Windows 10: Samsung 960 EVO NVMe performance...

My new system is based on the Asus Prime Z370-A MB, that has two PCIe (x4), M.2_1 and M.2_2. Both of the sockets are populated with Samsung SSD 960 EVO 250 GB M.2 cards.

Windows 10 setup had selected the M.2_1 drive and installed without any issues. Windows 10 is up to date, including the patch released today, and Samsung NVMe driver installed.

Checking the performance of the "C" drive with CrystalDisk looks good:





Well, until the other NVMe drive is checked that is:





In every category, the "E" drive performs better than the the "C" drive, especially in write performance , and it puzzles me.

The only difference visible between the two is the used space; "C" has 47 while "E" has 7GBs used space.

Could the disk space utilization be responsible for the different performance, or the "C" drive came with some slight factory fault(s)?

While there are temperature differences between the two NVMe cards, the "C" drive has a heatsink and actually runs 2-4 degree cooler than the "E" drive:





Any advice would be appreciated, TIA...

:)

 

Windows 10 not starting after I upgraded bios on MSi B250M Bazooka!

I upgraded my WD blue to an Samasung 850 Evo then a 960 Evo. I first cloned the Hard Disk to the SATA SSD & it worked, then I cloned it to the M.2 NVME SSD & it was starting from the SATA SSD. I had to format both the SATA SSD and Hard Disk to make
it work on the M.2 NVME. I did that by going into Disk Management & I had to get the drives hot swappable in the bios beforehand. I had my SATA data & power cables off on both Hard Disk & SSD. The NVME M.2 960 Evo worked I got the drives online & I formatted
them, otherwise it was booting of SATA drives instead of my NVME M.2 no matter how much I fiddled with it in the bios trying to set 960 Evo as the first drive to boot. I recently did a bios update to my MSi B250M Bazooka.

All the drives were installed I did not think to get them offline while I was updating the bios. The only bootable one being 960 Evo M.2 NVME the other 2 formatted. I have looked for a way to do this my self but I didn't make much progress. I managed
to reinstall Windows on the Sata 850 Evo.

My Computer is a CyberPowerPC ET8282-2162 Model C Series. I upgraded the processor to a i7 7700K from a i5 7400 & changed the video card to a 1080 Ti. I tried to figure out a way of creating a system restore on the 850 Evo & move the file to the 960 Evo
because it does not see a system restore file created. I tried all the troubleshooting ways possible. I am currently trying to get Windows media creation tool & run that off an USB see if that gives me any other options.

I tried to go to diskpart and listed the volumes & see if I can copy the recovery or system to the 960 Evo NVME M.2 drive but I don't know if it will work or it will still tell me I have no recovery file saved. I should of had 1 when I upgraded to
fall creators update but it's telling me I don't have one, I don't have other drives where I have Windows 10 clones because I deleted them if 960 Evo got corrupted somehow. Good thing is that I can still see the drive & I haven't lost any files. I have steam
games bitdefender installed & I still have another 7 months or so out of the subscription. I am afraid with a new install it will see it as a new computer. Is there a way to port over the Apps installed on the 960 Evo to the 850 Evo & then I can clone the
960 Evo with the fresh install.

I did the same thing on a CyberPowerPC ET8260-5322 I upgraded the CPU to a 1700X changed the ram to 3200 Mhz the videocard stayed the same 580 4Gb. I have the same 960 Evo & Hard Disk in it & I haven't have a problem with it. The Bios update on the
AMD setup was fine while the Intel suffered. On the AMD I have like 15 days of Kaspersky so I wouldn't of had a problem reinstalling Windows on that anyway + other apps as usual, which will take a while to install. I'd rather not do that if I can.

Thank you for your Help!

 

Directories with HEVC/x265 or Hi10p/10bit video sluggish

Same issue here since installing the Fall creator's update, and I have an absolute monster of a PC so non-sufficient hardware is NOT the cause of the issue (AMD 8c/16t Ryzen 7 1700 @ 4GHz, AMD R9 Fury X (which has native GPU HEVC decode support iirc),
32GB (16x2) 3200MHz CL15 G.Skill Trident Z DDR4, Samsung 960 EVO NVMe SSD).

 

There is nothing in your numbers, slight differences are to be expected.
The benchmarks are intentionally designed to bring out very small differences.

The small temperature difference, slight difference in interface, manufacturing batch, bin sorting and so on will give small differences in Drives.

Real world there is no difference, they are essentially both the same. Chill out, be happy.

 

There's always going to be performance fluctuation, see Performance Consistency - The Samsung 960 Pro (2TB) SSD Review

Your system (C) drive is generally going to have lower benches because the OS is doing stuff in the background. The lower amount of free space (potentially triggering garbage collection more frequently) as well as the DRAM cache already being filled with OS data before the benchmark is started would also affect performance.

 

Thanks for the chill, I am happy... *Biggrin

Your reasonings make sense, especially the part about the OS drive being active during the performance test.

The Anandtech's article about performance consistency is an interesting read, thank for the link. Since this PC does have a UPS backup, the Windows write cache buffer flushing is off.

Over all, I am not too impressed with this PC's performance. Coming from a PC with Samsung SSDs is probably the main reason. This PC is certainly faster than the SSD PC, but it isn't like stepping up from the HDD to the SSD. In my view, unless one need to run large programs, or works with large files, the SSD based PC is almost as good as the M.2 based PC. On the other hand the cost of the M.2 and the SSD drives is pretty much the same

With that said, creating a Macrium Reflect image of the C drive, around 25GBs, to the other M.2 drive , that was impressive. The image completed in 1:18, but that's not the purpose for the second M.2 drive...

