Unless you do huge workloads, throttling is a non-issue. The SM951 has been tested on this site and they came to that conclusion. If your workload is so high that throttling becomes noticeable, just get an adapter with a heatsink.
Oh yeah, because if there is one thing that you can easily do, that is to put an SDD cooler adapter in your laptop or tablet :)
It could work for a PC, but it adds to the inconvenience factor. And as the testing here at AT established, the larger SSD actually experienced performance degradation from installing a heat sink, even if it was able to sustain its maximum sequential speed. I assume that's because the product was optimized with thermal throttling in mind, so eliminating that puts it on a less than optimal setting.
Which is why I'd prefer the problem is solved at the factory, and the product is optimized to perform at its peak.
I don't expect it would take more than a heat spreader, it doesn't even merit any significant volume of heat sink. But a naked chip on its own that is so fast to overheat and throttle back - bad idea.
Well said, ddriver probably doesnt even own a NVMe SSD but has read somewhere that under heavy synthetic tests they throttle down to save the drive. I own both the 256GB 950 Pro in a i7 laptop and a 512GB 950 pro in a 5820k Desktop and never hit any thermal issues on either platform.
Don't be such a hypocrite. Multi-channel data acquisition at high sampling frequencies (MHz+) can easily choke this drive. Throttling kicks in like 30 seconds in, and you lose HALF of the throughput. And it won't cost much to fix that, engineering, BOM or end price wise. So why not have it? Can you actually name a good and valid reason?
A basic heat spreader will also reduce the thermal stressing of the components and reduce the chance of fatal failures.
So what did you add to the discussion and how productive was it?
Regardless of the target or use, hypothesizing on a potential (even if far-fetched) problem/bottleneck and proactively creating a solution for it is the realm of any good planner or think tank. While at first I thought you were arguing ddriver, I see I was mistaken; you make good points and I generally will agree with you. BUT Who designs the cooling systems? The chip manufacturer, or the PC's OEM? I think the established point of responsibility falls to the OEMs. In fact, I think Dell is already sending out 'kits' with some of their M2 upgrades....which is an adapter and a heatsink. In the future I think we would hope to see some designs from OEMs that include allowances for improved ventilation and/or thermal solutions without the need for additional 'kits'.
As a side note, I ALWAYS add heat sinks to anything that I 'think' may end up needing it. It's just good practice for so many reasons.
Imagine usage such as 4k video editing and effects creation. In this field you can never have too much storage or too much speed. As you can rarely manage to have massive fast storage, many cases the solution is 'kinda fast' large storage supplemented by nosebleed fast storage connected by thunderbolt or PCIe. In this scenario you could make good use of one of these (or a raided pair) as your workspace, which would be read and written pretty much constantly. So yeah, heat spreader, and probably a fan.
In most cases this is true. But I doubt anyone will be putting such a drive into your typical casual use notebooks. The more likely candidates are mobile workstation grade laptops, and those do the same workloads as desktop workstations.
Regardless of the target or use, hypothesizing on a potential (even if far-fetched) problem/bottleneck and proactively creating a solution for it is the realm of any good planner or think tank. While at first I thought you were arguing ddriver, I see I was mistaken; you make good points and I generally will agree with you. BUT Who designs the cooling systems? The chip manufacturer, or the PC's OEM? I think the established point of responsibility falls to the OEMs. In fact, I think Dell is already sending out 'kits' with some of their M2 upgrades....which is an adapter and a heatsink. In the future I think we would hope to see some designs from OEMs that include allowances for improved ventilation and/or thermal solutions without the need for additional 'kits'.
As a side note, I ALWAYS add heat sinks to anything that I 'think' may end up needing it. It's just good practice for so many reasons.
