*** Official Clevo P65xSA/SE/SG / Sager NP8650/51/52 Owner´s Lounge ***

Thanks for the info about the M.2 slots. I actually had a hunch that the SATA slot might be disabled if you used a PCIe card in the other slot, but I edited it out of the post because I was really unsure. And yeah, I didn't think you could use storage in the WLAN A keyed slot, but surely you can find stuff to put in it if you're uber paranoid and hate wifi.

And yes, I know anandtech has done some benchmarks on M.2 SSDs. The full sized cards fair quite a bit better than the 2242 cards that I have experienced (which are rather horrendous), but still not as fast as full 2.5" drives. I can't find any benchmarks on the MX550 M.2. If they were able to get a decent amount of DRAM on it for nand mapping and caching, then I'd feel more confident recommending them. But until I see confirmation of that, I'd err on the side of caution, and get the time tested 840 evo instead. It's cheaper, faster, and worse case scenario you can always go buy an M.2 drive later, you don't really lose much by getting the 840 now (unless you need the 1tb hdd backup).

 

It's all anecdotal really - numbers only tell you so much...

I went from a Mushkin Chronos 256GB SSD on my desktop (SATA-3, advertised as ~550MB/sec read) to a Plextor M2 256GB SSD (connected via PCIe, advertised as ~700MB/sec read). The Plextor has 512MB of cache, I don't know what the Mushkin has.

The difference is noticeable, especially when it comes to dealing with small files. As such, going for the Plextor M2 SSD was an obvious choice for my P650SG, as far as I'm concerned!

 

The big thing coming out of many reviews is that write speed drops when one chooses a 128-256GB unit versus a 512GB unit in M.2 A major reason is because the number of NAND ICs are removed and, correspondingly, DRAM on-IC with them. Depending on how old/new designs are compared, this may or may not impact the older 2.5" versions versus newer M.2 units, especially when DRAM was liberally used (on older, more costly devices), and newer devices are dropping in price (as NAND gets cheaper) so less DRAM is used.

DRAM is used for mainly keeping up burst write speed because NAND is slow as crap when it comes to writing. If you're largely reading, NAND can be quite fast, let alone one gets the huge boost in read random access. That's the whole purpose for NAND devices.

But yes, if you're heavily writing to NAND devices, then reduced DRAM is going to constrain performance. Of course, it's also going to kill your device more quickly too, especially if it doesn't have enough to coalesce the writes in the DRAM buffer.

I'd like to see the M550 M.2 v. mSATA v. 2.5" benchmarks as well. We're at the point they are putting 512GB in 5mm high (not even 7mm) and not using the full area of a 2.5" device. Also remember that 2280 M.2 device (176mm^2) has almost 20% more area than 30x50mm (150mm^2) mSATA device too.

It's really about what your usage is going to be, and what devices offer what performance for that usage. If you have a small M.2-SATA device just for largely booting, reading, etc..., then DRAM buffer because less of a consideration. If you're going to put your swap file, TEMP/TMP and other things that are regularly (almost constantly) written to, then no, DRAM buffer matters heavily.

Er, um, you guys would go for the TLC (840 Evo) over MLC (MX100)? Sorry, I just cannot recommend that given the underlying physics of the order of magnitude reduced lifespan of TLC v. MLC. Devices can play games with buffering for performance, and even coalescing for reduced writes, but even rotation cannot starve off the inevitable rate of errors in TLC over MLC.

To use your phrase ... "to err on the side of caution", stick with MLC over TLC when you have the option for almost same price (or even savings). NAND devices do not last as long in the real world of regular usage v. short-term, artificial tests using block writes, especially not with NT's layout and a FAT filesystem design are committing much smaller portions than the typical 32-512KiB elemental clusters of NAND cells. This has nothing to do with TRIM either.

This is one area where I just have to disagree. Knowing the realities of NAND at the layout-level, and how it has more errors logarithmically, I cannot recommend any TLC solution over MLC. But that's just me, for myself. Might be less of an issue if you're only getting a device just for largely reading, or even just Intel SRT (which is very infrequently written to). And in the case of Intel SRT, it's still backed by another storage device (it's just a NAND-based read cache).

M.2-PCIe is keyed in A, B, E and M, and logic location vary between keys. I need to read the specs again, but I bet B (PCIe x2) and M (PCIe x4) are in the same location, at least for the two, common 2 PCIe channels.

M.2-SATA is keyed in B and M at the same location. Hence why we're going to see M.2 storage devices (non-USB) that are B+M notched (work in either B or M slots).

Having read the manual, I just have to disagree, and here's why. The M.2 B-key slot defines PCIe x2. What the manual is likely saying with the "or" is that the P65x M.2 B-keyed slot does not support PCIe, only SATA.

Understand the manual is absolutely ambiguous with the use of "or." There are multiple ways to read it, and if there is anything I've learned dealing with non-English engineering translated to English documentation, it's misuse of conjuctions (as well as propositions) that leave ambiguous (or even flat-out incorrect logic). I don't see how using the PCIe x4 channels would "disable" a SATA channel.

The only way it would be is if an entire set of SATA channels were disabled because it was connected to a PCIe x1 channel that the PCIe x4 connection used. But that wouldn't be the case because all the mobile 80 series -- HM86, HM87 and QM87 -- all have six (6) SATA ports, four (4) SATA-6Gbps (the eSATA is usually 3Gbps to support longer cable lengths).

Until we know for sure, we should not be claiming either way. We should see what happens when a M.2-PCIe device is inserted into the M.2 M-keyed slot and a M.2-SATA device inserted into the M.2 B-keyed slot. Only then will we know. In fact, none of the reseller order configurators prevent a M.2-SATA from going in the B-keyed slot when there is a M.2-PCIe in the M-keyed slot.

Actually, you can, but it's not going to be a M.2-SATA B+M notched storage device. It would have to be a M.2-PCIe A[+M] notched storage device, or, an M.2-USB A[+B][+E] notched storage device.

 

Only read half your post, but DRAM absolutely has a rather large impact on read speeds, especially as the drives grow larger and it can't hold enough of the lookup table in the controller's SRAM. Sequential reads are mediocre, random reads drop off to the point where it is mildly pathetic (we're talking as low as 25mbyte/s, compared to the 840 evo's 100+mb/s). Writes, as you already pointed out, are even worse.