 

If you were interested you could boot from PE (for example Windows 10 Recovery Tools - Bootable Rescue Disk - Windows 10 Forums) then download CrystalDiskMark6_0_0.zip

As neither disk would be being used (PE runs in RAM) you could compare them with no other overhead. I just tried it and the zip version runs fine in PE.

 

Thanks for the suggestion, I'll add the unzipped CrystalDiskMark to Reflect, or other bootable flash drive and test it.

I am just curious, if there are substantial performance differences between the M.2 cards/drives. In my view, over 10% difference is substantial...

 

You don't know that yet though do you? Not till you test it.

Another thing to consider is running CrystalDiskMark from PE might show a lower write speed (as you are using some RAM that would otherwise be used as cache). I also very much doubt PE is as efficient as Windows when it comes to memory management or anything else.

Only therefore compare relative performance of your disks from PE - don't compare performance running from PE against performance running from Windows.

I doubt you'll learn anything interesting but in your shoes I would definitely test it and see what happened. I would want to know.

Perhaps you will find both are the same (so speed deficit is due to OS running) or perhaps you'll find one really is worse. Then you'll have to swap them around and test again in case it is connection not drive.

It would be interesting anyway I think. If you do it please update your results.

 

I've tested and I know, now, you will too... *chuckle

Test result for the "C" drive in PE, with just Windows drivers:





And here's the result for the "D" drive:





In general, the performance differences between the two drives are the same as previously; not the values, just the difference between the two drive. Keep in mind that the test in PE did not have the Samsung NVMe driver. The MS NVME driver is pretty good, it seems....

The next test would be swapping the drives, but I am not going there. One of the reason is that the "C" drive has a heatsink attached to it that came with the MB built-in. The other is that the results might be the same; in another word, the current performance difference would still be there, And that would be either the drive in itself, or the MB; neither of them will be changed for a 100MB/s performance difference. Besides...

I tend to side with PolarNettles... Once the 50GBs data is dumped to the "E" drive, the chances are that its throughput performance will be similar/same as the "C" drive.

Nonetheless, I do appreciate your suggestions/guidance, thanks...

 

Me too generally. He/She know more than me I reckon.

I'll stick my oar in anyway though on occasion with some random theory. Thanks for the update anyway.

 

Enjoy the drive, I was one of the lucky few who pre ordered one of these and had it delivered the day they were available to the public. If you ever raid drives, one drive is the primary drive and the other is the secondary. the primary drive always gets used more so its performance numbers will show a bit lower.

Your post got me wondering about what numbers my drive would post. I have one 2TB drive. (no raid) I had a tun of crap running, so my test wasn't optimal and my server hasn't been rebooted in about two weeks, so you get the idea. I think the larger drive has something to do with the better numbers, but i might be wrong. I hope my screenshot helps.

 

The storage size of the drive does impact the performance to a certain extent, at least in the case of Samsung EVOs. The 250GBs EVO has a write speed of 1,500 MB/s, while the larger EVOs do 1,900 MB/s. All EVOs have the read speed 3,200 MB/s.

For backup and restore purposes, I like to separate the OS/programs from the data. It's easier to manage and marginally faster in the case of the ~20GBs database. Hence the two separate 250 GBs EVO drives. Cost wise, it's a wash when compared to a single 500 GBs EVO that provide an increase in write speed.

I am not even convinced that overall performance wise the system is much better than my SSD based system. Yes with PCIe 3.0 x4 drives Windows boots in a jiffy, large programs start faster, loads up large databases faster, etc., with Samsung EVO/PRO drives. but beyond that in everyday use of the system, the performance increase isn't as noticeable. Certainly, there is a performance increase, but it's nowhere close to moving from the HDD to SSD drives, despite the six fold increase in read speed. My guess is that accessing small, around couple of MB or so programs/files with the PCIe drives isn't that much different from doing the same with an SSD drive...

Your 2 TBs PCIe 3.0 x4 drive is probably a Samsung Pro that has a nominal read speed of 3,500 MB/s and write speed of 2,100 MB/s. Your benchmark results are right around that.

I've looked at the Samsung Pro, but for less than the 2 TBs Pro cost nowadays, I've built the whole system and glad I did. It's doubtful, if the performance of this system would be better with a single, large PCIe drive...

 

Hi,

That's generally true and to be expected.
One major factor is proper cooling of NVMe drives which is crucial to prevent thermal throttling. A cooler that only lowers the drive's temp by a few degrees is useless if you're talking degrees Celcius.
Much depends of the size of the casing the motherboard resides in and the airflow present.

For my next build I'll also pick a motherboard with two M2 PCIe x4 slots that I'll configure in raid 0 for maximum I/O throughput. Either using Samsung's 960 Pro's or Intel's Optane modules. Not decided yet but in everyday use the Intels may well beat the PRO's from Samsung and are much cheaper too although of course much smaller in capacity.

Cheers, *Wink

 

Most likely they will be kept in memory anyway so it is irrelevant.

You'll see the benefit in benchmarks (or loading/searching big things) but not in normal use.

I used to write back office software for banks. Until about mid 2000s I would force currency exchange rates to stay in RAM. This was a file less than 1KB. After that there was no point as memory management kept it RAM anyway.

Usually if a program is slow is because it is written badly - for example looking to read todays date (or USD/EUR rate or whatever) for each of 10 million records rather than once for all. Not much you can do about that except ask the developers to change their program.

The problem with most PC programs (including games) is they don't bother - just up the requirements to cover crappy programming.