....and are subject to much harsher handling and environments, ie: left on inside a laptop bag, baked/frozen in a car, rained/spilled on, dropped, filled with cat hair, etc.
ddriver, I suggest you read a bit about how NAND and flash in general work. You need to drive currents and dissipate power to write cells, no factory can get around that. And if you're writing to many of them over a short time span you dissipate lot's of power and the device heats up. It's simple physics and there are few ways the factory could get around that: - heat sink (you don't want that) - write slower (I suppose you don't want that either) - use NAND with bigger cells (more expensive - I don't think you want that)
I suggest you cut on the condescending tone and notice the flaws in your statements:
- it is not the the flash chips but the controller that's overheating
- a modest heat sink / heat spreader is exactly the remedy to the problem, it doesn't need to be anything bulky, just an interface to spread the heat to a larger surface so it can be more efficiently displaced
Thereby increasing z-height. It's not like laptops are space constrained or anything. Probably does not fit within the parameters of the M.2 standard as well.
If anything I see the need for a heatsink as a laptop OEM problem. Any good laptop oem would be more than able to add a heatspreader/heatpipe/etc to their laptop to counter this issue.
I figure if it was as easy as you say ddriver, they would do it. To me it sniffs of a compromise they cant or are unwilling to make.
Yes, it's the controllers temperature which triggers the throttling. But what heats the controller? The NAND surely plays a significant role in this, under sustained writes.
And sorry for the tone. I got carried away, feeling your other posts on that topic were rather provocative / ignorant / agressive. Seeing things more relaxed today.
Yes, it's the controller that throttles BUT the controler is the one writing to the flash chips and driving those voltages, some of them can be very high for modern digital stuff, and that will heat up the controller.
I've got a Latitude E5470. It came with a sata drive and when I wanted to utilize the M.2 port I had to buy a little mount and stock heatsink for the M.2 drive from dell. It's nifty, basically a flat piece of aluminum with thermal padding on one side, and it uses the airspace of the drive bay, which is ventilated by the cpu fan. I'm sure other vendors have similar solutions.
3 GB/s sequential reads...holy monkey. I thoroughly enjoy the 'problem' that SSD advancements and price drops are now so prolific that I hardly get excited anymore.
I believe (looking through forum information) that Lenovo already offers the PM961 in 256GB, 512GB, and 1TB sizes for their Thinkpads such as the X1 Carbon and X1 Yoga. You can buy them by searching for the replacement parts, newegg even had 512GB version. Search for "Lenovo 512 GB Solid State Drive PCIe-NVMe".
I am going to order a X1 Yoga with the 1TB drive this week, I will be able to confirm if it's the Samsung PM961 or not.
You probably mean the 960 Pro, going by Samsung's naming convention.
Wanna hear something funny though? The 850 Pro is cheaper than the 840 Pro on Amazon and even physical stores. Unfortunately, SSD prices don't behave exactly like HDD. They're more like RAM.
It's not that bad, PCIe 4.0 is around the corner(2017) and in x4 it's 8GB/s! In my day we were happy if we could afford a 10K HD Raptor with 46MB/s. If you are "stuck" with a SATA SSD you shouldn't worry about the newest stuff coming out.
At this point, I don't see speed as mattering much. This is plenty fast for the next few years.
What I'm looking forward to is larger drives beyond the 2TB that have begun to roll out, and along with that, the continuing drop in price per GB. For most purposes, those are more important specs.
The fact is that drive speeds have evolved enournously over the past few years, from HDD speeds to the nominally faster, but not always, speeds of the early SSDs, to this.
I'm curious if they can get read latency down any lower. Average latencies on the SM951 were in the hundreds of microseconds, which hurts for random read. For laptop/mobile applications being able to do faster small reads probably matters a lot more than sequential.
I imagine when it comes out it'll be like the very early SSDs were--used mostly by enterprise customers who want speed at more or less any cost--but just the fact that that's achievable at all suggests we could someday get there for consumer apps. (Also wonder if the existence of something higher-end will put downward pressure on Flash SSD prices. Not sure; maybe SSD makers don't have that much margin to give as it is.)
