Wasn't the first time enough?

Not for them. They did it again in a USB drive. http://techreport.com/news/130...

but the anxiety after only 3 months is palpable.

I assume they're under warranty for the next while.

My fear would be that there is an intelesque hard limit on writes that bricks the drive, and ALL of your drives in a RAID array will hit that limit simultaneously.

TechReport did something similar and finished killing all their SSDs. Granted, their article included only a 840 (Pro and "non Pro"), but it is still significantly more than what they were rated for.

I am glad more people are conducting these experiments regularly to make sure that the manufacturers do not relax their QC.

Question is, can they afford to buy cheap expendable SSD drives now to boost performance

After reading about the SSD write endurance test:

http://techreport.com/review/2...

I'd be curious just to see how long they would last in real world (neither extremely brutal nor extremely mild) SAN/RAID applications. Like a shelf of 24 in a RAID-6 config with a couple of hot spares.

What would the actual failure rate be? Would the relatively low cost of say a Samsung 850 Pro be worth a higher failure rate when you consider what the cost vs. performance of a shelf of 24 SSDs would be? Would $2k a year on replacement disks be worth a storage subsystem capable of that kind of performance?

Or would it just overwhelm the SSDs and make the failure rate unsustainable?

They have a good overall reputation but some of them hijack your data when they fail.
See SSD life endurance test https://techreport.com/review/....
Anyway SSD's fail completely different from Hard drives. Most just vanish, some corrupt massively and others go in a final one chance read-only mode (select Intel consumer models).
Tested backups are a necessity here.

The GPU actually has thousands of cores. For example, the GeForce 980 ti has 2816 cores. Compared with a CPU, those cores are much simpler, and they run at a lower clock speed, but for extremely parallelizable operations, a GPU will stomp a CPU easily. That's the reason you see GPUs becoming a major factor in supercomputers. More generalized supercomputers are hybrid systems.

Both AMD and nVidia have been doing this for years with their Windows drivers. Why? Because apps like 3DMark and games like CS, Quake are used to benchmark video cards by reviewers.

I trust the name Toshiba. But I can't help but think any company aiming at the budged SSD market will skimp on wear leveling in favor of other attributes. Yes I saw the 7% overprovisioning note, but that was the whole of the attention given to a rather complex topic.

I wouldn't mind the somewhat slower access noted but in recommending an SSD for general system use I would be wary if a drive couldn't handle large volumes of throughput over its lifetime. Modern applications, not even Windows 8, are careful with disk usage. For example Windows 8 is happy to use an average 2% of drive access time for record-keeping alone on my own PC (I had to disable services to make that stop)

The average user is still running 15 junkware applications that tend to be written with any consideration but the user in mind. How much more important is it for them to minimize disk wear for an SSD? Since they're not, the'll need drives that can hack it. I saw numerous benchmarks in the fine article, but not the kind about which I care the most.

So I looked it up myself. Ten minutes of Googling found some information for other OCZ drives, but not this one. The only other thing I can add is this table of return rates for previous editions.

We've been using Samsung drives in "non production" status servers, embedded servers, etc. and have had a terrible time of it. The first drives we bought a few years ago (840 Pro) were good, but we've seen Samsung SSDs run entirely through their write capacity (as reported by SMART) and then go dead when not even mounted! Turns out we aren't the only ones to get bit by buggy Samsung drives.

And the 10 year warranty that interest me - running a 32Gig SSD drive for a week I lost 48K of the drive.

It's or bytes written.

"Samsung guarantees the 2TB 850 Pro for 10 years or 300 terabytes written (TBW), and the 2TB 850 EVO for five years or 150 TBW.

840 Series 120GB/250GB/500GB 3 years
840 PRO Series 128GB/256GB/512GB 5 years (73 TBW for enterprise applications)"

http://www.samsung.com/global/...

We've been using Samsung drives in "non production" status servers, embedded servers, etc. and have had a terrible time of it. The first drives we bought a few years ago (840 Pro) were good, but we've seen Samsung SSDs run entirely through their write capacity (as reported by SMART) and then go dead when not even mounted! Turns out we aren't the only ones to get bit by buggy Samsung drives.

It also turns out that Samsung drives are even blacklisted in the Linux Kernel

I welcome Samsung's excellent cost/size value proposition! I just wish their drives were solid enough for our actual use.

Perhaps you should update your lovely Wikipedia page, because it is outdated.

SATA 3.1 standard note at techreport
Webopedia info on SATA 3.x
Wikipedia's own entry on SATA 3.1
TechPowerUp article about SATA 3.1

Here's a press release from sata-io about it: in PDF format

Not only does TRIM via NCQ exist, it is in the recent specifications. You see, the thing about computer technology is that it keeps being improved. Outdated information doesn't stop that. It just becomes outdated.

The drive's media wear indicator ran out shortly after 700TB, signaling that the NAND's write tolerance had been exceeded. Intel doesn't have confidence in the drive at that point, so the 335 Series is designed to shift into read-only mode and then to brick itself when the power is cycled. Despite suffering just one reallocated sector, our sample dutifully followed the script. Data was accessible until a reboot prompted the drive to swallow its virtual cyanide pill.

One has to give it to AMD. Despite their stock and sales taking a battering, they have consistently refused to let go of cutting edge innovation. If anything, their CPU team should learn something from their GPU team.

On the topic of HBM, the most exciting thing is the power saving. This would potentially shave off 10-15W from the DRAM chip and possibly more from the overall implementation itself - simply because this is a far simpler and more efficient way for the GPU to address memory.

To quote:
"Macri did say that GDDR5 consumes roughly one watt per 10 GB/s of bandwidth. That would work out to about 32W on a Radeon R9 290X. If HBM delivers on AMD's claims of more than 35 GB/s per watt, then Fiji's 512 GB/s subsystem ought to consume under 15W at peak. A rough savings of 15-17W in memory power is a fine thing, I suppose, but it's still only about five percent of a high-end graphics cards's total power budget. Then again, the power-efficiency numbers Macri provided only include the power used by the DRAMs themselves. The power savings on the GPU from the simpler PHYs and such may be considerable."

http://techreport.com/review/2...

For high end desktop GPUs, this may not be much, but this provides exciting possibilities for gaming laptop GPUs, small formfactor / console formfactor gaming machines (Steam Machine.. sigh), etc. This kind of power savings combined with increased bandwidth cna be a potential game changer. You can finally have a lightweight thin gaming laptop that can still do 1080p resolution at high detail levels for modern games.

I know Razer etc already have some options, but a power efficient laptop GPU from the AMD stable will be a very compelling option for laptop designers. And really, AMD needed something like Fiji - they really have to dig themselves out of their hole.

I've actually fully refreshed (ie, done a dd back to the disk after saving it off) twice. The disk is still slower than molasses compared to its initial 20% full speed. Now, that doesn't mean it's slow by any means, but far slower than the 400MB/s peaks. In fact, this is true across the entire spectrum of SSDs, from what I can tell from various tests, including long term tests.

Now I have not run anywhere near a PB through my drive, but I'm probably near a few tens of TB, which is why I'm replacing the single 840 with a pair of 850s (EVOs all) And yes, I was aware the "fix" wasn't a fix, just a means of resetting the drive cells, which is the same thing I accomplished by wiping and restoring my drive prior to that fix coming out.

They don't die randomly. They die on an exact schedule, and about three times sooner than more durable brands.