SSD Configuration Gurus Needed (slow 850 EVO issue solved)

[UsandThem]

I just placed an order for my new 500 GB 960 EVO, but there is something I haven't been able to figure out for a year now. I usually can search and fix it (or figure it out) myself, but this has left me stumped. I see people uploading their Samsung Magician and CrystalDiskMark scores, and the scores are much higher. I'm not trying to win any benchmark forum competitions, I just want to know what the heck I am missing and what is causing the IOPS score to be so low?

I have 3 desktops, and two laptops. They all have Samsung 850 EVO SSDs as the system drives. Some were cloned from a HDD, and a couple were clean installs. The laptops have 250 GB versions, and the desktops have 500 GB ones. They all have Windows 10 on them. Here is a screenshot of my system listed in my signature. The problem is this:

My IOPS write scores are much lower than they were when benching them on Windows 7 machines.

What I have already tried:

-TRIM enabled
-Properly Aligned
-Disabled System Restore
-Used Intel Rapid Storage Drivers
-Used stand-alone Intel chipset driver (currently installed)
-Hibernation is disabled on all
-Set specific page file, and also let Windows handle the size
-Tried changing power profile from balanced to high performance
-Write caching is enabled
-Went into BIOS and disabled Aggressive LPM Support for SATA
-All of the drives are connected to Intel controllers (SATA3_0) connection
-Some drives are formatted as GPT and some are MBR

 

[VirtualLarry]

Doesn't the Intel AHCI driver disable write-back caching? Or do I have that backwards?

 

[BonzaiDuck]

We're all learning new things -- why did you disable Aggressive LPM support? I enabled mine.

I can't speak to the IOPS scores, but your Seq-read and write scores in Magician and CrystaldiskMark are consistent and those are the results you would expect for an 850 EVO SATA drive.

IF you are moving up to an NVMe M.2, I'd advise configuring it in a PCIE-x4 slot instead of the M.2 motherboard slot. either way, you are likely to lose use of one pair of SATA ports on the SATA controller, but if there are six ports on the mobo SATA, you'll still have four left. Using the M.2 slot MAY NOT disable ports, but it will share bandwidth.

The PCIE adapters for NVMe M.2 are cheap or chump-change, probably some better and some worse, but figure on something between $15 and $30 before tax and shipping.

I bought a 960 Pro and noticed that benchtesting either the Pro or EVO as an unbootable data drive gives scores most consistent with the product spec, while using them for an OS-boot-system disk (my case -- dual-boot Win 7/10) will show scores about 500 MB/s slower. So my Pro drive shows sequential read under Magician bench at about 3,000 instead of 3,500.

Also -- if you created a dual-boot Win7/Win10 and you want to move it to a 960 NVMe, use Macrium Reflect "Free" and do the clone from within Win 10 -- shut down your system through Windows immediately thereafter and remove the source disk. Smooth as butter on toast.

 

[UsandThem]

Doesn't the Intel AHCI driver disable write-back caching? Or do I have that backwards?

I don't think so. It's just a standard Intel 100 Series/C230 Chipset SATA AHCI driver

 

[UsandThem]

We're all learning new things -- why did you disable Aggressive LPM support? I enabled mine.

I can't speak to the IOPS scores, but your Seq-read and write scores in Magician and CrystaldiskMark are consistent and those are the results you would expect for an 850 EVO SATA drive.

IF you are moving up to an NVMe M.2, I'd advise configuring it in a PCIE-x4 slot instead of the M.2 motherboard slot. either way, you are likely to lose use of one pair of SATA ports on the SATA controller, but if there are six ports on the mobo SATA, you'll still have four left. Using the M.2 slot MAY NOT disable ports, but it will share bandwidth.

The PCIE adapters for NVMe M.2 are cheap or chump-change, probably some better and some worse, but figure on something between $15 and $30 before tax and shipping.

I bought a 960 Pro and noticed that benchtesting either the Pro or EVO as an unbootable data drive gives scores most consistent with the product spec, while using them for an OS-boot-system disk (my case -- dual-boot Win 7/10) will show scores about 500 MB/s slower. So my Pro drive shows sequential read under Magician bench at about 3,000 instead of 3,500.

I only disabled the Agressive LPM long enough to rule it out as the issue. It is enabled now.

Here is an example of another CrystalDisk upload of a 850 EVO. It is the second column scores that are the issue (and the low random write IOPS are lower than other people's uploads (most are 50k+):

I plan on installing the 960 EVO on one of M.2 slots for now. It's going on the bottom slot, and when filled it only disables one PCIEX4 slot (which I don't use). The top slot disables 4 SATA ports.

