I don't care that much about speed - ~600-800MB/s is fine with me.
What is NOT fine is it crawling at ~250MB/s - -barely 2x HDD speeds - after only 2 months of use.
And there being no plans or way of fixing it. And I paid HOW much for this?
At this rate, there is zero benefit in using the SSD for the OS disk, better to use the HDD and use the SSD as a fast scratch disk where it only ever is used for new files.
My advice to anybody considering Alienware: don't get the SSD option. Upgrade later when this fiasco is finally resolved (most likely with new hardware, or the next generation of SSD technologies.
RST not relevant for SSD on this system, only sees the HDD.
Has NVM driver from MS (come back to that later...)
Trim is enabled, my understanding is that garbage collection is redundant these days, and also that OP is invisible to the user and AFAIK will still be set as default for the SSD (Drive Magician doesn't work with this drive (OEM), and I haven't changed any settings). I have free space on the drive, anyway.
The problem as far I I've been able to find so far, is that as files age, their read speed decreases. Benchmarks that write fresh files won't show the problem, but HDTune will. The ATTO benchmark is still interesting though, see below.
Of interest is the fact that this system, as set up by default, appears to be using the MS Standard NVM Express driver, 10.0.10586.0 dated 2006...
Why isn't a more recent driver installed, where can I get it? I have all the updated drivers from Alienware.
Sorry, I didn't catch it was PCie/NVM ... so I guess that is right ... no Intel RST. However, if it was SATA based SSD, pretty sure it would be in the control-chain.
Right, Win-10 is fully SSD capable with Trim, Trash-Collection, etc. It should auto-detect SSD and setup appropriately.
Right, no Samsung Magician with OEM drives. I reported that to Dell long ago.
Further trials seem to indicate this is thermal throttling (which is surprising, since the benchmarks are quite short and ambient temperatures are not high). The SSD doesn't show temperature, but the HDD shows 33C, which seems a fair indication of the case internal temperatures (not very high).
A series of benchmarks below:
First, the first benchmark on system after it has been idling overnight.
A benchmark run immediately after the first, notice the speed reduces at a different location on the disk compared tot he first benchmark, other artifacts possibly related to the file structure are in the same positions:
Third benchmark, speed drops almost immediately:
It is possible to benchmark at the throttled state from the beginning:
And this is repeatable:
Let the system cool off for a short time, and we're back to being closer to the first benchmark:
I can reproduce the effect reliably - after running any disk intensive operation subsequent benchmark always show the same reduction in performance, even though ambient temperature isn't high (about 20-25C)
Whether this will ever show in normal use is going to be hard to identify, but this has got to be bad on systems with smaller enclosures like the 13 and 15" systems.
FYI, new XPS 15 9550 also comes with PM951. However, unlike teardowns of 17R3 I've seen which seem to have no thermal treatment, my and others 9550 has a thermal pad which covers entire M.2 card and also contacts the aluminum base and/or an insulated copper foil to spread heat. No one with 9550 has reported PM951 slowdowns like this that I know of.
(Waiting for CrystalDiskInfo to support temps reporting for the PM951.)
The 9550 DOES, however, have a problem with slow PM951 speeds when running on battery or after sleep. Max speeds good on a cold boot on A/C, though. Might be worth trying a test on a cold A/C boot if you're not doing that already.
HD Tune 2.55 (free) does not support Windows 10. It does not test write ether. It might mixing up write and read speeds. The purpose of the most freeware is to freak you out there is something wrong and buy full version.
You have spent a fortune on a laptop; can't you afford a $35 on software? If you do not spend any on principle, I suspect your system is full of free *** dragging everything down.
FYI, new XPS 15 9550 also comes with PM951. However, unlike teardowns of 17R3 I've seen which seem to have no thermal treatment, my and others 9550 has a thermal pad which covers entire M.2 card and also contacts the aluminum base and/or an insulated copper foil to spread heat. No one with 9550 has reported PM951 slowdowns like this that I know of.
(Waiting for CrystalDiskInfo to support temps reporting for the PM951.)
The 9550 DOES, however, have a problem with slow PM951 speeds when running on battery or after sleep. Max speeds good on a cold boot on A/C, though. Might be worth trying a test on a cold A/C boot if you're not doing that already.
That's interesting, shows that Alienware are aware of the problem and addressing it. Thanks for the info. Maybe carbon fibre doesn't work so well as a p[possible heatsink :-)
Haven't even tried benchmarking on battery, have only used mains power (in fact I rarely use this system on battery). What order of speed reduction are you seeing?
I guess the fact that this issue shows when benchmarking has little relevance for everyday use (or so one would hope).
HD Tune 2.55 (free) does not support Windows 10. It does not test write ether. It might mixing up write and read speeds. The purpose of the most freeware is to freak you out there is something wrong and buy full version.