You're talking of 3D XPoint. This will be superior for latencies and bandwidth, also reliability, however, density is a lot worse than NAND at the moment. The technology slots between NAND Flash and DRAM, on most issues: higher density than DRAM, but lower than NAND, faster access than Flash, but lower than DRAM, higher reliability than NAND (100k+ write cycles/cell), but lower than DRAM... So, expect lower storage volume, faster speed and higher prices with early applications about caching.
About time they finally got a 1TB model out, now that 2.5 inch ssds can hit 2 TB. Now, how about all the heat these things release? My 950 pro idles quite warmly, as does almost every m.2 ssd ive seen.
As long as our computing model revolves around the conceptual model of "slow" solid state memory and fast but volatile RAM, there is relatively little to be gained by increasing the speed of slow storage, as we've built decades of software to work around the slowness of the disk.
Even as I write thousands of servers are being deployed to act as memory-caches for various applications to completely bypass to slow storage. No matter how much faster you make your storage, if your application is built to cache everything in memory and not touch the disk, speeding up the SSD won't make a meaningful impact in it's speeds.
Even replacing DRAM with non-volatile memory only starts having real impact once software architecture actually starts taking this into account. For example hundreds of millions of investments poured into database engines will be rendered pretty much obsolete as it has been for the big part been built to work around the volatility of the RAM and slowness of disks.
I'm going for a 1TB SM961 in my HP Elitebook 840 G3, that' for sure, as soon as they are possible to get.
I'm buying the laptop this week and will probably run a hp branded 250GB Nvme (possibly sm951 or something) + a 850 pro 1TB sata until I can get my hands on a 1TB sm961. Next stop after that: 3D xpoint/optane.
I tested the Samsung PM961 1TB with the following results from Crystal Disk Mark. Very impressive, overall though it looks like the Intel 750 is better.
I redid the test and increased the random thread count set up to 8 from 1 before (per recommendation). Results shown below. Results are closer to what I was expecting (esp the IOPS), but I think the Intel 750 is still slightly better. It will be interesting to see what the SM961 can do. ----------------------------------------------------------------------- CrystalDiskMark 5.1.2 x64 (C) 2007-2016 hiyohiyo Crystal Dew World : http://crystalmark.info/ ----------------------------------------------------------------------- * MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s] * KB = 1000 bytes, KiB = 1024 bytes
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
48 Comments
Back to Article
ddriver - Wednesday, March 23, 2016 - link
Looks good, but they better deal with the heat throttling.hansmuff - Wednesday, March 23, 2016 - link
Unless you do huge workloads, throttling is a non-issue. The SM951 has been tested on this site and they came to that conclusion. If your workload is so high that throttling becomes noticeable, just get an adapter with a heatsink.ddriver - Wednesday, March 23, 2016 - link
Oh yeah, because if there is one thing that you can easily do, that is to put an SDD cooler adapter in your laptop or tablet :)It could work for a PC, but it adds to the inconvenience factor. And as the testing here at AT established, the larger SSD actually experienced performance degradation from installing a heat sink, even if it was able to sustain its maximum sequential speed. I assume that's because the product was optimized with thermal throttling in mind, so eliminating that puts it on a less than optimal setting.
Which is why I'd prefer the problem is solved at the factory, and the product is optimized to perform at its peak.
ddriver - Wednesday, March 23, 2016 - link
I don't expect it would take more than a heat spreader, it doesn't even merit any significant volume of heat sink. But a naked chip on its own that is so fast to overheat and throttle back - bad idea.HollyDOL - Wednesday, March 23, 2016 - link
With that transfer rate there is a decent chance you run out of drive capacity sooner then you hit heat throttling :-)Cellar Door - Wednesday, March 23, 2016 - link
Absolutely that you will run out - and who will be dealing with such huge data sets for such long periods of time on a laptop, on a consistent basis.ddriver just wants to be part of the conversation, whether he has anything productive to add to the discussion or not.
