topic Re: A question about the Intel NVMe SSD P3700 thermal throttling in Solid State Drives (NAND)

A question about the Intel NVMe SSD P3700 thermal throttling

idata — Mon, 26 Sep 2016 14:49:30 GMT

Hi,I have a question about the Intel NVMe SSD P3700 series specification document.

In page number 26, what is the exact meaning of descriptions of Byte 0 and Byte 1-4? (please see the picture below)

Is integer percentage in Byte 0 means the ratio of slow down due to the thermal or something?

What is throttling event count in Bytes 1-4 mean?

I need more clear descriptions for them.

Thanks for your kind support in advance.

Best regards,

Jaemin

Re: A question about the Intel NVMe SSD P3700 thermal throttling

idata — Mon, 26 Sep 2016 21:56:01 GMT

Hello Jaeminyx,

Thanks for posting in our forum.We would like to work on your request and give you a proper response about this SMART attribute, we will check with one of the disk here and will provide screenshots if possible.Please stand by for an update.Regards,NC

Re: A question about the Intel NVMe SSD P3700 thermal throttling

idata — Thu, 29 Sep 2016 19:38:15 GMT

Hello Jaeminyx,

According to the bytes meaning, could you please send us a log of the SMART details first? You may need to download the https://downloadcenter.intel.com/download/26221/Intel-SSD-Data-Center-Tool Intel® SSD Data Center Tool, here is the http://www.intel.com/content/dam/support/us/en/documents/solid-state-drives/ssd-software/Intel_SSD_DCT_3_0_x_User_Guide.pdf User's Guide.The information about these values is being investigated, what we can tell you is that for bytes 1-4 there is a signal that indicates when the drives passes the threshold. We will keep you updated.Regards,NC

Re: A question about the Intel NVMe SSD P3700 thermal throttling

idata — Tue, 04 Oct 2016 08:36:17 GMT

Here I attached a results from isdct command ("isdct show -smart" on ubuntu machine):

https://drive.google.com/file/d/0BxpwwgcCWPrqR0dvTnVHTE00SVJMdV9VZFFrNVgzY2JBZzhv/view?usp=sharing 0930.tar.gz - Google Drive

You can find two log, one for a whole log file (isdct_whole), one for a log only contains information related to a thermal throttling.

And something I really want to know is related to this graph:

https://www.google.com/url?hl=ko&q=https://docs.google.com/spreadsheets/d/1i6GDeQ1AaT8ZsYELePzacn_AZxIQZuAzFXus-nxmrmc/pubchart?oid%3D895791124%26format%3Dinteractive&source=gmail&ust=1475655174849000&usg=AFQjCNGAgpmYt9bGyMuzLzkgSdxPuKhNGw https://www.google.com/url?hl=ko&q=https://docs.google.com/spreadsheets/d/1i6GDeQ1AaT8ZsYELePzacn_AZxIQZuAzFXus-nxmrmc/p…

I ran some Fio benchmark (bs 4k/qd 16) to see an effect of high temperature on NVMe device (I removed heatsink on the device).

As you can see in the graph, As the temperature goes up, the thermal throttling percentage is also increased. However, the value of kiops seems like same as a result with moderate temperature.

So, I want to know how P3700's thermal throttling works? If it does something to cooling down when a temperature goes up (seems like it does something according to the spec), why the throughput is not affected?

Any information will be helpful for me.

