The sm961 vs the Samsung 960 pro

sram

[H]ard|Gawd
Joined
Jul 30, 2007
Messages
1,699
Are they essentially the exact same ssd disk? One oem and one retail? Or are there design differences?

I bought the sm961 and i'm benchmarking it now to make sure I have the right numbers. I have the 512 GB version. If the 960 pro is different or bettet/faster then I'll need to think this through. I'm also planning to build a RAID 0 array( for the heck of it and for bragging rights).....can I mix them? A 960 pro and sm961 in RAID 0 ?
 
Anybody?

To elaborate: When I first found out that the 960 pro is out, I immediately decided I'll get it for my next ultimate build (For bragging rights). I went to ebay looking for one for good price and I bought it. It ended up being the sm961. I just finished my build and I need to know what to really compare my benchmarks to. I didn't know at the beginning there is an OEM version and a retail version. Even if I knew, I would have thought they would be the same exact drive, which doesn't seem to be the case !!!

What numbers should I be expecting?

And what is the best driver to use? Should I use the magician software?

This will sound stupid but I'm willing to get the 960 pro the retail version just to say I have the best if the sm961 is really under-performing the 960 pro retail. I will post my results in few minutes but I know there are less than people get....there is some tweaking that I need to do.
 
I use ATTO disk benchmark v3.05.
https://www.atto.com/disk-benchmark/
I went with a pair of the retail 960 EVOs in RAID 0. Performance is phenomenal. I know this is not your configuration, but you did mention an interest in RAID. I think you'll be able to see some numbers for comparison here:
https://www.vortez.net/articles_pages/samsung_960_pro_raid_review,7.html

FWIW, my 960 evos in RAID 0 are returning peak read speeds in the 3200Mb/sec range, and peak write speeds in the 2800Mb/sec range.

If you're on the fence at all, ask yourself if you would in fact like to get to the Windows 10 desktop in ~8 seconds. ;)
 
Here are the numbers


p6rIayr.jpg
 
Do they look right to you? They are different from what I have been seeing.

And I don't know about this as ssd error I'm getting now. I didn't get it before when first benching. There is one thing I should say, the results are not consistent. They change every time I run the tests specially with HD Tune. What gives?
 
I don't know specific numbers - maybe check that link I gave earlier to compare your results to theirs.

I do know that I was only able to find one benchmark that gave consistent results with mine in RAID 0, which was Atto.

At the moment mine may be down. I'm having trouble with posting past the IDE initiation process... I fear I may have a dead drive after ~3 months.
 
I don't know specific numbers - maybe check that link I gave earlier to compare your results to theirs.

I do know that I was only able to find one benchmark that gave consistent results with mine in RAID 0, which was Atto.

At the moment mine may be down. I'm having trouble with posting past the IDE initiation process... I fear I may have a dead drive after ~3 months.

I'm really sorry to hear this about your drive. A hardware going bad is the worst that can happen to a computer enthusiastic.


I don't know but it is shown to you in this tab for your sm961, but mine shows a driver with a 2006 date. Can this be right!!!? I feel that's not right.

3hgYpz7.jpg
 
An update: I changed the power mode to "High performance" and things are looking better now. I think I'm fine. Results are consistent.
 
Toggle "Turn off Windows write-cache buffer flushing on device" - Meaning that it is checked, and not an empty box. It is in the policy tab of that very screen shot.

Run the benchmarks again to see what you get after that. I use the Samsung driver and max out. However, Samsung's Magician doesn't see the 960 with the 2.2 driver. With Window's NVMe driver SMART attributes can be read, temperatures, genuine verification is proper. Performance with both drivers was very similar. I also have every power and c-state enable in the EFI and get the benches, but do have on Windows balanced power setting. My 6700K goes to C7. It is bullshit, IMO, about power states affecting performance. Never had the issue with C-state, P-state, and LSPM ( enabled in EFI @ Windows moderate) but the BIOS/EFI was an area I can configure with much knowledge. I knew this was going to be fast but it actually goes beyond a bit of my expectation, and boot after post, flash for barely a blip and its desktop.

Edit: Hmn, Magician now shows the drive. Anyways, this is what I get with Magicians' benchmark. Note, Overclock's forums in the background for the 960 Pro (http://www.overclock.net/t/1617171/samsung-960-pro-evo-owners-club). Seems the image link isn't working so here: http://imgur.com/a/E1xXX
 
Last edited by a moderator:
I'm on duty at work now. When I get home I will post images of my benchmark after changing the power state. I remember my numbers were even better than some reviews!
 
Cool, I am curious if the cache setting is better. C-states and P-states vary from EFI and Windows installations too. If possible do, cache and then cache with power settings. Power settings shouldn't do much at all but YMMV
 
Sorry for being late. For some reason the issue with AS SSD came back even though the power plan is still high. Here are the numbers. They are more consistent now.

6shLFDN.jpg



They go down a little when I play with write-caching, so I kept that at default.
 
OOOO, just saw something 0_o.

You are on the 2700K? So, you are using PCIe 2.0 and then going through DMI 2.0? Or is this a different core logic that this is installed in? Because you are basically maxed out and will not get the output of newer architecture-If you had HEDT or Ryzen and routed to the processer, or the Z100 and higher core logics the number would be different. Also the PMC is really good at regulating P-States and C-states for the newer processor and preventing I/O from being demoted, but never noticed differences when I configured systems for high C-states.

Were these number with cache off?
 
No this is my new kaby lake build. I just didn't modify my signature yet. I'll make another thread for that.

Enabling/disabling write cashing doesn't really make a difference.
 
960 Pro should be better, without doubt, but a 960 would be better in performance with a HEDT or Ryzen processor due to the routed PCIe to processor. I/O from the PCH and through the DMI would not be that optimal for a 960 pro, IMO. Although, you can see in the overclock forums there are many that have low numbers like you high end and desktop. Your first numbers are very similar to what I saw in their forums. I always read experiences of others and compare. I can say that nothing of what is said there has been the experience I had also; and this includes here. So, the question is to return or not to return, I guess? Or continue to hopefully find if resolution is within the system configuration?

You didn't happen to clone a target drive to the source 960? Basically this was a fresh install of Windows with the installer creating the partitions and formatting?
BTW, there is 2.2 that was released not too long ago. However, I do not think it will resolve your issue: http://www.samsung.com/semiconductor/minisite/ssd/download/tools.html

I do everything that on many forums have posted not to do. From prefetch to power settings, and as you can see the numbers with my 960 EVO are nominal. So, it is either the NVMe itself, EFI, or/and system configuration-drivers and such-after Windows install. Matter of fact, can you do a screen shot of Device Manager's > System Devices? Sometimes not having the SMB driver installed can cause low numbers. There was even a recent chipset software installation build that had an issue not installing the driver and people were complaining of performance along with other issues. Then the new build that came out had a build date of 1/1/1970. Fucking Intel guys are smart but some recent issues with their drivers (like non compliant Windows driver for network cards with the latter version of Windows 10), a bit small in mistake size, but a multi-billion industry couldn't see these mistakes when us $5 in the bank account dudes can spot in in two seconds...hmmmm

Screen shot like this: http://imgur.com/a/DLTwC
 
Last edited by a moderator:
I don't know if this was mentioned elsewhere in the thread, but I thought I would share it. I once read that disabling C-states in the BIOS also helps with getting maximum PCIe SSD scores.
 
Back
Top