Redstorm - Wednesday, March 23, 2016 - link
Well said, ddriver probably doesnt even own a NVMe SSD but has read somewhere that under heavy synthetic tests they throttle down to save the drive. I own both the 256GB 950 Pro in a i7 laptop and a 512GB 950 pro in a 5820k Desktop and never hit any thermal issues on either platform.ddriver - Wednesday, March 23, 2016 - link
Don't be such a hypocrite. Multi-channel data acquisition at high sampling frequencies (MHz+) can easily choke this drive. Throttling kicks in like 30 seconds in, and you lose HALF of the throughput. And it won't cost much to fix that, engineering, BOM or end price wise. So why not have it? Can you actually name a good and valid reason?A basic heat spreader will also reduce the thermal stressing of the components and reduce the chance of fatal failures.
So what did you add to the discussion and how productive was it?
daemoch - Sunday, May 1, 2016 - link
Regardless of the target or use, hypothesizing on a potential (even if far-fetched) problem/bottleneck and proactively creating a solution for it is the realm of any good planner or think tank. While at first I thought you were arguing ddriver, I see I was mistaken; you make good points and I generally will agree with you.BUT
Who designs the cooling systems? The chip manufacturer, or the PC's OEM? I think the established point of responsibility falls to the OEMs. In fact, I think Dell is already sending out 'kits' with some of their M2 upgrades....which is an adapter and a heatsink. In the future I think we would hope to see some designs from OEMs that include allowances for improved ventilation and/or thermal solutions without the need for additional 'kits'.
As a side note, I ALWAYS add heat sinks to anything that I 'think' may end up needing it. It's just good practice for so many reasons.
CalaverasGrande - Wednesday, May 18, 2016 - link
Imagine usage such as 4k video editing and effects creation. In this field you can never have too much storage or too much speed. As you can rarely manage to have massive fast storage, many cases the solution is 'kinda fast' large storage supplemented by nosebleed fast storage connected by thunderbolt or PCIe.In this scenario you could make good use of one of these (or a raided pair) as your workspace, which would be read and written pretty much constantly.
So yeah, heat spreader, and probably a fan.
saratoga4 - Wednesday, March 23, 2016 - link
I/O loads in a laptop are probably even less likely to hit throttling than in a desktop PC.ddriver - Wednesday, March 23, 2016 - link
In most cases this is true. But I doubt anyone will be putting such a drive into your typical casual use notebooks. The more likely candidates are mobile workstation grade laptops, and those do the same workloads as desktop workstations.saratoga4 - Wednesday, March 23, 2016 - link
Given how unlikely throttling is on a workstation, I'd say that's a solved problem then.daemoch - Sunday, May 1, 2016 - link
Regardless of the target or use, hypothesizing on a potential (even if far-fetched) problem/bottleneck and proactively creating a solution for it is the realm of any good planner or think tank. While at first I thought you were arguing ddriver, I see I was mistaken; you make good points and I generally will agree with you.BUT
Who designs the cooling systems? The chip manufacturer, or the PC's OEM? I think the established point of responsibility falls to the OEMs. In fact, I think Dell is already sending out 'kits' with some of their M2 upgrades....which is an adapter and a heatsink. In the future I think we would hope to see some designs from OEMs that include allowances for improved ventilation and/or thermal solutions without the need for additional 'kits'.