Best regards,

Jaemin

Re: A question about the Intel NVMe SSD P3700 thermal throttling

idata — Tue, 04 Oct 2016 21:01:46 GMT

Hello Jaeminyx,

Thanks for all the information provided and questions, we will be working on all this and we will get back to you, please expect an update this week.Regards,NC

Re: A question about the Intel NVMe SSD P3700 thermal throttling

idata — Thu, 06 Oct 2016 14:01:21 GMT

Hello Jaeminyx,

After researching this situation we can tell you that when Thermal Throttle is "activated", the performance will drop to a certain level in order to mitigate any thermal challenges.Now, according to the graph, the drive initially operates at 75-80K (IOPs) at a temperature of ~50-60 Celsius, but after a certain period of time operating under that characteristics, it starts heating and that's when the Thermal Throttle "activates" until the IOPs start going down until the get below the 75K... You can actually see a drop in temperature kind of aligned with the drop in performance until both stabilizes (IOPs) slightly below 75K and Temperature at ~60C.Could you please let us know the form factor of the SSD you have? Can you send a picture of your drive?Regards,NC

Re: A question about the Intel NVMe SSD P3700 thermal throttling

idata — Fri, 07 Oct 2016 13:23:21 GMT

Here are pictures of my drive:

As I mentioned before... I removed the heatsink of my drive. And the form factor of my SSD is AIC.

So, "the drop in performance" you said means the interval after 10 minute in the graph? Then, what is the meaning of the throttling percentage? I can't found any relationship between the value of this attribute and the performance shown in the graph.

And, I attached the other experiment result I did:

https://mail.google.com/mail/u/0/?ui=2&ik=225ef0d506&view=fimg&th=1579e2a502bd56a1&attid=0.1&disp=emb&realattid=ii_1579e29352c2bce5&attbid=ANGjdJ92_rbEpQRLFR6YjpChy3ReG-4YV6sj-wBYjUl40RkX7m11PH5uXxqm0aVpA8pwV94bQOj38zccGcSD8fyrtbebtL6OcIUTK6-ZlylTTyzHEgH_WmuPhqwRoMc&sz=w1128-h698&ats=1475846229065&rm=1579e2a502bd56a1&zw&atsh=1

In this time, I ran some sequential write benchmark (also Fio, bs=64K, qd=64). As you can see in the graph, initially temperature was goes up fast, and it slowly goes down after around 7~8 minute. But, iops seems like stable all the time.

Why it is happen? I really want to know about the mechanism of thermal throttling in my drive.

Could you give me any information?

Thanks for your kind support again in advance.

Best Regards,

Jaemin

Re: A question about the Intel NVMe SSD P3700 thermal throttling

idata — Sun, 09 Oct 2016 04:49:19 GMT

Why it is happen? I really want to know about the mechanism of thermal throttling in my drive.

How can you expect Intel to explain the results of your "testing", when you have taken the SSD apart, and it can no longer work as it was designed to work?

How do you know the thermal throttling mechanism will still work as it was designed, when you have removed the heat sink?

The throttling mechanism is based upon this SSD being fully assembled, and used in the manner described in the datasheet. The engineers that designed it would not be able to take you seriously.

What was the air temperature and LFM speed of the air moving over the (remains) of the SSD during your testing? That is one of the specifications Intel uses in providing the normal operation specifications of this SSD.

You removed the heat sink, but you want to know how that affects the operation of the thermal throttling mechanism, that is designed with and meant to be used with the heat sink you removed?

I would be embarrassed to ask.

Re: A question about the Intel NVMe SSD P3700 thermal throttling

idata — Mon, 10 Oct 2016 14:05:24 GMT

Hello Jaeminyx,

It is really hard to give you any explanation about this when you have removed the heatsink from the SSD, this is not going to operate as it should if you remove parts from the disk.To help you further, we would like to bring this to our engineering department and check what they have to say about it, however, keep in mind they may not provide support to this drive.We will keep you posted.Regards,NC

Re: A question about the Intel NVMe SSD P3700 thermal throttling

idata — Fri, 14 Oct 2016 20:19:06 GMT

Hi Jaeminyx,

Our engineering department reviewed the thread and we need to let you know that since the drive was altered or modified the behavior may vary from the actual specs.Regards,NC

Re: A question about the Intel NVMe SSD P3700 thermal throttling

idata — Fri, 21 Oct 2016 20:31:09 GMT

Hello Jaeminyx,

We would like to know if there is any other question you may have or something else we can assist you with?Regards,NC