You have spent a fortune on a laptop; can't you afford a $35 on software? If you do not spend any on principle, I suspect your system is full of free *** dragging everything down.
ORLY?
What I choose to spend money on is my choice: there is absolutely no need for you to be pious and judgmental (incorrectly, as it happens) about it. And extrapolating from a couple of screenshots to a judgement on the likely state of my system is just plain absurd, no matter how smug it might make you feel.
Your comments also do the guys at HDTune a disservice.
Free CrystalDiskMark has given me and others good high-level results with NVMe SSDs like PM951, 950 PRO, etc. My results for the PM951 256GB in the XPS 15 9550 for comparison:
Oh, one other thing, as far as I know, ALL XPS 15 9550 models were shipped with PM951 running under RAID (Intel RST). Not sure why, but could be related to Trim support. Haven't tested with TrimCheck or a Hex Editor, but might be something to look into.
Free CrystalDiskMark has given me and others good high-level results with NVMe SSDs like PM951, 950 PRO, etc. My results for the PM951 256GB in the XPS 15 9550 for comparison:
Oh, one other thing, as far as I know, ALL XPS 15 9550 models were shipped with PM951 running under RAID (Intel RST). Not sure why, but could be related to Trim support. Haven't tested with TrimCheck or a Hex Editor, but might be something to look into.
Doesn't really mean much unless there is a difference between hot and cold.
For example, when cold I get:
Which is, in practical terms, identical to your benchmark.
However, if I run HDTune a couple or three times to heat the SSD up, I get a dramatically different result:
Which correlates with HDTune benchmark results. DO you ever see a similar degradation, with the heatsink installed, or does it just never happen? Just repeatedly run HD Tune until performance drops - in my case, about twice.
The problem is, Crystal Disk Mark doesn't heat the drive up like HDTune does.
My 17R3 shipped with RAID enabled, subsequently switched to AHCI with no noticeable effects. Trim Check shows trim is working just fine (as reported by fsutil, too).
eternalkp
314 Posts
0
December 31st, 2015 06:00
We all wish Alienware uses the SM951 instead. But the PM951 is not that fast to begin with.
www.samsung.com/.../MZVLV256HCHP;CID=AFL-hq-mul-0813-11000170
hindesite
276 Posts
0
December 31st, 2015 14:00
I don't care that much about speed - ~600-800MB/s is fine with me.
What is NOT fine is it crawling at ~250MB/s - -barely 2x HDD speeds - after only 2 months of use.
And there being no plans or way of fixing it. And I paid HOW much for this?
At this rate, there is zero benefit in using the SSD for the OS disk, better to use the HDD and use the SSD as a fast scratch disk where it only ever is used for new files.
My advice to anybody considering Alienware: don't get the SSD option. Upgrade later when this fiasco is finally resolved (most likely with new hardware, or the next generation of SSD technologies.
Tesla1856
8 Wizard
•
17.4K Posts
0
December 31st, 2015 15:00
What machine do you have? What OS are you running?
What HDD/SSD drivers are you running, and are you running Intel-RST (if so, what version).
Have your verified OS-Trim is enabled? Forced a Trash-Collection? How much of the drive is Over-Provisioned?
en.community.dell.com/.../19568143
Try ATTO DiskBench
hindesite
276 Posts
0
December 31st, 2015 17:00
17R3, Windows 10 (1511)
RST not relevant for SSD on this system, only sees the HDD.
Has NVM driver from MS (come back to that later...)
Trim is enabled, my understanding is that garbage collection is redundant these days, and also that OP is invisible to the user and AFAIK will still be set as default for the SSD (Drive Magician doesn't work with this drive (OEM), and I haven't changed any settings). I have free space on the drive, anyway.
The problem as far I I've been able to find so far, is that as files age, their read speed decreases. Benchmarks that write fresh files won't show the problem, but HDTune will. The ATTO benchmark is still interesting though, see below.
Of interest is the fact that this system, as set up by default, appears to be using the MS Standard NVM Express driver, 10.0.10586.0 dated 2006...
Why isn't a more recent driver installed, where can I get it? I have all the updated drivers from Alienware.
Tesla1856
8 Wizard
•
17.4K Posts
0
January 1st, 2016 01:00
Sorry, I didn't catch it was PCie/NVM ... so I guess that is right ... no Intel RST. However, if it was SATA based SSD, pretty sure it would be in the control-chain.
Right, Win-10 is fully SSD capable with Trim, Trash-Collection, etc. It should auto-detect SSD and setup appropriately.
Right, no Samsung Magician with OEM drives. I reported that to Dell long ago.
See this
http://www.sqlskills.com/blogs/glenn/beware-of-the-native-microsoft-nvme-driver/
See if this will install
http://www.samsung.com/global/business/semiconductor/minisite/SSD/global/html/support/downloads.html
Samsung NVMe Driver Installer
Tesla1856
8 Wizard
•
17.4K Posts
0
January 1st, 2016 01:00
I don't think so ... you can see it.
