Silicon Power XS70 1 TB SSD

[W1zzard]

Priced at $130 for 1 TB, the Silicon Power XS70 is one of the most affordable PCIe 4.0 SSDs. It still offers outstanding performance because of the Phison E18 controller and Micron 176-layer TLC NAND combination. An excellent heatsink is preinstalled, too. The drive never reaches its thermal throttle point.

Show full review

 

[Udyr]

Price at Time of Review:	$130 / $130 per GB

 

[ThrashZone]

Hi,
At least it also has a heat sink
Although some boards may not be able to use it.

 

[thewan]

I don't know why where I live, the 980 Pro is ~260 USD, while both the Corsair MP600 Pro LPX and the SN850 is ~210 USD. You would then think the Kingston KC3000 is higher priced than those two based on TPU reviews, but no it is ~175 USD, all 1TB after conversion. I bought one KC3000 1TB for my relatives new PC build. Hopefully this Silicon Power one arrives here and is priced better. Silicon Power is probably more well known here than say T-Force or even Crucial.

 

[ThrashZone]

Hi,
No region selected so you're the only one that could know.

 

[W1zzard]

I don't know why where I live, the 980 Pro is ~260 USD, while both the Corsair MP600 Pro LPX and the SN850 is ~210 USD. You would then think the Kingston KC3000 is higher priced than those two based on TPU reviews, but no it is ~175 USD, all 1TB after conversion. I bought one KC3000 1TB for my relatives new PC build. Hopefully this Silicon Power one arrives here and is priced better. Silicon Power is probably more well known here than say T-Force or even Crucial.

Yeah it depends on the region of course. Any drive in your list, if you can find it cheap, will be a good choice

 

[maxfly]

Well done
I appreciate your making it easier to sift through the glut of drives out there right now.
Its nice to see pricing continuing to flatten out compared to a year ago.

 

[Chrispy_]

I'm just glad that QLC nonsense appears to be falling out of favour with manufacturers and that half-decent TLC stuff is still competing with it on price.

Even for trivial stuff like Windows boot up time, as the chart gets updated with more and more models, the QLC junk is all the bunched up with the very old SATA drives as the slow stuff.

 

[Wooden Law - Black]

I'm just glad that QLC nonsense appears to be falling out of favour with manufacturers and that half-decent TLC stuff is still competing with it on price.

Do you really think that QLC is failing? In the enteprise market they are very popular, look at Intel drives for example, which are QLC in pTLC mode.

 

[Solid State Brain]

@W1zzard
This is more of a general question, but what benchmarking application is employed in the Random Access Performance test to bring out about 50 kIOPS with random QD1 4kB blocks as per this graph?

 

[Chrispy_]

Do you really think that QLC is failing? In the enteprise market they are very popular, look at Intel drives for example, which are QLC in pTLC mode.

Intel exited the entire SSD market in 2020 because it wasn't profitable enough.

Each jump from SLC > MLC > TLC > QLC scaled up in capacity less and less each time whilst controller costs and complexity scaled up more and more each time.

The fact that QLC has to operate in pTLC and pSLC modes at all proves that it's not really fast enough as a native NAND type. Operating QLC in pTLC mode is more expensive than just using purpose-built TLC NAND.

The need to meet endurance and performance specifications is why QLC is likely the last bits-per-cell increase we'll see in a long time, and it's only applicable to specific use cases - ie budget products with short warranties and light workloads, or enterprise products where the data requirements are WORM or extremely capacity-intensive.

 

[W1zzard]

This is more of a general question, but what benchmarking application is employed in the Random Access Performance test to bring out about 50 kIOPS with random QD1 4kB blocks as per this graph?

Diskspd, the graph is made in Excel

 

[Solid State Brain]

Diskspd, the graph is made in Excel

Thanks, I didn't know about this program. I generally use Flexible I/O Tester (fio), also available for Windows.

I just tried Diskspd on a Windows PC. Are you using something like this for testing 4 kB QD1 random reads?

./diskspd -b4k -o1 -r -Sh -c200M testfile.out

At the moment I can only try it with an old Samsung 840 120GB, but the above command gives rather realistic results for this SSD used as a system drive (6425 IOPS, 25.1 MiB/s). However, I find that Random Access Performance results in TPU SSD reviews give almost always significantly higher values than expected.

For instance, Samsung claims 22000 IOPS performance for 4K QD1 random read workloads for its 980 Pro, which is about half of what was tested some time back on TPU for the same SSD and also the Silicon Power XS70 of this review. I am currently using a Samsung 980 Pro 1 TB on Linux and 4K QD1 random read performance is definitely closer to Samsung data than ~50 kIOPS.

 

[Chrispy_]

Thanks, I didn't know about this program. I generally use Flexible I/O Tester (fio), also available for Windows.

I just tried Diskspd on a Windows PC. Are you using something like this for testing 4 kB QD1 random reads?

./diskspd -b4k -o1 -r -Sh -c200M testfile.out

At the moment I can only try it with an old Samsung 840 120GB, but the above command gives rather realistic results for this SSD used as a system drive (6425 IOPS, 25.1 MiB/s). However, I find that Random Access Performance results in TPU SSD reviews give almost always significantly higher values than expected.