TLC NAND is fine. 100 gigabytes of writes per day on a 500GB 840 Evo will last you 15 years. If that isn't long enough for you, then go use tapes.

 

As always ... assume as you will.

If I make 3 responses, some people complain and/or point out that I make 3 posts (sometimes even using a separate post to do such).
If I make the extra effort to consolidate 3 posts into 1 response, even ordering the quotes to fit each response (cut'n paste), some people will complain about it's length.
And few notice I trim to keep size down either, while still maintaining and respecting context (which cannot be said for several others).

So ... declare all you want, assume all you want ... I actually do try to reduce overall response size.

Depends on the basic allocation, layout, etc... For the most part DRAM is being increasingly bundled in the NAND IC itself, with no, additoinal, external DRAM being offered.

E.g., if you compare an older, more costly 2.5" NAND device that had DRAM off-IC, to a newer, M.2 NAND device -- one that is cheaper and uses same in both 2.5" and M.2 versions -- with DRAM on-IC with the NAND, then yes, you're going to get apples-to-oranges in performance. No argument.

It's not just simply M.2 v. 2.5", but if the 2.5" offers additional, off-IC DRAM for additional buffer.

I'll quote you ...
"random reads ... as low as 25mbytes/s"

I haven't seen a NAND device that comes remotely close to breaking 100MiBps on random reads. They are typically around that 25MiBps, maybe getting closer to 50MiBps these days. That's why we use NVRAM (e.g., battery backed RAM) devices in front of them in storage solutions. But compared to platter, which is usually only 1-2MiBps, maybe 5MiBps best, at random reads, that's over an order of magnitude faster.

Maybe it is that the 840 Evo devices come with so much DRAM, to coalesce writes to the point they prevent their TLC NAND cells from being written to much at all (although I'd have to ask what happens if I lose power to the unit?). In that case, if you're reading solely from DRAM nearly all the time, sure, you can get in excess of 100MiBps. I'll concede that.

But I'd like to see it in the real-world, under the tests I typically deal with when it comes to NAND in a 3-tier storage solution. I just never seen much beyond 25MiBps any way, so I end up putting NVRAM in front of it for a reason. Understand what I'm saying here. Not just dumb, perfect, 4KiB block reads ... but actually trolling through a file system, like a FAT-based file system under NT/Windows, with random sized reads, not just random reads.

That's what always makes the difference, and why I don't see these numbers you guys are talking about.

Sorry, disagree. Too many device failures over the last few years in too many production environments, even in just portable applications. There's no way anyone can test a new device like it's being used 2 years straight like a typical OS.

TLC NAND allocations are only good for a median of 100 writes before the errors cause the block to become infeasible compared to 1,000 for MLC. You can only rotate so much. Basic physics at work.

Heck, there's a reason we still use SLC, heavily so, in several applications.

 

http://images.anandtech.com/graphs/graph7173/56406.png

From anandtech's review of the 840 evo. Even Samsung's MLC, and SanDisk's offerings are hitting or passing the 100MB/s random reads mark. And your 'real world usage' example doesn't have much merit -- if the performance of the 840 evo is 4x in 'dumb, perfect, 4K block reads', do you think it'll be less than 4x difference in real world usage? Random sized reads so they aren't perfectly aligned is just going to thrash the already stressed lookup table.

As for your point on TLC, fair enough. I've yet to have any problems with it, but my usages aren't exactly write intensive.

 

Yep, as I assumed ... fixed, exacting, 4KiB random read. I live in the real world. That never happens.

No doubt. I can run some Linux dd and probably get even better numbers!

EDIT: In fact ... (I found the article that graphic is from) ...

'Our first test writes 4KB in a completely random pattern over an 8GB space of the drive to simulate the sort of random access that you'd see on an OS drive (even this is more stressful than a normal desktop user would see). I perform three concurrent IOs and run the test for 3 minutes.'

Sorry, that made me laugh. And this is just for the writes.

Spoiler

If you've ever tuned Linux, you know that the defaults are 50s/15% to "wakeup" pdflush, and 300s/60% to "commit" a flush.

I.e., 5 minutes and 9.6giB (60% of 16GiB) to force a flush when there is ongoing I/O fill/wait (which there is, you're writing continuously). In other words, it doesn't really matter what disk scheduler is used, how many current writes you do, the patterns of the data, etc... the Linux kernel won't even flush that until it's all "seemingly written" (but not), if you have 16GiB RAM (unless you tune those defaults way down).

Oh, but wait ... I can use O_DIRECT, and bypass! Er, yes and no. O_DIRECT only works if everything is perfectly aligned. So of course for an artificial test using 4KiB blocks, it will work, most definitely. But not real-world. Which brings me back to my earlier points. You really can't test this ... unless you use real apps with real user loads, with a real file system atop of the block storage, etc...

NT's I/O scheduler isn't nearly as aggressive (or tunable for that matter), and I use Linux as an example because I haven't tuned NT in several years now, so I might be out-of-date on specifics. In fact, most people don't bother with tuning NT Server I/O these days because it's almost always running atop of VMware (older RHEL5+VMware HyperVisor), although bare-metal Workstations running Windows must still be tuned.

Yes! Oh Goodness yes!

It's why even SD cards -- which are much, much slower (so the bottleneck isn't in the test software or elsewhere) -- don't even get half their speed when truly random sizes are involved, let alone a file system is involved! Anyone who has done fixed sized dd v. random-sized reads/writes v. through the filesystem layer, deals with this.

Storage is one of those things where any layers of complexity can be a major performance hit.

Spoiler

Linux 2.6 sported a completely new subsystem called DeviceMapper (DM) specifically to remove any overhead to storage organization. It's what LVM2, MD, Multipath and countless other userspace actually modify. They merely change DeviceMapper tables/mappings which work at full, direct, block speeds.

Unfortunately file systems are still atop of them.

And goodness knows Ext file system performance has massively jumped through many performance tunings, many gained from Red Hat hiring Sandeen and others from SGI who designed XFS back in the '90s for massively parallel video servers. But it's still a major basket of overhead in the end.

btrfs is yet another story, merging volume and file system management into one.

