Who has the fastest…SSD controller?
But what about thermal throttling...? Who cares as long as the benchmark looks good...

Who has the fastest…SSD controller? But what about thermal throttling...? Who cares as long as the benchmark looks good...

Phison, Silicon Motion and Samsung are battling themselves in the fight about having the fastest PCIe Gen. 4x4. We see a lot of astonishing performance figures according to CD test benchmarks. 7000MB/s are exceeded already…see the link.

Anyhow, many customers are facing issues here. In expectation of reaching these specified performances, systems will be designed on that. However, when making prototyping tests, the reality looks completely different.

The performance will be only reached fresh out of the box and for short time. After short period the performance is rapidly crashing down.

Why?

1.   When the SSD is new, all NAND blocks are empty. The controller can write them directly without clearing at first stage. However, as soon as NAND blocks are filled once, the controller needs to start cleaning data fragments, the so-called Garbage collection. This is reducing the performance at first.

2.   By keeping the controller on high speed, all channels are stressed, thus it generates heat! If the heat is exceeding a certain security threshold, it starts to shut off channels to reduce workload in order to lower the temperature. The so-called thermal throttling. Performance needs to be reduced otherwise the controller dies due to high heat.

You can see an example of thermal throttling, which was tested by "techpowerup.com" on a Gen3x4 M.2 SSD.

https://www.techpowerup.com/review/crucial-p1-nvme-m-2-ssd-1-tb/7.html

Es wurde kein Alt-Text für dieses Bild angegeben.

All in all, these features are necessary, but mean that the peak performance is not a realistic sustained performance. For users in the industrial segment with passive cooling systems etc. thermal throttling needs to be considered and the choice of the right SSD should not be done according to “best case” datasheet figures.

By the way, even the linked report says: “…the CrystalDiskMark 7.0.0 bench tool only demonstrates the peak performance value. So it is not an indicative of performance in real-world applications.”

 

Daniel Menzel

Menzel IT GmbH | Hochleistungsinfrastrukturen

4 年

Sch?ner, kurzer Fachartikel. Gef?llt mir.

Sehr sch?n Ricky. Guter Post.

要查看或添加评论,请登录

Ricky Gremmelmaier的更多文章

  • Meet me at the D&E event 2017!

    Meet me at the D&E event 2017!

    You can find me at the D&E Event in Belgium and Netherlands next week. If you want to learn more about different NAND…

    1 条评论

社区洞察

其他会员也浏览了