As a side note, I ALWAYS add heat sinks to anything that I 'think' may end up needing it. It's just good practice for so many reasons.
daemoch - Sunday, May 1, 2016 - link
Accidentally double posted the above comment, sorry. Ill delete it once i figure out how.daemoch - Sunday, May 1, 2016 - link
....and are subject to much harsher handling and environments, ie: left on inside a laptop bag, baked/frozen in a car, rained/spilled on, dropped, filled with cat hair, etc.MrSpadge - Wednesday, March 23, 2016 - link
ddriver, I suggest you read a bit about how NAND and flash in general work. You need to drive currents and dissipate power to write cells, no factory can get around that. And if you're writing to many of them over a short time span you dissipate lot's of power and the device heats up. It's simple physics and there are few ways the factory could get around that:- heat sink (you don't want that)
- write slower (I suppose you don't want that either)
- use NAND with bigger cells (more expensive - I don't think you want that)
ddriver - Wednesday, March 23, 2016 - link
I suggest you cut on the condescending tone and notice the flaws in your statements:- it is not the the flash chips but the controller that's overheating
- a modest heat sink / heat spreader is exactly the remedy to the problem, it doesn't need to be anything bulky, just an interface to spread the heat to a larger surface so it can be more efficiently displaced
doggface - Wednesday, March 23, 2016 - link
Thereby increasing z-height. It's not like laptops are space constrained or anything. Probably does not fit within the parameters of the M.2 standard as well.If anything I see the need for a heatsink as a laptop OEM problem. Any good laptop oem would be more than able to add a heatspreader/heatpipe/etc to their laptop to counter this issue.
I figure if it was as easy as you say ddriver, they would do it. To me it sniffs of a compromise they cant or are unwilling to make.
MrSpadge - Thursday, March 24, 2016 - link
Yes, it's the controllers temperature which triggers the throttling. But what heats the controller? The NAND surely plays a significant role in this, under sustained writes.And sorry for the tone. I got carried away, feeling your other posts on that topic were rather provocative / ignorant / agressive. Seeing things more relaxed today.
extide - Friday, March 25, 2016 - link
Yes, it's the controller that throttles BUT the controler is the one writing to the flash chips and driving those voltages, some of them can be very high for modern digital stuff, and that will heat up the controller.Methodical - Monday, March 28, 2016 - link
I've got a Latitude E5470. It came with a sata drive and when I wanted to utilize the M.2 port I had to buy a little mount and stock heatsink for the M.2 drive from dell. It's nifty, basically a flat piece of aluminum with thermal padding on one side, and it uses the airspace of the drive bay, which is ventilated by the cpu fan. I'm sure other vendors have similar solutions.jsntech - Wednesday, March 23, 2016 - link
3 GB/s sequential reads...holy monkey. I thoroughly enjoy the 'problem' that SSD advancements and price drops are now so prolific that I hardly get excited anymore.hojnikb - Wednesday, March 23, 2016 - link
I really wish samsung made retail versions of this. Atleast PM961 could be branded as 950EVO or something.Also, B-key versions would be awsome. Pretty much all pcie drives are m-key only, which means some laptops wont accept them.
Arnulf - Wednesday, March 23, 2016 - link
Probably because PCIe drives are supposed to be top performers and gimping them to PCIe x2 would negate that?Billy Tallis - Wednesday, March 23, 2016 - link
Do many laptops actually route two PCIe lanes to B keyed M.2 2280 slots, or are they just providing SATA?hojnikb - Wednesday, March 23, 2016 - link
Thats actually a very good question. Well, they could, given that b-key supports sata and 2 lane pcie.cptcolo - Sunday, June 12, 2016 - link
I believe (looking through forum information) that Lenovo already offers the PM961 in 256GB, 512GB, and 1TB sizes for their Thinkpads such as the X1 Carbon and X1 Yoga. You can buy them by searching for the replacement parts, newegg even had 512GB version. Search for "Lenovo 512 GB Solid State Drive PCIe-NVMe".I am going to order a X1 Yoga with the 1TB drive this week, I will be able to confirm if it's the Samsung PM961 or not.
arayoflight - Wednesday, March 23, 2016 - link
Now waiting for the 1050 pro, so that I can buy the 950 pro for cheap.lilmoe - Wednesday, March 23, 2016 - link
You probably mean the 960 Pro, going by Samsung's naming convention.Wanna hear something funny though? The 850 Pro is cheaper than the 840 Pro on Amazon and even physical stores. Unfortunately, SSD prices don't behave exactly like HDD. They're more like RAM.
damianrobertjones - Wednesday, March 23, 2016 - link
"Looking Ahead: PCIe 3.0 x4 May Become a New Performance Limiting Factor for SSDs"Of course... they can then announce something else, create a few different form factors, increasing their $$$$ each step of the way.
