Since NVMe supports up to 64K queues with 64K commands each, would NVMe SSDs show any scaling beyond what AHCI SSDs are usually tested at, at just queue depths of 32-128?
@Flunk: "Who cares? That's like saying how fast would that Ferrari be without tires? It's a situation that is never going to happen and is totally fabricated."
I'm not pickin up what you're puttin down here.
Queue depths for normal home usage scenarios are usually between 1 and 5. Running one of these in a normal home usage case would be, to use your example, like driving a Ferrari without tires. Bumping it up just beyond the standard AHCI queue depths into the 32 - 128 range, would be like putting some (perhaps cheap) tires on those previously bare rims.
I bought one of these for my 5820K based system. I am an average gamer with some business activities as well. I found no difference in perceptible speed while booting, opening applications, zoning in games, etc compared to a Samsung 840 Pro.
Or even the X300; I put a 256GB in my HTPC just as a stopgap until I could afford an M.2 NVMe (it was on special offer at the time, cheaper than any other product), but it runs so well I don't think I'll bother changing it now, doubt I'd ever notice the difference. It exhibits very consistent performance (flat 422MB/sec in HDTach right across the graph; high reads in AS-SSD, somewhat less on writes, scores 852 overall, actually better than the X400 in some cases), and I say this being well used to "better" models like the 850 Pro (I have several). I also like its fast sleep/recover speed. I wonder if the X400 behaves as consistently. It's slower than the 840 Pro though, so Chaser's comment is interesting.
I like the idea of having an NVMe for some of my systems, but then I look at the cost and think, a SATA would give double the capacity at a lower price.
If anyone's interested, I've accumulated some test results for various SSDs here (AS-SSD, CDM, HDTachRW and ATTO):
"At the business end of the review, we can clearly see that the Toshiba OCZ RD400 M.2 NVMe SSD is a good choice for use in the Intel Skylake NUCs such as the NUC6i5SYK."
Yeesh. A $310 storage drive for a $675 system? How is that a good choice? This conclusion has completely ignored what a normal consumer would do.
Come on, Anandtech. Let's pick up the critical analysis here beyond reading off benchmarks.
Just in case someone posts who has NOT bought an M.2 drive recently:
The entire RD400 series is overpriced for a normal consumer to "get NVMe", which this conclusion admits no one will notice without reading benchmarks! The 128GB RD400 model is $140 on Amazon. Yuck. There are dozens and dozens of perfectly fine AHCI M.2 SSDs.
Though not really relevant for the the NUC, there are many motherboards that disable an available SATA port if the M.2 drive uses AHCI. Whereas using NVMe frees up that SATA port and allows an additional drive to be supported. (Not sure if the same applies when using the PCIe add-in card--probably depends on the motherboard)
So the price premium for NVMe can be worth it even if there is no performance benefit.
That has been done in our OCZ RD400 review for small transfer sizes.
For typical NUC workloads like media editing, we have that covered in our thermal stress testing workload does. Even though temperatures go up to almost 70C, there is no evidence of throttling.
In the conclusion you refer to PCIe 2.0 where in the specifications it is PCIe 3.0, you cannot not get 2635MB/s through a PCIe 2.0 x 4 connection. "Kingston HyperX Predator PCIe 2.0 x4 AHCI SSD"
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
16 Comments
Back to Article
tipoo - Tuesday, July 5, 2016 - link
Since NVMe supports up to 64K queues with 64K commands each, would NVMe SSDs show any scaling beyond what AHCI SSDs are usually tested at, at just queue depths of 32-128?Flunk - Tuesday, July 5, 2016 - link
Who cares? That's like saying how fast would that Ferrari be without tires? It's a situation that is never going to happen and is totally fabricated.tipoo - Tuesday, July 5, 2016 - link
I care academically ¯\_(ツ)_/¯XZerg - Tuesday, July 5, 2016 - link
never say never... 640K ought to be enough.Programmers tend to limit use case scenarios as much as possible if there is no technology to make a good user experience.
