SSDs - State of the Product?

Santilli

Hairy Aussie
Joined
Jan 27, 2002
Messages
5,060
If you read a little further:
"

I found this, did not try it myself.

Not the Sabertooth but a similar ASUS board with a P6X58D Premium. But I made the Samsung 950 Pro work with it easy with an Asus M.2 PCI-E card and now also working with a riser cable due to space constraints. Speeds are of course PCI-E 2.0 at around 1700 mbps sequential read but still a huge increase in speeds.

AFAIK the 950 Pro is the only NVMe drive that works with X58 without bios modding cause it has a Legacy Boot Option built in. The 960 Pro/EVO doesn't have this option AFAIK.

Pics and instructions on this link:
links are gone....

https://www.asus.com/ca-e...HYPER_M2_X4_MINI_CARD/

https://audiocricket.com/...asus-p6t-se-mainboard/

"

??
 

Santilli

Hairy Aussie
Joined
Jan 27, 2002
Messages
5,060
May look into this:
2. Install a PCIe x1 SATA 3 card and install Samsung 850 Pro and have it bottleneck over PCIe 2.0 x1, but be ensured AHCI
 

LunarMist

I can't believe I'm a
Joined
Feb 1, 2003
Messages
14,816
Location
USA
Holy crap that is a 2008 CPU and chipset! It should be in a museum or relegated to some secondary purpose.
Maybe replacing the ancient X25Ms with a single modern SSD will help a little.
 

Santilli

Hairy Aussie
Joined
Jan 27, 2002
Messages
5,060
Motherboard has Sata 2. Raid 0 gives 450 mb/sec, X-25's
IIRC, that's about the limit of Sata 2. Sata 3 or 6, whichever, is around 600MB/sec.

When I looked at the options, I also came to think either a Mirror 1 Raid, or a single newer drive might be a better idea.

I use little more then half the 297 gigs for the boot drive.

Recently put a high end T-Link wifi card in it, two PCIE cards, one to USB C and 3.1, and the other 3.0, 5 ports external, and mainly two for the interior, which makes the front ports on the case work at 3.0, as they are designed.

Windows Picture viewer is having a hard time pulling up pictures off the various drives, 6 right now, and that's the only real speed issue I have.

Money is a bit tight, so I'm not likely to replace a setup that is doing everything I want it to do.

The card and the tv do an excellent job of playing 4k, or upscaling Bluray, so the picture is excellent to incredible.
 

Santilli

Hairy Aussie
Joined
Jan 27, 2002
Messages
5,060
Crystal Disk Mark Evo 860 285 mb/sec. vs. 440 for Intel Raid 0. However, the access time for everything seems instant with the 860. Pictures viewers' lag is gone.
Samsung Disk Migration works perfectly.
Samsung Magician appears to work as well.
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
12,856
Location
USA
Crystal Disk Mark Evo 860 285 mb/sec. vs. 440 for Intel Raid 0. However, the access time for everything seems instant with the 860. Pictures viewers' lag is gone.
Samsung Disk Migration works perfectly.
Samsung Magician appears to work as well.
I'm sure I've mentioned this to you in the past...the raw MB/sec isn't a great way to gauge a storage solution, it's one measurement of many to consider. You should compare IOps at 4KB between SSD and HDD and then you'll have a better idea of how fast SSDs can manage much smaller blocks of data in random patters. The random access is the true burden of storage performance, not monolithic contiguous access.

Example:
Samsung EVO 860 SSD: Max. 98,000 Read IOPS / 90,000 write IOPS
7,200 RPM SATA 75-100 IOPS
 

Santilli

Hairy Aussie
Joined
Jan 27, 2002
Messages
5,060
Odd you brought that up. I was looking through my old pictures, and found my first Crystal Diskmark test of the X-25M raid, and I have the current Evo 860 test:
X-25's
493. 330. 21. 268.

860
285. 214. 214. 40.

I ran a 7200 Sata and the numbers:

111. 3.6 3.6 1.4

so 4k random 860 is 10 times faster then X-25s, and about 60 times faster then a standard drive.

If I raid 0 another drive, I might get a gaudy seq read, but little increase in actual 4k random access.

I found a guy that did just that, and the 4k Random reads were barely 40% faster then the single drive.

Also with my ancient motherboard, I wonder if the onboard controller is incapable of taking advantage of drives designed 10 years after the chipset, and raid controller came out?
 

Santilli

Hairy Aussie
Joined
Jan 27, 2002
Messages
5,060
Hi Handruin:
Thank you for the recommendation.
I just looked at the CDM on the Samsung 960 and 970 MVME drives.
While the sequential tests are around 3500 mb/sec, the 4k randoms are only 3 times faster then the Sata drives, 600 mb/sec.
The seq test goes up 7 times, 4k random 3x.
What do you think is causing the 4k access times to not increase with the seq tests.?
 

Handruin

Administrator
Joined
Jan 13, 2002
Messages
12,856
Location
USA
Hi Handruin:
Thank you for the recommendation.
I just looked at the CDM on the Samsung 960 and 970 MVME drives.
While the sequential tests are around 3500 mb/sec, the 4k randoms are only 3 times faster then the Sata drives, 600 mb/sec.
The seq test goes up 7 times, 4k random 3x.
What do you think is causing the 4k access times to not increase with the seq tests.?
There's a lot going on in there that isn't described so I can only make some assumptions in the testing. For example, if your goal is to get the highest sequential transfer, you might want to do a test using a high queue depth (maybe 32 or 64) with a block size of 1MB or larger (not 4K). Higher queue depth typically equates to better performance but results might vary. Larger block sizes can be easier for the controller to manage which is why you may see larger jumps in sequential performance when testing with larger blocks. You also have to know more about how the data is stored on the internal NAND in their page size to optimize to its advantage if your goal is to get the fastest possible speed...which is unlikely to be applicable to real-world situations. That will test the limits of sequential transfer of a device as well as the bus overhead and bandwidth limitations. Most things if life are more random so the true strength of a storage device will be how it handles random data access patterns at a queue depth of 1 when compared to sequential access with a large queue depth.

You can not also compare SATA3 to NVMe and expect linear changes. NVMe is much newer and more efficient as a protocol in addition to being enabled on a much faster bus. The Samsung 970 EVO NVMe claims to have the ability to process 500,000 random 4K reads and 450,000 random 4K writes. That's 5x faster on both reads and writes when compared to a SATA SSD. That's quite significant. Real-world may be difficult to notice the difference. At this point you are just being motivated by paper specs.

You can also do the paper math to see what this means in a worst-case situation:

500,000 * 4096 = 2048000000 bytes / per sec
2.048 GB/sec random 4K reads is pretty damn good on paper.

As for your final question, I don't quite understand what you're asking me. Why would 4K access times increase with the sequential tests?
 

Santilli

Hairy Aussie
Joined
Jan 27, 2002
Messages
5,060
Hmmm. This seems to have a LOT to do with the controller used.
I have an old drive I'm trying to get data off. Wouldn't work with bios set to Raid.
Changed to ide. It works.
The 860 evo CDM tests are:

238. 44.68 44.16 39.60

The computer seems a little bit less snappy.
the 4k random is down 80% or so...
 

Santilli

Hairy Aussie
Joined
Jan 27, 2002
Messages
5,060
Got done working with the old drive, or at least one of them.
Decided to go back to ACHI mode.
Computer blue screens, and won't boot.
Switched to Raid, and have the following numbers:
279. 210 207. 40.

So random are now 5 times faster.

See if it's noticeable.
 
Top