Lower performance on my SSD

[Steelbom]

Hi everyone,

I've run a few benchmarks on my two Samsung 850 EVO's (250GB and 500GB) and have noticed that performance is far less than that of my friend's 250GB model.

My 250GB model is getting the following (1 run, 1 GiB) (read/write) Seq32T1 513/116, 4KiBQ8T8 353/61, 4KiBQ32T1 147/60, 4KiBQ1T1 23/63 whereas his drive is getting 540/517, 399/340, 269/234, 35/109. Both on Windows 10, same settings. Same firmware. Similar capacity filled up. My 500GB does much better, but even his 250GB is outperforming it in many of the write tests. Both are Sata 3 6Gbps, AHCI is enabled, trim is enabled, same firmware.

My 250GB has had 15TBs of data written to it whilst his has had 3.5TBs written. Could that be causing this? What can I do to investigate further? I'm not getting any errors and Samsung Magician says everything is A-OK.

 

[VirtualLarry]

My 250GB has had 15TBs of data written to it whilst his has had 3.5TBs written. Could that be causing this?

Internal fragmentation of the mapping tables? Certainly. But even so, your sequential write speeds seem below what I've read in benchmarks as the steady-state write speeds.

So, I do think something is wrong in your case.

What rigs do each of you have, and what Windows Power Plan are you set to? That can make a difference for SSD benchmarks.

I assume that the same benchmark program and version was used on both systems in question?

Attached is my Intel 545s 512GB SATA6G 2.5" SSD benchmark. Rig is a Ryzen R5 1600 w/32GB DDR4-3000, native SATA6G port, Win10 1809 64-bit Pro. Drive is basically new, just had Win10 installed on it, although it was a secondary drive for the benchmark run.

Also attached is my OS drive, an Adata SP550 240GB 2.5" SSD, which has had 10TB+ written to it. It's about half full right now. Benchmark looks great, not bad for an older 2D TLC drive at all. I'm personally very impressed by it, in everyday usage.

Edit: The device in Device Manager for my SATA controller, is called "Standard AHCI SATA Controller", FWIW. "Microsoft" is the provider. For some reason, I have two entries like that. I guess one is the on-die SoC SATA on the Ryzen CPU, and one is the B450 chipset SATA.

I do have four black SATA ports and two grey SATA ports on the board.

 

[UsandThem]

What driver are you using?

I had a similar issue a while back, and I finally figured out (with help here) that it was because I was using the Intel driver instead of the generic Windows one.

 

[Steelbom]

Internal fragmentation of the mapping tables? Certainly. But even so, your sequential write speeds seem below what I've read in benchmarks as the steady-state write speeds.

So, I do think something is wrong in your case.

What rigs do each of you have, and what Windows Power Plan are you set to? That can make a difference for SSD benchmarks.

I assume that the same benchmark program and version was used on both systems in question?

Attached is my Intel 545s 512GB SATA6G 2.5" SSD benchmark. Rig is a Ryzen R5 1600 w/32GB DDR4-3000, native SATA6G port, Win10 1809 64-bit Pro. Drive is basically new, just had Win10 installed on it, although it was a secondary drive for the benchmark run.

Also attached is my OS drive, an Adata SP550 240GB 2.5" SSD, which has had 10TB+ written to it. It's about half full right now. Benchmark looks great, not bad for an older 2D TLC drive at all. I'm personally very impressed by it, in everyday usage.

Edit: The device in Device Manager for my SATA controller, is called "Standard AHCI SATA Controller", FWIW. "Microsoft" is the provider. For some reason, I have two entries like that. I guess one is the on-die SoC SATA on the Ryzen CPU, and one is the B450 chipset SATA.

I do have four black SATA ports and two grey SATA ports on the board.

Thanks for your reply - both Intel systems - Windows 10, similar CPU models. Same benchmark and version used, correct.

I have enabled over provisioning on the smaller model (15%) -- I had issues setting this up and I wonder if this was part of the problem. Though his drive does not have anything but the default OP from factory. My results have improved a lot after doing this but are still a bit lower than his -- particularly in the write column - he is getting 517, 340, 234, 109. But perhaps that's normal for a newer drive?


