SSD - Crucial MX500 vs Samsung 860 EVO - SATA

[grass on grass]

Hello,

I am considering two different 2TB drives for an older computer.  I am not replaceing this computer at the moment. I currently have a 1TB samsung SSD and want to upgrade it to 2TB (both SATA).    The crucial MX500 is quite a bit cheaper than the Samsung 860 EVO so i am considering that option.  when checking UserBenchmark.com, some of the metrics weight heavily in Samsung's favour.  The issue is that i don't really understand which metrics they are employing apply to my specific workflow.

 

I do not want a loss in performance that will slow down my workflow noticeably.  My main/primary concern is, when i open a new program or start new web searches, i don't like waiting for the computer to keep up with my requests.  Sometimes i will do several of these requests at once.   I dont transfer alot of files making sequential transfer times not too important for me.  I don't care about boot times.

 

My usage:

1. 100% work, 0% gaming for the foreseeable future.
2. many more applications open at the same time that the average persion 100% of the time using a significant amount of RAM.  excel, word, picture viewer, adobe reader with large files usage heavy, some other applications, video editing sometimes, but not often.  THere are many others that i have not listed.
3. heavy web browsing with many many many tabs open at the same time
4. I think my queue depth is larger than most people, but i am not certain.

 

Can someone clue me into which metrics i should pay most attention to for this type of workflow? and/or any other relavent information that will help along with the buying decision. 

Yes i know the best option is really to upgrade the whole rig, but the timing is not right.

Thanks

-GonG

[Electronics Wizardy]

Don't use userbenchmark, its not very good, there are much better reviews.

 

They both preform basically the same in almost all workloads.

 

9 minutes ago, grass on grass said:

I think my queue depth is larger than most people, but i am not certain.

What queue depth do you see now? Look in resource monitor.

 

Id get the cheaper mx500, there basically the same.

 

If you really want faster, get a nvme drive, there not much more money, and are a lot better at many things, and have lower cpu overhead aswell.

[grass on grass]

2 hours ago, Electronics Wizardy said:

Don't use userbenchmark, its not very good, there are much better reviews.

 

They both preform basically the same in almost all workloads.

 

What queue depth do you see now? Look in resource monitor.

 

Id get the cheaper mx500, there basically the same.

 

If you really want faster, get a nvme drive, there not much more money, and are a lot better at many things, and have lower cpu overhead aswell.

The userbenchmark was from user review section which has thousands of samples for each option, not the reviews from editors (which i didn't even look at).  It is true they are similar, but i was wondering fo rmy specific type of workload.  the userbenchmark has many different types of metrics

[Electronics Wizardy]

1 hour ago, grass on grass said:

The userbenchmark was from user review section which has thousands of samples for each option, not the reviews from editors (which i didn't even look at).  It is true they are similar, but i was wondering fo rmy specific type of workload.  the userbenchmark has many different types of metrics

There is a lot of bias at userbenchmark, just ignore them completly and look at better reviews.

 

There basically all sata limited for most uses, so it really won't matter performance wise, If you want a fast drive, don't get a sata drive, get a nvme drive.

 

What is your normal io usage on your system when working? What queue depth? Iops? My guess is its not too io heavy, so it won't matter.

 

Look at this review,https://www.tomshardware.com/reviews/samsung-860-evo-ssd-review,5446-2.html, the performance is so close it just won't matter. Here is anouther review, basically the same result https://www.anandtech.com/show/12408/the-samsung-860-evo-m2-2tb-ssd-review/3

[NewMaxx]

There's just a couple of basic parameters: controller, DRAM, flash, and software (firmware and SLC cache design). Both drives have DRAM and comparable 64L 3D TLC, the differences are with the controller and the software. For the controller: the 860 EVO's MJX is tri-core with specialized cores (read, write, host) while the MX500's SM2258 is single-core. The latter is more efficient but can struggle with heavier workloads, e.g. when fuller. For SLC: the 860 EVO has a hybrid cache, that is part static and part dynamic, while the MX500's is purely dynamic. Dynamic is more flexible and better for consumer/general workloads, static is better for steady state. However, the interface (SATA) and protocol (AHCI) are quite limiting so in most cases these drives will perform very similarly in subjective ("real world") terms. This is especially true at higher capacities where you have plenty of dies to saturate the controller, generally 32 dies (1TB) is sufficient.

 

You shouldn't use UserBenchmark for a number of reasons, however for SSDs it's just showing synthetic results which heavily lean towards SLC and ideal (e.g. not full-drive) performance. UB can be useful for demonstrating differences, for example under "Nice To Haves" you'll see the 860 EVO stands out at DQ/ADQ Mixed 4K-64Thread Mixed IO - this is because of the more powerful controller as mentioned above. But UB doesn't tell you anything about full-drive or steady state (TLC) performance. Generally I would expect the 860 EVO to do better there thanks to some static SLC and the controller, although the MX500 can be more efficient otherwise. Also, at 2TB it's really less of a concern with SATA. So if the MX500 is a lot cheaper it's a pretty easy decision.

 

When reading you're basically going by TLC read latency (~75µs) either way, these drives have the same page size (16KiB), both have partial read (e.g. <=4KiB) capability, the MX500 has data-at-rest power-loss protection (differential module), both use LDPC, RAID parity, etc. That is to say the technology is very similar and you probably won't be putting the drives into a position where their differences are noticeable. As long as you have DRAM and comparable flash they will perform similarly. Their largest gaps will be with writes and mixed I/O due to the controller and SLC differences but you're unlikely to push them hard enough to expose that.

[grass on grass]

39 minutes ago, NewMaxx said:

For the controller: the 860 EVO's MJX is tri-core with specialized cores (read, write, host) while the MX500's SM2258 is single-core. The latter is more efficient but can struggle with heavier workloads, e.g. when fuller.

are heavier workloads defined only when the drive becomes more full or is there another definition for heavy workloads, such as running many processess at once?

[NewMaxx]

5 minutes ago, grass on grass said:

are heavier workloads defined only when the drive becomes more full or is there another definition for heavy workloads, such as running many processess at once?

Well as I pointed out on UB's results, with 64 threads (heavier workload) with mixed I/O (read + write simultaneously) the MJX is clearly faster. DQ/ADQ means "depth queue" so this means high queue depth. The vast majority (95-98%) of client workloads (and that includes workstations even) are at QD4 or lower with most (>50%) being QD1. You're not going to get anywhere near 64 threads or QD. I mention the full-drive state because with dynamic SLC you're converting to/from TLC, emptying/folding SLC, and dealing with background maintenance (garbage collection) which can increase the amount of load on the controller with mixed I/O, but it's still not going to be anywhere near the UB results. But it is more visible in a full drive state where you push the SLC cache and controller, but even then it's like 5-10% difference in bandwidth/latency between the two drives. It's difficult to notice tens of microseconds for most people. That's a simplification of course. I've found a SATA SSD is plenty capable of hosting a half dozen or more VMs without a problem, "heavy" certainly means prosumer at the minimum.

[SurvivorNVL]

Just going to throw this out there; I have had no issue with longevity with Crucial, Adata, or Western Digital SSD's; NVME or SATA.  I have had, on the other hand, Samsung SSD's fail on me multiple times in the last four years.

[GDRRiley]

the MX is the better buy but the 860 is a faster drive for some workloads like high res video recording

[grass on grass]

well,

i did it.  went with the crucial MX.  thanks for everyone's input!

GonG