For instance, Samsung claims 22000 IOPS performance for 4K QD1 random read workloads for its 980 Pro, which is about half of what was tested some time back on TPU for the same SSD and also the Silicon Power XS70 of this review. I am currently using a Samsung 980 Pro 1 TB on Linux and 4K QD1 random read performance is definitely closer to Samsung data than ~50 kIOPS.

I wouldn't read too much into synthetic benchmarks, that's why they're called synthetic - they're not real.

All that matters is that the test, no matter how realistic or not it may be, is consistent. It provides a meaningful heirarchy of drives from bad to good. What the absolute numbers say are irrelevant, as they're only valid within the particular limitations of that synthetic, repeatable workload.

 

[Solid State Brain]

@Chrispy_
The hierarchy might look different if QD1 is not actually QD1, or 4k not exactly 4k, or a buffer is being used (usually SSD benchmarks are with direct, unbuffered I/Os), depending on benchmark settings. It might be something worth looking into, if anything at least for the sake of correctness.

For what it's worth, from a test I made on my own Samsung 980 Pro on Linux using the fio benchmark it looks as if the results on Techpowerup are shifted one queue depth position, i.e. QD1 is actually QD2. But they also cap at lower values; I don't know what to make of that.

 

[Wooden Law - Black]

Intel exited the entire SSD market in 2020 because it wasn't profitable enough.

I'm talking about what they sold, in fact. In the 2020 they sold like more than 10 milions of QLC SSDs in the enteprise market...
Also, QLC didn't fail, QLC is the basis of X-NAND, it gives the flexibility for pSLC, pMLC and pTLC mode.

The fact that QLC has to operate in pTLC and pSLC modes at all proves that it's not really fast enough as a native NAND type.

So following this reasoning, also TLC isn't "fast enough as a native NAND type" since it works in pSLC mode?

 

[Viruzz]

OP, what type of NAND the XS70 has, Fortis Grade like updated Rocket 4 Plus or Media Grade like addlink AddGame A95?

 

[W1zzard]

OP, what type of NAND the XS70 has, Fortis Grade like updated Rocket 2 or Media Grade like addlink AddGame A95?

Checking, to confirm with Silicon Power

Edit: it's FortisFlash according to Silicon Power

 

[Viruzz]

Checking, to confirm with Silicon Power

Edit: it's FortisFlash according to Silicon Power

WOW, thank you that just moves it up a notch into best SSDs of this period like re-released Rocket 4 Plus

 

[Wooden Law - Black]

OP, what type of NAND the XS70 has, Fortis Grade like updated Rocket 4 Plus or Media Grade like addlink AddGame A95?

Both 4 Plus and A95 use the same flash with the same grade (Micron B47R - 176L TLC - FortisFlash).

 

[Viruzz]

Both 4 Plus and A95 use the same flash with the same grade (Micron B47R - 176L TLC - FortisFlash).

Hi,
The A95 uses Media Grade B47R

As we watched the numbers coming in, we had our eye on the 4K Q1T1 random read because we suspected that it might be arrayed with Media Grade 176 Layer Micron based on its relatively low pricing. 76 MB/s Q1T1 random read on our AMD test system confirmed our suspicions that the A95 is arrayed with Media Grade B47R flash. Makes sense based on its price point

Read more: https://www.tweaktown.com/reviews/10002/addlink-addgame-a95-2tb-ssd/index.html

 

[Wooden Law - Black]

Read more: https://www.tweaktown.com/reviews/10002/addlink-addgame-a95-2tb-ssd/index.html

I don't think that he means B47R FortisMax.

 

[Viruzz]

I don't think that he means B47R FortisMax.

What do you mean?

There is 2 types of B47R NAND, one made for Enterprise, Fortis Grade and one made for Consumer products, Media Grade, its identical, almost identical, the media grade is tad slower and has lower endurance.
The reviewer said that A95 comes with Media Grade B47R flash,
Also he has benchmarked and compared A95 vs Rocket 4 Plus, which has higher 4K speeds, so thats the main difference between Fortis and Media grade

 

[Wooden Law - Black]

What do you mean?

There is 2 types of B47R NAND, one made for Enterprise, Fortis Grade and one made for Consumer products, Media Grade, its identical, almost identical, the media grade is tad slower and has lower endurance.
The reviewer said that A95 comes with Media Grade B47R flash,
Also he has benchmarked and compared A95 vs Rocket 4 Plus, which has higher 4K speeds, so thats the main difference between Fortis and Media grade

Micron's flash is usually classified into two grades: FortisFlash (the most common and used in the consumer market) and FortisMax. FortisFlash has a lower endurance than the FortisMax in terms of PEC (P/E cycles).

 

[Viruzz]

Micron's flash is usually classified into two grades: FortisFlash (the most common and used in the consumer market) and FortisMax. FortisFlash has a lower endurance than the FortisMax in terms of PEC (P/E cycles).

Ok now i get it, so the reviewer calling it Media Grade, means that its FortisFlash and the Rocket 4 is FortisMax?
It does have Lower q1t1 reads, 76 vs 88