970 evo nvme running slower than it should.

[Sonikku]

I had used Userbenchmark a couple days ago after getting my new machine setup and it was boasting reads and writes in the 2700-2600 range. However, now after putting in the last parts in my build userbenchmark is indicating it is now "performing well below expecations" with a read of 904 and write of 864.

https://www.userbenchmark.com/UserRun/12695915

Now, you could argue that that's still "good enough". But it's still less than half of what it was at a couple days ago. Since then all I've done is update the drivers and firmware on Samsung's website and installed a new video card. What happened?

 

[Sonikku]

Well I restarted the machine and...



What in the actual...? That's even better than the bench I got initially! And now userbenchmark again.

https://www.userbenchmark.com/UserRun/12696762

2472 read, 2322 write.

It's all over the place. What the heck is going on here?

 

[mikeymikec]

The last time I read about benchmarking SSDs, it was suggested that they should be allowed to run idle for 30 minutes before running the test (presumably to get garbage collection out of the way). On Win10 I'd probably switch off background apps as well as the monitor being switched off and sleep mode in order to minimise background activity. I'd also run Task Manager on the performance tab to make sure the system isn't deciding to do a bit of background maintenance during the so-called idle time.

 

[Rifter]

My SSD, also a samsung nvme drive, has somewhat erratic results from windows, it must do a bunch of crap in the background with either caching or whatever else.

Linux results however are repeatable and accurate to what most reviews show. So im not to worried.

If this erratic results bothers you i would run some benchmarks in linux, if you dont dual boot just use a live USB image, and then if those results are matching up with reviews id say you are good to go.

 

[piasabird]

Is this an AMD Ryzen system we are talking about? Some of the AMD chipsets are not so great when it comes to M.2 devices. Heat can also interfere with M.2 performance. Overclocking which produces more heat can be a factor also. I watched a video done by Carey Holzman a YouTuber from Phoenix, AZ and computer tech of 30 years look at this issue and he concluded at the end that some newer AMD chipsets did not work well with M.2.

 

[Sonikku]

It's an Intel system and it's a non issue now.

 

[LTC8K6]

Looks like caching is fooling people?