I'll probably eventually pick up an adapter (currently eyeing the KryoM.2). But my wife is probably going to kill me over my new $250 "toy".

 

[VirtualLarry]

Those last pictures that you posted, those are of 2TB 850 EVO drives. No wonder they're faster.

Also, the pictures that you may have seen of 850 EVO drives, were probably from the original batches. They've "optimized" the V-NAND mfg since then, and they've moved to a more dense arrangement, which cuts down on write bandwidth, as I understand it.

So, make sure that you're making an apples-to-apples comparison.

Also, the 250GB model should be roughly half the performance of the 500GB model. That should be unsurprising to you.

 

[BonzaiDuck]

Doesn't the Intel AHCI driver disable write-back caching? Or do I have that backwards?

Backwards or not, it raises some additional questions or factors of interest, given my own experiences described in my post after yours.

 

[VirtualLarry]

That said, it's not out of the question that drivers or other OS components could have an effect as well.

My SM951 M.2 PCI-E AHCI 128GB MLC SSD, was giving me spec-rated sequentials, but my 4K QD32 read/write were about half.

Somehow, between Win7 updates, Win10 1511, and 1607, my speeds went up to what I had seen in a YouTube video.

Not really sure what "fixed" it.

 

[UsandThem]

That said, it's not out of the question that drivers or other OS components could have an effect as well.

My SM951 M.2 PCI-E AHCI 128GB MLC SSD, was giving me spec-rated sequentials, but my 4K QD32 read/write were about half.

Somehow, between Win7 updates, Win10 1511, and 1607, my speeds went up to what I had seen in a YouTube video.

Not really sure what "fixed" it.

The systems work fine and aren't slow, but I just noticed over the last couple of years when people uploaded CrystalDisk and Magician scores, my 4k QD32TI scores seemed lower (especially the writes). One of the 850 EVO drives were a carry-over from a Win 7 build, and the IOPS write scores were higher.

Who knows. It just has peaked my interest on the 'why'. I missed those other scores were larger drives. Here is review of the 500 GB version:

http://www.pcadvisor.co.uk/review/solid-state-drives/samsung-850-evo-500gb-review-3639493/

 

[Justinus]

TBH I've seen variations in SSD speeds based on motherboard model, chipset, bios settings, the alignment of the sun and the moon, and solar flares.

Seeing performance like that from a 6700K Z170 setup? I'm perplexed as usually it's in a laptop or on a budget chipset/setup that I see significant performance losses. The EVO drives have been spectacularly consistent from my experience.

Have you tried any other SATA ports (besides the asmedia garbage ones)?

Make sure to check for a firmware update through magician when you get your 960 EVO, mine updated the day after I got it giving me a 3-4% increase in random iops in the magician benchmark and crystaldiskmark.

Edit: Reread op, improved reading comprehension.

 

[UsandThem]

TBH I've seen variations in SSD speeds based on motherboard model, chipset, bios settings, the alignment of the sun and the moon, and solar flares.

Seeing performance like that from a 6700K Z170 setup? I'm perplexed as usually it's in a laptop or on a budget chipset/setup that I see significant performance losses. The EVO drives have been spectacularly consistent from my experience.

Have you tried any other SATA ports (besides the asmedia garbage ones)?

Yeah, all the drives are Intel SATA controllers. The systems the 850 EVOs are on are:. 2 x Skylake, 2 x z97, and z87. The motherboards are Gigabyte, Asus, and ASRock (plus two Dell laptops).

It's a odd one for sure. I've seen a few other posts at various hardware sites with the same issue, but a solution was never found.

Oh well, they are fast and work perfectly for what I use them for, but I always like to figure out issues like this. Maybe this will just be one that I can't.

 

[Justinus]

Yeah, all the drives are Intel SATA controllers. The systems the 850 EVOs are on are:. 2 x Skylake, 2 x z97, and z87. The motherboards are Gigabyte, Asus, and ASRock (plus two Dell laptops).

It's a odd one for sure. I've seen a few other posts at various hardware sites with the same issue, but a solution was never found.

Oh well, they are fast and work perfectly for what I use them for, but I always like to figure out issues like this. Maybe this will just be one that I can't.

Have you tried the suspect drive in another computer? I thought those performance figures looked familiar and I finally realized why.

I went to a coworkers place to help him tune up his system and his 850 EVO 250GB benches extremely similarly to yours. I assumed it was his AMD board having a weak SATA controller since nothing seemed out of place when I walked through the normal troubleshooting steps. I'm more interested now if there's an issue with the specific slow drives or something else going on.