On the Windows end, Microsoft has long been trying to develop the CairoFS, cached in-memory, SQL-based store for NTFS since '94 (for NT4 released in '97). The subsequent WinFS was finally abandonded 10 years later, and there are still major bottlenecks that show up in "real world" performance -- and I mean major ones compared to other platforms.

Real-world is filled with all sorts of sizes. Plus you have the 2-3 reads for the table, directory block, etc... in FAT. If it's inode, there can actually be even more reads (although far less writes to the same blocks than FAT).

Again, it all depends how the system is used.

Before Windows 7, when Microsoft brought the overlay approach forward from Embedded NT/XP, it was pretty bad on anything but Embedded NT/XP, for FAT/NTFS. Not even SLC lasted 90 days median for a Windows desktop OS. I did those tests back in '06-08 with full (non-embedded) Windows XP, the new Windows Vista and, for the heck of it, Windows 98, using a storage solution that claimed all sorts of things with their "advanced rotation."

As always ... I stress people read many of the guides out there and disable/reconfigure all sorts of stuff. Don't put your swap file on NAND (disable it if it's all you got), point TMP/TEMP environments elsewhere (if possible, RAMDISK if it's all you got), etc...

With modern, typical Windows desktop usage, 3 years should be the overwhelming reality for MLC, possibly 5+, depending on RAM sizing (how much Windows thrashes). I'm not saying TLC will last 1/10th as long, there is likely mitigation going on. But it still factors into the ... again, to requote ... " to err on the side of caution."

In 2 more years we'll see how TLC is fairing. But understand vendors are still shipping MLC for a reason ... including Samsung with the 850 Evo Pro.

I will totally support anyone recommending the Samsung 850 Evo Pro. But you're going to pay for it. Hence why one has to look at MLC v. MLC, TLC v. TLC, and understand why something costs something. E.g., not just MX100 being cheaper than M550, but also M550 often being cheaper than 850 Evo Pro too -- all MLC designs.

 

@ [email protected]

Simple lesson... less is more. Regurgitating Wikipedia will just make peoples eyes glaze over.

And TLC does just fine as does 256GB M.2 drives.

The whole "sky is falling" over SSD writes is over and debunked ten times over. Most users won't be able to see a degradation or failure due to TLC NAND for 8-10 years with their usage patterns. Heavy write users (like frequent hibernation 3-4 times daily, video processing, etc) should opt for MLC or SLC.

 

Noted. I'm cutting and using Spoiler and other tags. But at some point, if I don't post the details, they are lost. Even my 2 statement posts aren't read.

Spoiler

Ergo the "context" aspect earlier. People miss it because they just skip my post altogether, not because of my length, and definitely not because it prompted 3-4 other posts that end up being even bigger (especially if I'm fully quoted).

Or Google, yes, yes, I've heard it before on forums, user lists, etc..., "Bryan just Googles."

Spoiler

Until someone else who knows me, in-person, tells them off-line ... "dude, he's an engineer that has designed some of this stuff over the years." In some cases, I've stated many people I know personally and have worked with in-person. In other cases, like M.2 details, I hit up the SNIA, who's specificatoins and standards I regularly read. These are coming from my mind, and I only Google if I'm trying to find a link to the document I've previously read.

So ... do I take your statements as ...
- An "assumption," because you don't know me in real life?
- An "insult" because you accuse me of getting things out of Wikipedia or via Google searches?
- Other?

I'll still never forget the hallmark moment years ago when one person who claimed I must have gotten my answer from Google because it was the first hit that led to a TLDP HOWTO until someone pointed out the obvious ... I was the co-maintainer of the official TLDP HOWTO. It's one of those things other people link to when they go, "Who is this guy that thinks he knows everything?"

Seriously, I don't post to do anything but share my viewpoint. Everyone is free to differ. I don't post to be popular. I only post for the 1-3 people that it might help. Some of my biggest critics over the years have become some of my biggest advocates. Because they never understood why I post the detail I do ... until they've been in a situation, and it was all there, in one post, including answering 2-3 follow-up questions.

If I posted to be popular, I wouldn't get the questions off-thread and work off-line I do, when I'm in the minority on something, usually based on first-hand experience. I fully admit where I don't have experience. I also admit when I'm proven incorrect too. I'm just an engineer, not a marketing guy. If I'm wrong, then others -- often the majority -- will be correct. If not, then I earn a little trust. Over months, and eventually years, that trust grows. That's how it's earned, and I put in the time in many technical groups.

Have they been left on and used for 15 years, 70 hours/week, which is typical? I cannot stress this enough, real world is very, very different than artificial tests that claim 15 years of lifespan.

NOTE: I never said anything about M.2 drives being not "fine," someone else did. They, not me, claimed lower performance with M.2 devices v. 2.5".

And that's another issue I often deal with, what others say being attributed to me. I.e., if it's see in an argumentative light by an argumentative person, it must be Bryan that said it.

HINT: I'm never trying to be argumentative. I'm stating my view, interpretation, even noting where I "disagree" or "differ" and letting it stand on its own, for others to decide. And I duly note I am often in the minority on many things. Doesn't mean I will conform to majority, but it doesn't mean I'm saying the world is wrong either.

Spoiler

I have literally been in those meetings over NAND, NOR, etc... over the years with various, global suppliers for embedded, advance storage and even workstation systems and solutions. Those tests are not only about as far from "real world" as you can get, but using a system for years results in far, far different patterns.

I'm still reaming from a decision in '07 with Deutsche Bank in this very case of embedded design. I made various concerns known about everything from the NOR selection to firmware update, and ... well ... I made the risks known. 6 months later ... I hate being right, because it means I failed to make the risks known. That's all. Risk always exists, it just about degree and impact.

As I used to say ... if the Shuttle Transport System (STS) could only fly when there was no risk, it would never, ever fly, not even once. All risks were mitigated in the original design, from the original, External Tank to the original, Solid Rocket Boosters (SRBs). Most people think otherwise, but they don't look at the details, like Material Reviews (MRs) of part changes at NASA, like O-Ring sealant in '85 and CFC-less insulator in '97. You have to have been in them, and at their massive rate (hundreds to thousands per year), to understand what I'm talking about.

I use that as an example, because everyone I've met (especially non-engineers) assume what happened with the Shuttle from popular media.

Risk is about degree and impact. I don't like TLC for my systems unless they are read-only caches.

