Question Real world SSD performances listings

[cool_quiet_computing]

Hi,
I am wishing to add a fresh SSD, for a specific project, to my Ryzen 7950x based PC, to replace my current project SSD, a 980 Pro 500GB.

Not willing to be seduced by promises of 14GBps in sequential tests, I am interested more on random and real world test results. I found Tom's list of SSD hierarchies very informative :

https://www.tomshardware.com/features/ssd-benchmarks-hierarchy

Looking at that list, the Corsair MP600 Elite 2TB is faster than the T705, both having 25% higher IOPS than the 980 Pro. I guess that translates as being 25% faster.

I am wondering if the T705, or the Sabrent Rocket 5, which is not listed yet, will have the best improvement. I don't really understand if IOPS is the best indicator, or whether I should look at online ATTO tests to decide the better performance.

My specific project generates Excel files of ~60MB. I was examining the online ATTO 64MB results ( at the bottom of the ATTO table ) and was impressed at the Sabrent Rocket 5 speeds. Read/writes appear to be 13GBps / 11.79GBps. I don't have a big grasp of the testing procedures and am not well versed in ATTO testing,

In my current setup, 60 MB files take about 2-3 seconds to open / save using the Samsung 980 Pro and ideally I would like a new SSD to reduce this to 1 second (Not looking for 13GBps performance, I will be happy with effectively 60MBps.) If there's not at least a 3x performance increase, it's not worth the investment.

Is that how you as an expert would select a drive?

I have considered using a second Asus ROG SQ7 1TB but random test results have not been published.

(my previous build, a 12 year old Phenom-II based CPU, was taking 75 s to open / save the same files and so I selected the 7950X CPU for the new system, based on the L2 + L3 cache being larger than the file size )

your thoughts would be appreciated

Thank you

 

[USAFRet]

My specific project generates Excel files of ~60MB. I was examining the online ATTO 64MB results ( at the bottom of the ATTO table ) and was impressed at the Sabrent Rocket 5 speeds. Read/writes appear to be 13GBps / 11.79GBps. I don't have a big grasp of the testing procedures and am not well versed in ATTO testing,

In my current setup, 60 MB files take about 2-3 seconds to open / save using the Samsung 980 Pro and ideally I would like a new SSD to reduce this to 1 second (Not looking for 13GBps performance, I will be happy with effectively 60MBps.) If there's not at least a 3x performance increase, it's not worth the investment.

A lot of that is RAM and CPU. Or the software involved. Excel and/or whatever else is running at the time.
Not the drive.

Your current 980 Pro is near the top of the performance ladder.
Swapping in a PCIe 5.0 drive, and keeping everything else exactly the same, you'd see little if any difference.

 

[cool_quiet_computing]

Thank you USAFRet

i had seen for example 1,400,000 recently quoted as Read / Write IOPS on the Sabrent 5 and am sure the IOPS of the 980 Pro is about 500,000.

So many zeros, so little real world relevance.

I no longer try to judge SSD speeds by any specific number, although I do try to look at the ratios between them to estimate relative performance improvement

 

[USAFRet]

Thank you USAFRet

i had seen for example 1,400,000 recently quoted as Read / Write IOPS on the Sabrent 5 and am sure the IOPS of the 980 Pro is about 500,000.

So many zeros, so little real world relevance.

I no longer try to judge SSD speeds by any specific number, although I do try to look at the ratios between them to estimate relative performance improvement

Going from HDD to SSD of any flavor, huge increase.

Between the different flavors of SSD? Not so much.

Especially if going between NVMe drives.

With the exact same parts and software, and changing from a 980 Pro to "something faster"....you'd need a stopwatch to tell a real world user facing difference.
Maybe not even then.

 

[hotaru.hino]

i had seen for example 1,400,000 recently quoted as Read / Write IOPS on the Sabrent 5 and am sure the IOPS of the 980 Pro is about 500,000.

So many zeros, so little real world relevance.

Perhaps not to the consumer, but it does matter for those that run database servers, or anything else where tons of requests are made for small amounts of data.

 

[cool_quiet_computing]

Perhaps not to the consumer, but it does matter for those that run database servers, or anything else where tons of requests are made for small amounts of data.

Hi Hotaru Hino,

Does the heat generated by Gen 5 SSDs mean they are not suitable for servers ?

 

[cool_quiet_computing]

Going from HDD to SSD of any flavor, huge increase.

Between the different flavors of SSD? Not so much.

Especially if going between NVMe drives.

With the exact same parts and software, and changing from a 980 Pro to "something faster"....you'd need a stopwatch to tell a real world user facing difference.
Maybe not even then.

your picture reminds me - i need to get a backup disk too

 

[USAFRet]

your picture reminds me - i need to get a backup disk too

https://www.worldbackupday.com/en

 

[hotaru.hino]

Hi Hotaru Hino,

Does the heat generated by Gen 5 SSDs mean they are not suitable for servers ?

Heat has never been a problem for servers because noise is not a factor as most server racks are in an isolated room. So they have fans that blow stupid amounts of air at all times, in addition to whatever air conditioning is being pumped into the room.

 

[USAFRet]

Does the heat generated by Gen 5 SSDs mean they are not suitable for servers ?

"server room"

Ambient temp much below normal room temp.
Or trying to be.

Same as high end CPUs, etc.

Build the environment around the hardware needs.
Ignore the noise levels, because people don't sit in there all day (usually).

 

[geofelt]

Do not be much swayed by unrealistic vendor synthetic SSD benchmarks.
They are done with apps that push the SSD to it's maximum using queue lengths of 30 or so. They are done on new/clean drives for repeatability.
Most desktop users will do one or two things at a time, so they will see queue lengths of one or two.
What really counts is the response times, particularly for small random I/O. That is what the os does mostly.
For that, the response times of current SSD's are remarkably similar. And quick.

 

[cool_quiet_computing]

HOW TO AWARD BEST ANSWERS: CLICK HERE!!

been trying to find that Tick symbol as per your instructions but no luck yet ... i will give it another day or so....

 

[USAFRet]

been trying to find that Tick symbol as per your instructions but no luck yet ... i will give it another day or so....

Try it now...😉
(changed from Discussion to Question)

 

[Maxxify]

I am wondering if the T705, or the Sabrent Rocket 5, which is not listed yet, will have the best improvement. I don't really understand if IOPS is the best indicator, or whether I should look at online ATTO tests to decide the better performance.

These two drives have the same hardware, so will end up close in reality. The Rocket 5 is optimized a bit differently from what I understand. In any case, I'm not sure IOPS is what you'd want. If random reads are your goal, especially 4KB, then nothing beats the OG Samsung 990 PRO, but they've switched to newer flash that isn't quite as fast as of Q3/Q4 last year. At this point, something with 2400 MT/s Micron TLC is probably the closest you can get - that would be the Crucial T500, T705, Rocket 5, etc. The T500 still has DRAM and is very efficient, but it has weaker long tail performance than the T705 and Rocket 5. The Rocket 5 looks to have the best sustained write performance. But it's possible Hynix's upcoming flash, on the Platinum P51, will be faster yet, and that drive has very high IOPS as well. (announcements of the drive recently omitted/lacked IOPS, but we've known the PCB01 aka Platinum P51 to be 2M/2M for a while now)

 

[DJMCM]

Do you know SSD-Tester? https://ssd-tester.de/m2_ssd_test.php
Here you can also find real test results from practice.