[GTALUG] NUC NUC NUC

[D. Hugh Redelmeier]

| From: Russell Reiter via talk <talk at gtalug.org>

| On Mon, May 20, 2019 at 11:42 AM D. Hugh Redelmeier via talk <
| talk at gtalug.org> wrote:

| > Why do you think that this was heat-related?  It might be, but that
| > would not be my first guess.  (I am not an expert on this.)
| 
| Apparently, at least on Crucial products, there is built in thermal
| monitoring which will throttle speeds.
| 
| Here's a link to the 1TB Nvme. On sale for the next few days for $147.00,
| about $40.00 less than I paid.
| 
| https://m.newegg.ca/crucial-p1-1tb/p/N82E16820156199?item=N82E16820156199&m_ver=1
| 
| Here's a link to a review of Crucial's Thermal Throttling capacities

Wow, prices sure have dropped.  (I wish I had more sockets that would
take NVMe.)

| https://www.techpowerup.com/reviews/Crucial/P1_NVMe_M.2_SSD_1_TB/7.html

Thanks.  Quite interesting.  I hope your heat sink helps.

Looks like a good choice for a drive.

Something I didn't know:

=>	Thermal throttling is an issue for nearly all M.2 NVMe SSDs,
	and the Crucial P1 is no exception. Not cooled and fully
	loaded, it will heat up quickly and start throttling after a
	bit more than a minute at full load. Now, don't get scared. In
	that time, the drive processes almost 100 GB of data. Again,
=>	highly unlikely in a consumer scenario. Still, I would have
	wished for a higher temperature limit and a more graceful drop
	in performance during thermal throttle. Samsung, for example,
	has implemented that very well.

The problem I was thinking of is shown in "write intensive usage":

<https://www.techpowerup.com/reviews/Crucial/P1_NVMe_M.2_SSD_1_TB/6.html>

But they suggest it hits after 140GB of writes, not 100G you reported:

| > | the SSD. Copying a 100gb image to the 1TB drive really hit performance
| > tho

You can see a very large performance cliff at 140GB.

So either what you are observing is not this cliff, or you are doing
something else (reading the image from the drive?) to move the cliff
earlier.

The heat cliff should be much much sooner, I think.

Their diagram showed the thermal cliff happened (without any fan)
after about 60 seconds of 160MB/s writing.  That would be about 96GB.
Funny that you'd notice that with a 100GB image since only the last
4GB should be slow.  That should only take another 7 or 8 seconds
(instead of 2.5 seconds).  That isn't something I'd notice.

Copying a 200 GB image should be a LOT slower than a 100 GB image (if
there is no fan).  The second 100GB should take almost 200 seconds.

After 140GB, the caching cliff should hit.  Surprisingly, this doesn't
show up in the article's Thermal Throttling graphs.

There's something fishy here.

| > SSDs come with different performance trade-offs.  Most inexpensive
| > SSDs have (on-board) controllers with only small amounts of RAM.  This
| > makes them slow down a lot after a modest burst of intensive writing.
| > That's a fine trade-off for many of us but not for all workloads.

The Crucial drive has a 1 GiB RAM chip onboard, not like the cheap ones
I was thinking of.

I think that the cliff for cheap ones comes much earlier, due to
running out of mapping RAM.