Who said "sky is falling? I'm an engineer. I live in risk. I won't save $20 to take on something I know is inherently designed to have greatly reduced lifespan. But that's just me. Your applications, uses and other consideratons may differ.

E.g., for the 3 of you recommending the 840 Evo, I'm the 1 guy recommending the 850 Evo Pro, or the M550** if money is a concern. Why is that so ... "problematic"?

**NOTE: I don't nromally recommend the MX100, but the user was already considering it in the one case, and it's MLC, not TLC.

Why can't we just agree to differ? Why did you restate my post is as if the "sky is falling"?

Being in the minority doesn't mean I don't belong or am otherwise saying the "sky is falling." So what does that say about me? Or you saying that I'm saying the "sky is falling"? I stated my opinion, based on experiences. That's it. If you call that the "sky is falling," what can I do about that? Context. I said I "disagree." Accept that we disagree. Move on. Don't try to make my statements into what they aren't.

 

Are you guys saying I wont be seeing these speeds on my Samsung M.2 XP941? If not, that combined with the TN screen is a deal breaker for me.

Max Sequential Read Up to 1170 MBps
Max Sequential Write Up to 930 MBps
4KB Random Read Up to 122,000 IOPS
4KB Random Write Up to 72,000 IOPS

SAMSUNG XP941 Series MZHPU512HCGL-00000 M.2 512GB PCIe Gen2 5Gb/s up to 4 lanes MLC Internal Solid State Drive (SSD) - Newegg.com

 

hm... what was the topic of that thread again?
man i really like the way you write, all this posing stuff of numbers and personal preferences...
you buy every 2 to 4 years a new laptop and you are really discussing if the lifetime of a ssd is more or less 5 years?
how many times in 5 years do you move your 100% full ssd data from that ssd and back on? do you really think it will be more then 100 times??
rofl...
(little edit )

 

Define "seeing these speeds"?

First off, no one does sequential reads/writes, period. I honestly cannot think of a single application, sans cloning raw storage, block-by-block.
Second off, your storage devices ability to queue and service I/O operations per second is really a matter of device(s) and system usage.
Third ... what are you comparing to?

We can all compare spec sheets or even have fun running Direct I/O benchmarks on perfectly aligned boundaries ... but that really doesn't tell us jack as far as how something will perform for us. There are so many layers and levels of abstraction, from OS block (buffered/recordered writes and cached reads) to file system to application layers and those aspects.

And even getting past quantitiative ... there is the "qualitative" that is more feel and words. Someone here has already mentioned how a M.2-PCIe "feels" different -- e.g., better, more responsive, etc... That's really the most personal test for oneself and hard to argue with once has an experience and resulting agreeance (or disagreeance).

Spoiler

Maybe in the future something like Valgrind ( Wikipedia**) will be extended to, and supported by, secondary storage device vendors, so we can profile exact latency and throughput for running applications through the storage subsystem, and not just for CPU registers, various caches and memory.

**NOTE: I'm only linking to Wikipedia for those interested in an introduction, not that I learned about it from Wikipedia. If you do any system tuning, especially profiling programs on specific hardware, you will use Valgrind. It's extremely useful for more than just coders.

I don't understand. Are you buying the Clevo just to get a M.2 M-Key slot (PCIe x4)? I really don't want to directly influence your purchase/no purchase decision based on this limited information. What are your requirements?

Furthermore, one can't even really speculate on the future v. today's legacy AHCI-compatible M.2-PCIe products like the XP941 and othres.

Spoiler

E.g., no one knows how much of an impact NVMe will make over AHCI ( Wikipedia**). Same goes for any change to a dedicated, EEPROM-optimized "memory interface" (possibly closer to the local CPU-memory system interconnect?), and away from a PCIe I/O interconnect (which is not on the local system interconnect), let alone one not bridged to PCIe (e.g., SATA).

**NOTE: Again, only linking to Wikipedia as an example of what NVMe (optimized for randomly addressable EEPROM interfaces, like NAND) is allegedly designed to offer over AHCI (optimized for secondary storage with greater latency, like spindles).

That said ... for me, and me only ...

I'm personally saving some money and getting a M.2-SATA m550 for now, waiting for the M.2-PCIe options to increase and come down in price. I'm also interested in seeing what happens with NVMe.

E.g., Hynix is already sampling their 1TB, M.2-PCIe device with PCIe 3.0 x4 (60% faster than PCIe 2.0 x4). This includes Hynix claiming their new M.2-PCIe storage device will be offered in a NVMe version, not just AHCI.

Other people are going to opt for a 2.5" device, which must be SATA, as some models which may very well have much larger DRAM buffers than are possible to fit on a M2 card. That is also a very, very valid consideration, and will vary between models.

 

Eh, all 1080P MSI are m.2 + ONE 7mm HDD + TN panels, ALL of them from the 3000E GT72 to the 1600E gs60. .

M2 is not mature enough right now, they need just like CPUs to manufacture the M2 controllers on your SSDs with like 14nm so that they generate less heat and are even more powerful. that will probably be the case next year , be patient.



And i agree with HTwingnut : TLC FTW, a high capacity TLC will never give up on you before at least ~6-7 years of daily usage. AND they cost way less. MLC are more powerful and durable that's why they keep getting sold, but in 6 years I'll probably have a new SSD .

 

The problem is that we're talking a lot about component selection. This now brings in all the storage and the capabilities, concerns, excitement, interests, issues and other things for 2.5" SATA, M.2-SATA and M.2-PCIe options in the P65x.

My statements are what they are. Assume I'm the village idiot if it helps.

My personal laptop, yes.

Spoiler

Currently I scope and install thousands of NAND devices per year in various solutions, from other notebooks and workstations to multi-tier storage subsytsems. Most of my experience today is related to trading systems, high performance computing and other environments, storage-wise, all in the peta-bytes where dozens of tera-bytes of NAND caches are required.

(cutting everything else, irrelevant backstory)
NOTE: The problem with "credentials" is that someone always has more, which is why I don't like "explaining my background."

I'm not saying that to say I "know more." I'm saying that, now, in response to you, to point out my "experience" goes beyond buying a personal notebook every 2-4 years, and adding or changing it's local storage every 6-18 months.