BurntMyBacon - Wednesday, July 6, 2016 - link
@Flunk: "Who cares? That's like saying how fast would that Ferrari be without tires? It's a situation that is never going to happen and is totally fabricated."I'm not pickin up what you're puttin down here.
Queue depths for normal home usage scenarios are usually between 1 and 5. Running one of these in a normal home usage case would be, to use your example, like driving a Ferrari without tires. Bumping it up just beyond the standard AHCI queue depths into the 32 - 128 range, would be like putting some (perhaps cheap) tires on those previously bare rims.
Chaser - Tuesday, July 5, 2016 - link
I bought one of these for my 5820K based system. I am an average gamer with some business activities as well. I found no difference in perceptible speed while booting, opening applications, zoning in games, etc compared to a Samsung 840 Pro.Impulses - Tuesday, July 5, 2016 - link
All pretty light usage cases, I wouldn't expect anyone to notice a difference in that scenario even between a 950 Pro and a cheap Sandisk X400.Chaser - Tuesday, July 5, 2016 - link
Ya think?mapesdhs - Wednesday, July 13, 2016 - link
Or even the X300; I put a 256GB in my HTPC just as a stopgap until I could afford an M.2 NVMe (it was on special offer at the time, cheaper than any other product), but it runs so well I don't think I'll bother changing it now, doubt I'd ever notice the difference. It exhibits very consistent performance (flat 422MB/sec in HDTach right across the graph; high reads in AS-SSD, somewhat less on writes, scores 852 overall, actually better than the X400 in some cases), and I say this being well used to "better" models like the 850 Pro (I have several). I also like its fast sleep/recover speed. I wonder if the X400 behaves as consistently. It's slower than the 840 Pro though, so Chaser's comment is interesting.I like the idea of having an NVMe for some of my systems, but then I look at the cost and think, a SATA would give double the capacity at a lower price.
If anyone's interested, I've accumulated some test results for various SSDs here (AS-SSD, CDM, HDTachRW and ATTO):
http://www.sgidepot.co.uk/misc/ssdtests.zip
ikjadoon - Tuesday, July 5, 2016 - link
"At the business end of the review, we can clearly see that the Toshiba OCZ RD400 M.2 NVMe SSD is a good choice for use in the Intel Skylake NUCs such as the NUC6i5SYK."Yeesh. A $310 storage drive for a $675 system? How is that a good choice? This conclusion has completely ignored what a normal consumer would do.
Come on, Anandtech. Let's pick up the critical analysis here beyond reading off benchmarks.
ikjadoon - Tuesday, July 5, 2016 - link
Just in case someone posts who has NOT bought an M.2 drive recently:The entire RD400 series is overpriced for a normal consumer to "get NVMe", which this conclusion admits no one will notice without reading benchmarks! The 128GB RD400 model is $140 on Amazon. Yuck. There are dozens and dozens of perfectly fine AHCI M.2 SSDs.
keoki - Tuesday, July 5, 2016 - link
Though not really relevant for the the NUC, there are many motherboards that disable an available SATA port if the M.2 drive uses AHCI. Whereas using NVMe frees up that SATA port and allows an additional drive to be supported. (Not sure if the same applies when using the PCIe add-in card--probably depends on the motherboard)So the price premium for NVMe can be worth it even if there is no performance benefit.
bj_murphy - Tuesday, July 5, 2016 - link
No testing of performance consistency?ganeshts - Tuesday, July 5, 2016 - link
That has been done in our OCZ RD400 review for small transfer sizes.For typical NUC workloads like media editing, we have that covered in our thermal stress testing workload does. Even though temperatures go up to almost 70C, there is no evidence of throttling.
bj_murphy - Wednesday, July 6, 2016 - link
Ah, I understand. I hadn't clued in that the RD400 was separately reviewed in another article. Thanks for the clarification, Ganesh.Redstorm - Wednesday, July 6, 2016 - link
In the conclusion you refer to PCIe 2.0 where in the specifications it is PCIe 3.0, you cannot not get 2635MB/s through a PCIe 2.0 x 4 connection."Kingston HyperX Predator PCIe 2.0 x4 AHCI SSD"