I have attached a test of my 250GB and 500GB SSDs. The 250GB is doing much better now that over provisioning has been enabled... I may try to disable it again and see if it still works fine.

What driver are you using?

I had a similar issue a while back, and I finally figured out (with help here) that it was because I was using the Intel driver instead of the generic Windows one.

View attachment 5556

Thanks for your reply - I checked and am using the Microsoft drivers for both.

 

[VirtualLarry]

Those benchmark pics that you posted, look totally to me, at first glance, within the normal range of benchmarks for modern SSDs.

I think that you're fine, now.

Maybe you just needed to run some TRIM passes, and let the drive internally de-frag for a while.

 

[Steelbom]

Ah great, I thought these results seemed okay. But yeah before I added the OP to the 250GB model it was doing very poorly.

Adding OP to the 500GB model also improved performance slightly. I have about 34GB less space on my 250GB model now though which is the OS drive, so I'm going to have to try and cull some things.

 

[VirtualLarry]

I have about 34GB less space on my 250GB model now though which is the OS drive, so I'm going to have to try and cull some things.

So, it is / was close to full? That can definitely affect write speeds, in the negative sense.

My friend's old 30GB (yeah, that used to make SSDs that small) SATA SSD with Win7 64-bit (can you believe that it used to fit on an SSD that small? Updates!), got fairly full after some years worth of updates, so we finally had to upgrade him to a 120GB SSD, which is finally, a few years later, getting fairly full again. His older 30GB SSD started "stuttering", and affected watching online videos. (Did you know that streaming-video software, caches to the HDD/SSD, instead of JUST to the RAM? At least, the software back then did. He has 16GB of RAM now.)

 

[VirtualLarry]

I have about 34GB less space on my 250GB model now though

Is that directly from Over-provisioning? It over-provisioned 34GB of space, on a 250GB capacity drive? That sounds rather excessive. I wouldn't make the over-provisioning on a drive that size, more than about 10GB, if that. Maybe even 5GB. Unless you have a really write-heavy workload, like a database or e-mail server, or a surveillance station or something.

 

[Steelbom]

So, it is / was close to full? That can definitely affect write speeds, in the negative sense.

My friend's old 30GB (yeah, that used to make SSDs that small) SATA SSD with Win7 64-bit (can you believe that it used to fit on an SSD that small? Updates!), got fairly full after some years worth of updates, so we finally had to upgrade him to a 120GB SSD, which is finally, a few years later, getting fairly full again. His older 30GB SSD started "stuttering", and affected watching online videos. (Did you know that streaming-video software, caches to the HDD/SSD, instead of JUST to the RAM? At least, the software back then did. He has 16GB of RAM now.)

Yeah crazy... things have improved a lot! Prior to the OP I was at 79% capacity, after culling some things and doing an OP of 15% (34GB) I am now at 80% capacity.

Is that directly from Over-provisioning? It over-provisioned 34GB of space, on a 250GB capacity drive? That sounds rather excessive. I wouldn't make the over-provisioning on a drive that size, more than about 10GB, if that. Maybe even 5GB. Unless you have a really write-heavy workload, like a database or e-mail server, or a surveillance station or something.

I chose an amount of 15% after seeing some benchmarks showing performance consistency with different levels of provisioning... however it may be completely unnecessary in my case. I wouldn't say I have a particularly write heavy work load.

Do you think it's worth lowering it to 5%? (I added a 15% OP to my 500GB one as well)

 

[VirtualLarry]

If you are a "Pro Gamer", and don't want any unwanted stuttering from your storage, then you could probably keep the 15% O.P. But if you are short on storage space, and only want to be able to see decent sequential write rates, 10% or possibly even 5% might be enough. I mean, if you don't absolutely need the extra space, you can keep the 15% O.P. It's not really hurting anything, and may be helping your write speeds to a greater extent than a 5-10% O.P. might be. I guess, read some reviews, I've never felt the need to manually O.P. space on a modern SSD, although I like to keep roughly 25% free at most times.

 

[Steelbom]

Icic, thanks for your help!