Understand I do have about two-hundred (200) 2.5" drives and NAND devices, plus a good dozen (12) mSATA NAND devices in this very room at home I'm typing from. That's because I have about twenty (20) servers, from low-power AMD Jaguars to eight (8) core Opteron 4000 series, in my home lab. Virtually all I do with storage is work-related, even when prototyping a solution at home. In fact, I would figure I spend around $10K/year of presonal money on work-related prototyping, including $3-4K/year on storage.

Just a few years ago, mainly pre-Windows 7, it was really a median more like 3 months on SLC. Now, on typical MLC with buffering writes and various rotation, the roughly 1,000 write lifespan of the typical NAND block can result in a good median of 3 years lifespan with the majority of devices.

How many times in a hour do you think your table, dentries and other things get updated in a FAT file system without you even saving a file to disk? Ever profile it?

I have.

Spoiler

It sucks. And it is absolutely horrendous without the overlay, ported forward from Embedded NT/XP, into Windows 7+. There are other issues with how FAT file systems are laid out (although NTFS' extents helps somewhat), how Microsoft (really it starts with Gary Kindall of CP/M) stuffs everything into a single C: drive, etc...

And that's before even looking at putting a swap file on a NAND device, let alone the default TEMP/TMP environment under \WINDOWS, etc...

E.g., it's not uncommon for Windows to modify hundreds or even thousands of blocks when only a few KiBs of me ta-data have actually been changed. Understand that's 0 data actually being written, just a running system doing various things.

 

I assume you mean the M.2-PCIe options, correct?

Otherwise NAND is NAND, SATA is SATA, and it's just about design and, as discussed, any differences in amount of DRAM for write buffering (which can benefit read caching too).

As I mentioned/asked earlier ...

Is the "heat problem" in M.2 the actual 20nm NAND devices?
After all, we're not seeing a lot of heat issues with M.2-SATA devices.

So ... is it possible that the "heat problem" is the on-board, added PCIe bridge in M.2-PCIe?

Spoiler

I.e., you don't need a PCIe bridge if your device is just doing SATA. SATA, like Ethernet, is a pretty dang simple, 32-bit ASIC + serial PHY chips.

E.g., Mellanox pretty much makes "THE" ASIC and reference chipset for anything 40Gbps (~5GBps -- yes, 5 gigabytes, ignoring encoding and overhead) these days -- whether Ethernet, FibreChannel or Infiniband (they use the exact same ASIC for all 3).

But a PCIe bridge ... that's another story. And a commodity PCIe bridge at certain, sustained throughputs could generate a lot of heat, let alone is probably still on a 40nm TSMC process. So that would be my very first question.

A lot of people I know have lost data in 18 months on MLC over in just the last 2 years. MLC NAND, let alone TLC NAND, has only reached the point that it is considered "affordable" by PC OEMs in the last year. Hence why everyone is considering it for new PC system purchases in 2014+.

Spoiler

Also, most people have a tendency to use most of the storage they purchase, so I've never understood the "high capacity" claim. Unless you mean, of course, the newer, higher density NAND ICs leading to less rotation (and less DRAM buffer) in the smaller capacities because they are using fewer chips? That really doesn't change from prior generations either (lower capacities v. higher capacities).

Now if you're a gamer, and most of your data is read only binaries and data files that are not constantly generated/re-written, then you might be okay. It all depends on usage, if you disable many things and follow guides to reduce what Windows writes, etc... In fact, if you've been running Windows 3 years completely on a MLC NAND device for everything you do, then you know your usage.

NOTE: Android NAND Usage/Experience != PC Windows NAND Usage/Experience. Not only is Android running a Linux kernel, but it's a very embedded, read-only Linux image approach with minimal modifications to the file systems for the system portions. I've had to explain that to people who said to me, "my tablet lasted 2 years before I had any issues with its internal SSD. Same thing with BSD UNIX-based Apple iOS."

I cannot recommend TLC at this time, until I see the majority of people also reporting 3 years without any loss. Which is why I won't give anyone any estimates on how long something will last. And the last 12-15 years -- even just the last 3 -- are why I don't trust vendors who do either. That's my view, and my view only I speak of.

Hopefully I will be proven wrong over the next 3 years of daily usage with TLC devices.

 

A lot of people i know who have TLC doesn't have lost data in the last 18 months. I don't know what the f#ck your friends are doing with their SSDs but i sure won't do the same .

IF the added PCI bridge is there, it's for a reason that maybe you ignore ?

SSD in 2014 are way better that SSD in 2011 even if NAND is NAND and SATA is SATA, thanks to wear-leveling algorithms, a better nm for transistors, better controllers (with better firmwares) I assume the same will happen with M2






Normally the 25 -11 Pcspecialist told me they'll get their SG from clevos, but again, another person told me december 3rd and another december 9th (on the live chat, and another 25 -11 again on the same live chat module) i guess we'll see tuesday ! if on tuesday/wednesday they post 'in building' you'll probably hear my in the US joy screaming from France. Because they stated that november 11th they'll get the SE version but they got them november 1st.

 

I just found mySN UK have 2.5% cash back programme if you are buying through the quidco cash back website. I thought it's worth sharing since some members buy their computer from them.

http://www.quidco.com/mysn-co-uk/

 

Great news if true wickette
Personally I'm waiting for the SG version too.
3K/4K screen ( preferably the SHARP Igzo display), GTX 980M and at least a 500GB SSD
( probably the Crucial MX100, it's cheap and fast enough for my uses)
Would have loved to see a nice IPS 1080p offered but that's not happening here in the US.
Still waiting to see what CPU options we have too...if any.

 

Wait ... wait ... wait.

You mean the Samsung 840 Evo has been around 18 months?
Or ... do you mean ... the 840 ... pre-Evo?!
I.e., Be careful ... if you go there on the 840 ... pre-Evo!

What? I have no idea what you just said/asked.

Spoiler

If you're connecting to PCIe, you have to have a bridge. That bridge could be in the NAND controller, which requires a redesign of the NAND controller, but it must exist to talk to the PCIe. And then it's going to have all those lines, logic, etc... Understand a PCIe x4 controller isn't something you just slap on like a SATA end-device controller.

I.e., with NAND controllers with integrated PCIe 3.0 bridges (and even NVMe firmware) are now being announced these last few months. I suspect current designs have their PCIe bridge external to the NAND Controller, even if possibly in the same package. Yes, different feature sizes (e.g., 40nm or even 65nm for PCIe bridge, 20nm for NAND).