It's un-partitioned space at end of drive. You can see it in Control Panel / Administrative Tools. However, that's with SATA SSDs.
hindesite
276 Posts
0
January 1st, 2016 15:00
Further trials seem to indicate this is thermal throttling (which is surprising, since the benchmarks are quite short and ambient temperatures are not high). The SSD doesn't show temperature, but the HDD shows 33C, which seems a fair indication of the case internal temperatures (not very high).
A series of benchmarks below:
First, the first benchmark on system after it has been idling overnight.
A benchmark run immediately after the first, notice the speed reduces at a different location on the disk compared tot he first benchmark, other artifacts possibly related to the file structure are in the same positions:
Third benchmark, speed drops almost immediately:
It is possible to benchmark at the throttled state from the beginning:
And this is repeatable:
Let the system cool off for a short time, and we're back to being closer to the first benchmark:
hindesite
276 Posts
0
January 1st, 2016 15:00
Drive has 4 partitions (EFI, OS, and 2x recovery of different sizes) no free space visible anywhere.
Maybe this generation of SSDs deal with it differently.
hindesite
276 Posts
1
January 1st, 2016 15:00
I think this is thermal throttling.
I can reproduce the effect reliably - after running any disk intensive operation subsequent benchmark always show the same reduction in performance, even though ambient temperature isn't high (about 20-25C)
Whether this will ever show in normal use is going to be hard to identify, but this has got to be bad on systems with smaller enclosures like the 13 and 15" systems.
Techgee
2 Intern
•
623 Posts
0
January 1st, 2016 17:00
FYI, new XPS 15 9550 also comes with PM951. However, unlike teardowns of 17R3 I've seen which seem to have no thermal treatment, my and others 9550 has a thermal pad which covers entire M.2 card and also contacts the aluminum base and/or an insulated copper foil to spread heat. No one with 9550 has reported PM951 slowdowns like this that I know of.
(Waiting for CrystalDiskInfo to support temps reporting for the PM951.)
The 9550 DOES, however, have a problem with slow PM951 speeds when running on battery or after sleep. Max speeds good on a cold boot on A/C, though. Might be worth trying a test on a cold A/C boot if you're not doing that already.
ManlyElectronic
4 Posts
0
January 1st, 2016 18:00
HD Tune 2.55 (free) does not support Windows 10. It does not test write ether. It might mixing up write and read speeds. The purpose of the most freeware is to freak you out there is something wrong and buy full version.
You have spent a fortune on a laptop; can't you afford a $35 on software? If you do not spend any on principle, I suspect your system is full of free *** dragging everything down.
hindesite
276 Posts
0
January 1st, 2016 19:00
That's interesting, shows that Alienware are aware of the problem and addressing it. Thanks for the info. Maybe carbon fibre doesn't work so well as a p[possible heatsink :-)
Haven't even tried benchmarking on battery, have only used mains power (in fact I rarely use this system on battery). What order of speed reduction are you seeing?
I guess the fact that this issue shows when benchmarking has little relevance for everyday use (or so one would hope).
hindesite
276 Posts
0
January 1st, 2016 19:00
ORLY?
What I choose to spend money on is my choice: there is absolutely no need for you to be pious and judgmental (incorrectly, as it happens) about it. And extrapolating from a couple of screenshots to a judgement on the likely state of my system is just plain absurd, no matter how smug it might make you feel.
Your comments also do the guys at HDTune a disservice.
Techgee
2 Intern
•
623 Posts
0
January 2nd, 2016 00:00
Free CrystalDiskMark has given me and others good high-level results with NVMe SSDs like PM951, 950 PRO, etc. My results for the PM951 256GB in the XPS 15 9550 for comparison:
Oh, one other thing, as far as I know, ALL XPS 15 9550 models were shipped with PM951 running under RAID (Intel RST). Not sure why, but could be related to Trim support. Haven't tested with TrimCheck or a Hex Editor, but might be something to look into.
hindesite
276 Posts
0
January 2nd, 2016 03:00
Doesn't really mean much unless there is a difference between hot and cold.
For example, when cold I get:
Which is, in practical terms, identical to your benchmark.
However, if I run HDTune a couple or three times to heat the SSD up, I get a dramatically different result:
Which correlates with HDTune benchmark results. DO you ever see a similar degradation, with the heatsink installed, or does it just never happen? Just repeatedly run HD Tune until performance drops - in my case, about twice.
The problem is, Crystal Disk Mark doesn't heat the drive up like HDTune does.
My 17R3 shipped with RAID enabled, subsequently switched to AHCI with no noticeable effects. Trim Check shows trim is working just fine (as reported by fsutil, too).