Stan11003 - Wednesday, March 23, 2016 - link
It's not that bad, PCIe 4.0 is around the corner(2017) and in x4 it's 8GB/s! In my day we were happy if we could afford a 10K HD Raptor with 46MB/s. If you are "stuck" with a SATA SSD you shouldn't worry about the newest stuff coming out.melgross - Wednesday, March 23, 2016 - link
At this point, I don't see speed as mattering much. This is plenty fast for the next few years.What I'm looking forward to is larger drives beyond the 2TB that have begun to roll out, and along with that, the continuing drop in price per GB. For most purposes, those are more important specs.
The fact is that drive speeds have evolved enournously over the past few years, from HDD speeds to the nominally faster, but not always, speeds of the early SSDs, to this.
damianrobertjones - Wednesday, March 23, 2016 - link
We all know that they could be MUCH faster but, as with a lot of things, the slow gradual crawl towards new tech ensures maximum $$$$.saratoga4 - Wednesday, March 23, 2016 - link
I'm curious if they can get read latency down any lower. Average latencies on the SM951 were in the hundreds of microseconds, which hurts for random read. For laptop/mobile applications being able to do faster small reads probably matters a lot more than sequential.twotwotwo - Wednesday, March 23, 2016 - link
In announcing their new kind of NVRAM Intel talked about latencies on the order of 10µs (http://www.anandtech.com/show/9470/intel-and-micro...I imagine when it comes out it'll be like the very early SSDs were--used mostly by enterprise customers who want speed at more or less any cost--but just the fact that that's achievable at all suggests we could someday get there for consumer apps. (Also wonder if the existence of something higher-end will put downward pressure on Flash SSD prices. Not sure; maybe SSD makers don't have that much margin to give as it is.)
twotwotwo - Wednesday, March 23, 2016 - link
That link should be: http://www.anandtech.com/show/9470/intel-and-micro...frenchy_2001 - Thursday, March 24, 2016 - link
You're talking of 3D XPoint. This will be superior for latencies and bandwidth, also reliability, however, density is a lot worse than NAND at the moment.The technology slots between NAND Flash and DRAM, on most issues: higher density than DRAM, but lower than NAND, faster access than Flash, but lower than DRAM, higher reliability than NAND (100k+ write cycles/cell), but lower than DRAM...
So, expect lower storage volume, faster speed and higher prices with early applications about caching.
jjj - Wednesday, March 23, 2016 - link
Nice that random write gets a big boost vs the 950 Pro.TheinsanegamerN - Wednesday, March 23, 2016 - link
About time they finally got a 1TB model out, now that 2.5 inch ssds can hit 2 TB. Now, how about all the heat these things release? My 950 pro idles quite warmly, as does almost every m.2 ssd ive seen.zepi - Thursday, March 24, 2016 - link
As long as our computing model revolves around the conceptual model of "slow" solid state memory and fast but volatile RAM, there is relatively little to be gained by increasing the speed of slow storage, as we've built decades of software to work around the slowness of the disk.Even as I write thousands of servers are being deployed to act as memory-caches for various applications to completely bypass to slow storage. No matter how much faster you make your storage, if your application is built to cache everything in memory and not touch the disk, speeding up the SSD won't make a meaningful impact in it's speeds.
Even replacing DRAM with non-volatile memory only starts having real impact once software architecture actually starts taking this into account. For example hundreds of millions of investments poured into database engines will be rendered pretty much obsolete as it has been for the big part been built to work around the volatility of the RAM and slowness of disks.