In fact, what is likely is that they are re-using existing PCIe 2.0 bridges (x4) from various Mini-PCIe designs (x1, so 4 blocks in the layout) in the initial M.2-PCIe designs. Hence why all new, native designs are PCIe 3.0 and sporting NVMe.

The "heat" issues are heavily M.2-PCIe and there are not remotely the same issues with M.2-SATA.

Wait ... wait ... wait ...

Just what does that have to do with 2.5" v. M.2-SATA? Your point was, and I'll quote the full context ...

That was the context where I used the phrase "NAND is NAND and SATA is SATA."

So ... back to comparing 2.5" SATA v. M.2-SATA ... what is the difference in "maturity"? SATA is 7-data pins (including all grounds on the data lines). That's it. Nothing special. It runs to the same controllers and same chips in the same designs whether 2.5" or M.2-SATA right down to the same pin-outs. Plus they both use AHCI.

So ... other than ...

How is M.2-SATA "immature" compared to 2.5" SATA? That's what I want to know, and why I asked for clarification when you just said "M.2." Understand it's these types of "broad statements" that really make me concerned what is being said. One can't just "blanket" all of M.2 with such a statement.

M.2-SATA today is very mature, the exact same maturity, as 2.5" SATA or even mSATA.

Wear-leveling just reduces the error rate visible to the user. It does not change the fundamental physics of MLC v. TLC when it comes to the errors that continually occur in every NAND element of the IC. At some point, the element breaks down because there are just too many errors, a median of 1,000 and 100 writes in MLC and TLC, respectively. Wear-leveling doesn't change that, it only mitigates how much the user is affected by it.

Which goes back to the fact that TLC can never be better than MLC, even against earlier designs of MLC. In fact, the main "complaint" is about the number of elements that go back and all of the mapping and everything else that results in decreased performance over lifespan.

Er, sort of, but I'm being anal.

Like 840 pre-Evo v. 840 Evo ... but go on.
(I admit my last 2 statements are purposely argumentative, I recognize it and understand I'm just kidding around there).

Ummm ... they are the exact same when it's a SATA pin-out! 2.5" SATA, mSATA and M.2-SATA are all the exact same chips! There's really no "different M.2 chips" for a SATA design from 2.5" SATA or mSATA.

The great thing about M.2-SATA, like mSATA before it ... is that we get to use the existing 2.5" SATA chips! Remember, mSATA was an alternative pin-out in Mini-PCIe so one didn't need to have a PCIe bridge on-card just to use an existing SATA NAND controller and NAND ICs!

Yes, we purposely regressed with mSATA from Mini-PCIe (although PCIe x1 wasn't really any faster, using AHCI).

 

What i meant is that all my friends who have TLC (m500/x100/840evo) NEVER had any problem with them. I personnally hate the 840evo (even withouth the faulty firmware ), but because for a rollercoaster curve, performance chart, i bought a sandisk ultra 2 960Go, it comes with a 3 year warranty, since it's a high capacity i expect it to have a way better life cycle (7-9 years easily), TLC is a natural response from the industry to the high price of MLC ssd, it's a great equilibrum between Price (yeah I'm a student) and performance. no need to remind is why we paid less over and over . nevertheless for ultra important files (that will probably be still here for all my professional life) and for extra performance for windows i bought a small 240go MLC sandisk extreme pro.

I'll update you on the forum if they begin building it. I'll do a review of the laptop if i get it fast enough, don't worry i'm rather good at doing reveiws, (this is one of my review in french ( [Review] Msi Gt60 2Od Gtx 780M - Les tests des membres - Portables4Gamers or there P150) )

AND if i get it and if HTwingnut can't do it, i will do a how-to for changing the screen to IPS, i already have my LG ips panel right next to me.


I expect my computer to be shipped within 2 weeks. (no 3k/4k panels or 4870hq) but we're never saved from a delay .

 

Hey again everyone, finally decided to get the P650, the big question is.. GTX 970M or 980M? (casual gaming) I want it to be futureproof for the next 3-4 years, but the 980M is an extra €300, is it worth it? Will the extra 4mm height added to the chassis be really noticeable? Would greatly appreciate your opinions, thanks.

 

I seriously doubt anyone can notice an extra 4mm unless they are side by side as for the 980M if your only into "casual" gaming the 970M should be more than enough but for futureproofing for tomorrow's more demanding games the extra money spent on the 980M is well worth it in my opinion.
The only question is how well it will cool the the extra power of the GPU (980M)and as far as I can tell it should be fine since it does a great job cooling 970M.
Is the 980M worth the extra 300 euros well that's a decision you will have to make yourself.
I have a hard time deciding myself but I know the the 970M is good enough but I always want the latest and greatest, either way you can't go wrong.

 

If you plan on keeping up with recently released games for 3-4 years, then definitely get the 980m. I honestly don't think the extra 4mm will matter a whole lot.

I'm opting for the 970m personally only because I like how cool it runs, and the cost too. I tend to upgrade my laptops frequently though, so in a year or so I will likely upgrade again if the right components fill a laptop. I don't game as much as I used to, life is too busy these days, but I do like to game when I can, but it's not my top priority for a laptop. In any case the 970m should be able to manage any game thrown our way in 2015 at 1080p with near max detail.

 

Thanks for your replies guys, greatly appreciated. 970M is more than enough but I think it's a better idea to get the best option you can afford especially in a laptop, so with that in mind I'll get the 980M version, which hopefully runs at good enough temperatures like the P650SE.

 

only the CPU should ""worry"" you (it's stil lower than my big fat ex-GT60), they managed the 970/980m temps clevo like a champ, i'm convinced of that .

The 980M is future proof, the 970M is high (with some things ultra ) proof for at least the next 3 years.

with both cards you're going to be happy, you should more make your choice regarding the games you play : if you're an enthusiast player of let's say DICE games (star wars battlefront FTW !!! ) and Bioware games than go with the 980M to fully enjoy their games , if you're more DOTA valve, blizzard go with the 970M it should be more than enough even in my opinion for source 2 upcoming games and Overwatch.