 

[UsandThem]

Have you tried the suspect drive in another computer?

No, but all 5 of the 850 EVOs I have all score the same across all the systems. So I don't think the issue is with them, but some type of Windows 10 setting/configuration issue.

 

[UsandThem]

Ok, I remembered I had my old 830 EVO in the computer part closet, so I hooked it up to the Intel SATA_1 port, and these were the results:

So something is holding back the 850 EVO 4k Q32T1 scores. Now, I just have to figure out what it is.

 

[BonzaiDuck]

I haven't much fretted over my SATA SSD scores for whatever drives I've used: 840 Pro, 840 EVO, MX100, Patriot Blaze or Pyro, ADATA SP550.

I might still be a little puzzled over the performance of my 960 Pro 1TB as a dual-boot disk, including the system-reserved and FAT GUID volumes together with two large OS volumes leaving 100GB of unallocated space.

I tested the 960 Pro and the 960 EVO 250GB as simple data drives before deciding what to do with them, and they both met their specs within 100 MB/s more or less.

When the Pro became my dual boot drive, these are my CrystalDiskMark scores of a half-hour ago, taken to show on this thread. I think my web-pages space is getting full, or I would post the image. Don't have that much time, so here they are as text:

Seq Q32T1: read 2991, write 2040
4KQ32T1: read 1230, write 920.3
Seq: read 1692, write 1818
4K: read 53.01, write 288.5

This using the default 1GiB test.

I haven't decided there is anything wrong with this. Does anyone think there is something wrong?

 

[grimpr]

Remove the Intel AHCI driver and revert back to Microsoft AHCI, Intels driver is quirky.

 

[UsandThem]

Remove the Intel AHCI driver and revert back to Microsoft AHCI, Intels driver is quirky.

I'll give it a shot later and report back.

 

[UsandThem]

Seq Q32T1: read 2991, write 2040
4KQ32T1: read 1230, write 920.3
Seq: read 1692, write 1818
4K: read 53.01, write 288.5

I haven't decided there is anything wrong with this. Does anyone think there is something wrong?

Seems slower in some areas, and faster in others based on this review:

http://www.thessdreview.com/featured/samsung-960-evo-m-2-nvme-ssd-review-250gb1tb/3/

 

[Justinus]

Seems slower in some areas, and faster in others based on this review:

http://www.thessdreview.com/featured/samsung-960-evo-m-2-nvme-ssd-review-250gb1tb/3/

That looks like 960 EVO numbers and Bonzai got a 960 Pro. Those numbers are also really bad, my 1TB 960 EVO is dominating what they have for the 1TB.

 

[UsandThem]

That looks like 960 EVO numbers and Bonzai got a 960 Pro. Those numbers are also really bad, my 1TB 960 EVO is dominating what they have for the 1TB.

You're right. BonzaiDuck has posted so many different SSD combinations over the last few weeks in various threads, I got all turned around!

 

[UsandThem]

Remove the Intel AHCI driver and revert back to Microsoft AHCI, Intels driver is quirky.

Winner, winner chicken dinner here folks. It appears the standard Microsoft SATA AHCI driver outperforms the Intel chipset specific driver. Who'd ever thunk that?

 

[Charlie98]

Interesting...

 

[jhansman]

I'm not a BM'er, but my 500GB 850 EVO is by far the fastest SSD I've used on my system, which is by no means state of the art (Phenom II x6, Asrock 890FX Deluxe 5, 16 GB DDR3), but Win10 Pro just flies on it. And yes, I'm running the MS drivers (always have) as well with no issues. I don't OC, and am not a gamer, so my system doesn't get that much stress (Photoshop/Lightroom/audio processing), but it handles everything I throw at it with apparent ease.

 

[UsandThem]

I'm not a BM'er, but my 500GB 850 EVO is by far the fastest SSD I've used on my system, which is by no means state of the art (Phenom II x6, Asrock 890FX Deluxe 5, 16 GB DDR3), but Win10 Pro just flies on it. And yes, I'm running the MS drivers (always have) as well with no issues. I don't OC, and am not a gamer, so my system doesn't get that much stress (Photoshop/Lightroom/audio processing), but it handles everything I throw at it with apparent ease.

I don't care much about winning benchmarks either, but when a drive is only scoring 20k on random write IOPS, there is some type of issue. By just using the MS driver instead of the Intel one, it is now almost 88k. I just wanted the drive to perform up to its potential, and it simply amazes me how bad the Intel AHCI drivers are (at least in this case).