JKJK - Monday, April 18, 2016 - link
I'm going for a 1TB SM961 in my HP Elitebook 840 G3, that' for sure, as soon as they are possible to get.I'm buying the laptop this week and will probably run a hp branded 250GB Nvme (possibly sm951 or something) + a 850 pro 1TB sata until I can get my hands on a 1TB sm961.
Next stop after that: 3D xpoint/optane.
nagi603 - Sunday, March 27, 2016 - link
2280 again. When will we have 2242? Not all notebooks / motherboards can actually fit 2280...nerd1 - Tuesday, March 29, 2016 - link
Stick with mydigitalSSD drives for now.daemoch - Sunday, May 1, 2016 - link
Found this as a link from another site. They say it's built on Samsung’s third-gen 48-layer V-NAND.http://www.computerbase.de/2016-03/samsung-sm961-m...
TheinsanegamerN - Thursday, June 23, 2016 - link
I wonder if we will actually be able to purchase the 1TB model this time.cptcolo - Saturday, July 30, 2016 - link
I tested the Samsung PM961 1TB with the following results from Crystal Disk Mark. Very impressive, overall though it looks like the Intel 750 is better.-----------------------------------------------------------------------
CrystalDiskMark 5.1.2 x64 (C) 2007-2016 hiyohiyo
Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
* KB = 1000 bytes, KiB = 1024 bytes
Sequential Read (Q= 32,T= 1) : 2817.261 MB/s
Sequential Write (Q= 32,T= 1) : 1246.530 MB/s
Random Read 4KiB (Q= 32,T= 1) : 447.458 MB/s [109242.7 IOPS]
Random Write 4KiB (Q= 32,T= 1) : 401.962 MB/s [ 98135.3 IOPS]
Sequential Read (T= 1) : 1224.242 MB/s
Sequential Write (T= 1) : 1145.173 MB/s
Random Read 4KiB (Q= 1,T= 1) : 38.463 MB/s [ 9390.4 IOPS]
Random Write 4KiB (Q= 1,T= 1) : 144.299 MB/s [ 35229.2 IOPS]
Test : 1024 MiB [C: 14.3% (136.0/952.6 GiB)] (x5) [Interval=5 sec]
Date : 2016/07/30 11:47:06
OS : Windows 10 Professional [10.0 Build 10586] (x64)
Samsung PM961 1TB (NVMe M.2)
cptcolo - Saturday, July 30, 2016 - link
I redid the test and increased the random thread count set up to 8 from 1 before (per recommendation). Results shown below. Results are closer to what I was expecting (esp the IOPS), but I think the Intel 750 is still slightly better. It will be interesting to see what the SM961 can do.-----------------------------------------------------------------------
CrystalDiskMark 5.1.2 x64 (C) 2007-2016 hiyohiyo
Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 bytes/s [SATA/600 = 600,000,000 bytes/s]
* KB = 1000 bytes, KiB = 1024 bytes
Sequential Read (Q= 32,T= 1) : 2882.852 MB/s
Sequential Write (Q= 32,T= 1) : 1260.938 MB/s
Random Read 4KiB (Q= 32,T= 8) : 1137.597 MB/s [277733.6 IOPS]
Random Write 4KiB (Q= 32,T= 8) : 912.272 MB/s [222722.7 IOPS]
Sequential Read (T= 1) : 1239.127 MB/s
Sequential Write (T= 1) : 1161.739 MB/s
Random Read 4KiB (Q= 1,T= 1) : 39.088 MB/s [ 9543.0 IOPS]
Random Write 4KiB (Q= 1,T= 1) : 144.358 MB/s [ 35243.7 IOPS]
Test : 1024 MiB [C: 14.3% (136.0/952.6 GiB)] (x5) [Interval=5 sec]
Date : 2016/07/30 12:37:07
OS : Windows 10 Professional [10.0 Build 10586] (x64)
Samsung PM961 1TB (NVMe M.2)