 

deleted

 

Depends on cost but shouldn't really bottleneck it, but some games are sensitive to CPU speed: http://forum.notebookreview.com/gaming-software-graphics-cards/763237-gtx-980m-limited-cpu.html

 

yep but tbh only games that were released a while ago are obviously concerned.

if you don't mind hitting more than 60 fps (not really useful anyway) than the 4710HQ won't disappoint you .


for thermal optimization i will cap all my games at 60fps (on ips that's not really useful more than 60).
and on battery cap at 30fps

 

Sorry guys, but I'm really confused with the storage options I have. I'm buying the P650SE on friday and it comes with 1TB HDD. Initially I planned to buy an extra 256gb M550 for OS, games etc, and then use the HDD for movies/storage. Read some of the previous posts, and some recommend just to sell the HDD and get a 500gb 2,5" SSD instead. Don't know what's best right now. 500GB should be enoug space for me, atleast for a while, so would that be the best?

About the height of the SSD, 7mm vs. 9mm. Is it enough to check the specifications at the retailer, and if it says less than 7mm height of the SSD, that should be good? If i want to fit another SSD in the future I mean..

Also, do I need like a "casing" or something, or should I be fine with just ordering the SSD and replace it with the HDD?

Really appreciate answers. Thanks

 

It's all up to you. You have many options when it comes to storage. If you feel like you really need the 1TB drive that comes with the laptop, and don't mind sacrificing performance a lil, then pick up an M.2 drive. Personally, I don't need that much space right now, so I'd rather swap out the HDD for a 512GB SSD. Performance matters more to me than raw storage. And with this layout I can always get a second 512GB SSD down the line when I need more.

And yeah that should work fine for checking 7mm, they usually put the info somewhere on the item description page though. And I don't think you need any mounting bracket for 7mm drives, only for 9.5mm drives (which comes included with the laptop).

 

Go with 980M. Worth the extra 300 bucks. It will easily support new games released in next 2-3 years. 4mm increase in overall thickness will not make any difference.

 

Okay, okay, so let's get this straight ...

The M500** and MX100 are MLC, not TLC.

So, just to recap ...
- You made a claim about 18+ months with consumer usage on TLC
- You don't know which products are TLC, let alone the 840 Evo hasn't been consumer available around that long
- And it took several responses for us to figure this out
- And ... this was all in response to me when an user had already picked a MLC solution (MX100), and several people were saying 840 Evo

In other words ... my point was that until I see proven, consumer usage, I will consider ...
- TLC a big risk for myself, plus
- When someone has already chosen a MLC product, but others try to persuade them that TLC is "better," "safer," etc...

That's not "the sky is falling" or anything else. That's merely being a technologist that has a viewpoint, does not fudge experience and actually takes the time to know the products they recommend, often from actual usage. Someone here picked a MLC solution and then we had the "840 Evo is God" (to use a polarizing phrase like "the sky if falling") start.

SIDE NOTE: I have many M4 and M500** MLC units myself. I've yet to have M500**part fail. It's probably influencing my personal bias to go with the M550 in my P65x purchase, even though I don't have much time with M550 parts.

**Since you guys are so big on AnandTech ...

'Ultimately [the M500] is Micron's answer to TLC for the time being. Rather than sacrificing endurance to get to lower price points, the 20nm 128Gbit 2bpc MLC NAND device at mature yields should deliver competitive pricing at higher endurance. Indeed this is the message behind Crucial's M500. The company isn't targeting Samsung's SSD 840 Pro, but rather the TLC based 840.'

- http://www.anandtech.com/show/6884/crucial-micron-m500-review-960gb-480gb-240gb-120gb

And now here it comes ...

Not even going to respond to that.

If there's one thing I've learned from from the industry when it comes to warranties, it's this ... it is never tied to actual product lifespan.

I.e., the fiscal "loss" to offer long warranties is minimal, especially when consumes claim them 6-18 months down the road. The only time it matters, fiscally, is when a product fails an abnormally high number of times in the first 6 months. Hence why I never quote warranties.

E.g., Several MLC vendors have even moved to 10 year warranties, and that doesn't mean anything to me. It merely tells me they've mitigated the fiscal risk in offering such, not that the product will last that long.

Then why is everyone still producing MLC products?

Which is a good mitigation. Finally, something we can agree on and is based on sound experience.

 

The only time height is a mounting consideration is when it is volume-space mounting, and you need a "spacer" to hold a 7mm in a 9.5mm volume-space (or a 5mm in a 7mm, etc...). If the 2.5" drive is mounted by screws on the side or underneath, then it doesn't matter if it's 7mm or 9.5mm -- or 5mm or 12mm or possibly even 15mm for that matter.

The P65xSE was designed with 7mm (or smaller) in mind, so a 2.5mm spacer for 9.5 is not even a consideration for you. Now I haven't seen a video where someone has removed the 2.5" drive cage from the P65xSE yet. Although I fully admit I might have overlooked a video where someone did remove it and show that detail. We also have resellers with the units now, so maybe that can provide that first-hand experience.

Understand I'd like to see such to not only confirm the mounting approach, but to also see the orientation of the SATA connector edges. E.g., if they are inverted from one another, that means one can have a 5mm + 9.5mm in the P65xSE. If they are the same, horizontal orientation, then they are likely offset 7mm and one must always use 7mm (or smaller) drives when inserting 2 drives.

Most of the 9.5mm "adapters" for 7mm devices are nothing more than a frame of plastic that is secured by screws or, worse yet, just adhesive. Here's one that is screwed in (I fully admit, this was a Google search -- but I knew what I was looking for as I have about 10 of them here at home).
- http://images.anandtech.com/reviews/storage/Intel/520/_DSC8970.jpg

 

mythlogic said they're "exploring options" for a 1080p IPS panel and that it'll probably be around $150 for the upgrade, which they'll do for you per the upgrade plan. I'm most likely just getting the TN panel version then with a 970m but now I don't know which SSD to get after reading the last few pages...

 

I have two MLC Crucial M550 512GB M.2 sticks collecting dust until the SG model comes out. Only $450 for the two on ebay.

 

If you're ordering from Mythlogic ... why not ask them? They seem to do everything in-house.

They might also be able to confirm the 512GB M.2-SATA and 2.5" SATA M550 MLC NAND devices both have 512MiB DRAM and perform the exact same, if you're debating on M.2 v. 2.5". Or if you're considering a TLC like the 840 Evo, they might be able to advise on their M.2-SATA v. 2.5" SATA options too.

FYI, the MX100 MLC NAND is 2.5" only. I haven't used any of the most expensive, high-end MLC NAND devices like the Samsung 850 Pro and SanDisk Ultra Plus in a notebook, so I don't know if they are available in M.2. /me Googles

Spoiler

I went with Mythlogic and the M550 M.2-SATA solution for my primary storage, including Windows. I decided I boot Windows so infrequently I'm going to put C:\Windows on it. I've also yet to have a M500 mSATA-30x50mm or 2.5" SATA device fail me yet.

I'm still going to have swap file, TEMP/TMP and temporary/variable data on a 2.5" platter. I.e., I'm migrating an existing 2.5"x7mm 1TB platter (yes, 7mm high, there are now 2-platter 2.5"x7mm drives).

I will consider adding a 2nd M.2 and 2.5"x7mm NAND devices in the future, likely when 1 and 2TB MLC devices become available in M.2 and 2.5"x7mm, respectively.

 

b.j., we need to get you a woman.

 

That doesn't have to be the case. One can get mSATA and/or M.2-SATA (B-keyed) slot adapters to 2.5" bays or even flexible PCB (for mounting anywhere) for $10-30.

I picked up a mSATA + M.2-SATA (has both, although only 1 can be used at a time) to standard 7+15-pin data+power adapter for under $20 the other month so I can read/recover any mSATA or M.2-SATA device. With an external eSATA port, or USB to SATA adapter in the worst case, I could recover anything with any system.

Just a consideration in case a system ever goes "belly up" and you cannot wait on the warranty turn-around because you need your data.

E.g., M.2 adapter in a full, mountable 2.5" housing (although one of the more expensive ones) ...
- http://www.startech.com/HDD/Adapters/M.2-NGFF-to-2.5in-SATA-Adapter-with-Housing~SAT2M2NGFF25

I have something exactly like this (didn't buy from this guy), which is an open bay (and requires one to secure on their own, if it is to be mounted).
- http://www.ebay.com/itm/like/261466873862?lpid=82

There are all sorts of combo, dual and even quad adapters all over eBay. This was interesting, although with only a Molex power connector (no +3.3V, only +5V and +12V), I'm concerned M.2-SATA cards might require +3.3VDC.
- http://www.microsatacables.com/m-2-...ii-x-4-port-adapter-with-3-5-inch-metal-frame

Just keep in mind if you buy a PCIe slot adapter, it needs to either be a M.2-PCIe, have it's own SATA controller on-board the card or the device can support falling back to a PCIe mode. I only have mSATA devices, and will have M.2-SATA shortly, so I only need the SATA pin-out (and power).

 

There are 7mm aka "slim" drives and standard 9.5mm drives. The bracket accommodates two 7mm or one 9.5mm. You don't need any special hardware.



About SSD discussion: anyone that has their SSD's fail after 18 months are unlucky, plain and simple. It's been shown, 200TB+ written to an SSD keeps on working. Performance degraded, but data is intact. The failure mode is usually the controller dies, not the NAND.

 

Alright I think I'm gonna go with this setup, everything else standard:

System Memory: 16GB (2 x 8GB) , PC3-12800, 1600MHz SODIMM
m.2 Slots: Samsung 128GB XP941 m.2 PCIe x4 Solid Sate Drive
Hard Drive: 1TB 7200 32MB Cache SATA Notebook Hard Drive

Mythlogic said the XP941 is a lot faster and the heating isn't anything to worry about. The largest file transfer I'll be doing will be the initial one between my old laptop and the new one. Won't really be running anything remotely demanding on the SSD besides maybe some matlab homework. If I buy my own HDs, how much money could I save overall? About $100? How difficult is it to install the SSD myself? I imagine not very? Also is there much difference between the fancier sounding 4x4 memory and the 2x8? Thanks for help.

 

I think my wife of 20 years might mind.

Spoiler

The picture I put on my desk at clients or the background of my desktop, in you case you wonder if I'm a real person and what my passions are.


But go ahead ... have fun at my expense. Even accuse me of getting this out of a stock magazine photo.

 

Right here: http://forum.notebookreview.com/sag...50se-gtx-970m-htwingnut-s-review.html#storage

 

Big time thanx!

Although the bay doesn't have the SATA connectors, from the holes for side mounting, it looks like to me the drives are spaced 7mm apart in the same, horizontal orientation. So this means if you have 2 drives, you must have one no larger than 7mm, and something like a 9.5mm + 5mm is not an option. One can, of course, have a single 9.5mm (or even 12mm, possibly even 15mm?).

Again, big time thanx!

 

Yes, that's the configuration option. One 9.5mm or two 7.0mm because they are stacked on top of each other. The 9.5mm has to go in the top bracket, but it will interfere with any drive below it.

 

Does anyone know if there are any US resellers planning to stock the P651SG?

Also has anyone connected an external display to the display ports yet to see if either is connected directly to the dGPU?

 

It's not so much the screw mount, but where the SATA card edges are.

If they are stacked atop of each in the same, horizontal orientation (top to bottom), then the SATA card edges would have fixed off-sets of 7mm (near bottom of each drive).

But ... if the tops of each other were facing, inverted horizontal direction (top to top), then the SATA card edges (as well as screw mounts) would be at the edges of the cage (near bottom of each drive). That would allow one to combine drives as long as the total size is under 14-15mm (or whatever it is) -- possibly a 5mm + 9.5mm.

But from what I saw in the screw mounts, it looks like they have fixed offsets of 7mm, period, no flexibility.

 

I ordered my P651SE (XMG P505 Pro) with the Crucial m550 256gb m.2 drive. I can do some tests if anyone wants. I saw some talk on the m550 a few pages back. Really hope they ship it on Monday, and then would expect it Wednesday.

 

Yeah, about 100 to 150 bucks depending on the deal.

Yup. Pop open the cover, identify the m.2 x4 slot, plug in the SSD card and screw it in place.

No difference performance wise. But better to go with two 8GB ram modules as you will have 2 unused ram slots for future upgrade.

 

Now that makes me wonder if the SG can support two 9.5mm HDDs. Because we are gaining 4 mm of extra space height wise. Would be nice if Sager could consider this in their design.

 

No, but it can supposedly support 9.5mm plus 7mm, rather than the SE's 2x 7mm.