Samsung 970 or new Phison PCIE Gen. 4

B

bose301s

Weaksauce
Joined
Jul 26, 2019
Messages
65
Building a new PC based on Ryzen 9 3900X and an X570 motherboard, looking at which storage to get, looking at the Samsung 970 series, either EVO Plus or Pro vs. the new Phison PCIE Gen 4 drives like the Corsair MP600 or Sabrent Rocket. Wondering what people think.
 
I have a ton of information about this subject on Reddit, I'm not sure where to begin here. I would generally advise against the 4.0 drives (E16) right now. The EVO Plus is very fast but quite expensive. If you're looking for everyday use, including gaming and light content creation, you're fine with a E12 or SM2262/EN drive, of which there are about two dozen to choose from. Check my spreadsheet.
 
If you want price, go with an E12, if you want performance go with an E16 (?). I'm as big a Samsung fan as anyone around (all of my builds up to my last one are running Sammies) but the price/performance ratio of these new drives are hard to argue against. These Phison are game changers and great for the customers - but not so much for the bottom line of other SSD makers.
 
I ordered the Sabrent E16, but seeing the Inland E12 drive for $108 vs. $200 I may return the Sabrent for it.
 
bose301s said:
I ordered the Sabrent E16, but seeing the Inland E12 drive for $108 vs. $200 I may return the Sabrent for it.
Click to expand...
I think I picked up my inland e12’s for around 90 with a micro center coupon. I know the price is slowly creeping up now though.
 
Got an inland premium 512 e12 based drive for my server, it's 1:1 or even slightly faster than my 970 evo 512 in my pc. Even that is only just barely better than my samsung 850 sata ssd. Outside of benchmarks and transfers over my 10gbit fiber connection to the server, it makes no difference at all. I honestly don't understand why anyone at all is considering the e16 drives. Price is creeping back up but this is still the drive to get (or its equivalent e12 based brethren): https://www.microcenter.com/product...80-pcie-nvme-30-x4-internal-solid-state-drive
 
Ordered the Inland, going to return the Sabrent, loving that you can do the drop off at Kohl's now for free instead of having to pay to return it.
 
Nenu said:
Do the E16 drives support 512e?
Click to expand...
Sabrent's Sector Size Converter supports their E16 drives as well. Most SSDs actually report 512 native or 4K native, not 512e. Either way the SSD doesn't use any of them internally so it makes no difference apart from cloning.
 
Monstieur said:
Sabrent's Sector Size Converter supports their E16 drives as well. Most SSDs actually report 512 native or 4K native, not 512e. Either way the SSD doesn't use any of them internally so it makes no difference apart from cloning.
Click to expand...
Yeah I have since bought a Sabrent E12 and converted it to 512, thanks anyway.
 
Maxx said:
I have a ton of information about this subject on Reddit, I'm not sure where to begin here. I would generally advise against the 4.0 drives (E16) right now. The EVO Plus is very fast but quite expensive. If you're looking for everyday use, including gaming and light content creation, you're fine with a E12 or SM2262/EN drive, of which there are about two dozen to choose from. Check my spreadsheet.
Click to expand...
Can you elaborate on this note in your spreadsheet re: Inland Premium."Incorrect write @ 1TB; 3-yr"

I have the 1TB version about 2/3rds full and can only get up to about 1300MBs writes. I did the upgrade to firmware 12.3 when it was released which initially improved writes up to about 3000MBs, after a while it fell off to the 1300MBs writes and after imaging/backing up the the drive and then zeroing out the drive I wrote the image back and for a while writes were good again, but eventually fell off. I only get the 1300MBs write if I run drive maintenance/TRIM right before, otherwise it's under 1000MBs.
 
zpackrat said:
Can you elaborate on this note in your spreadsheet re: Inland Premium."Incorrect write @ 1TB; 3-yr"

I have the 1TB version about 2/3rds full and can only get up to about 1300MBs writes. I did the upgrade to firmware 12.3 when it was released which initially improved writes up to about 3000MBs, after a while it fell off to the 1300MBs writes and after imaging/backing up the the drive and then zeroing out the drive I wrote the image back and for a while writes were good again, but eventually fell off. I only get the 1300MBs write if I run drive maintenance/TRIM right before, otherwise it's under 1000MBs.
Click to expand...

Inland had the wrong sequential write speed listed for that SKU. It's as fast as any other E12 drive in reality.

The speed you're seeing is the native TLC speed, which is roughly 1200 MB/s at 1TB (amount of dies varies with capacity, and thus does speed). The 3000 MB/s speed is only for the SLC cache. E12 drives have about 30GB of dynamic cache when empty but it gets smaller as the drive is filled (the SLC is just the TLC in single-bit mode, so it takes up three times the capacity). When this cache is exhausted it drops into direct-to-TLC mode. It juggles the SLC cache in the background at this point, so actually speed is closer to 1050 MB/s or so if it's flat-lined (that is, if you are looking at any given second's average). You can see that here with a drive that has the same hardware.

This is unfortunately one of the things people don't really grasp about TLC-based drives. I don't blame them, it's complicated. It's further confused by the fact that SLC cache design is not uniform: the WD SN750 has only a small, static cache, the SX8200 Pro has a large, dynamic cache, and the 660p has both! And it's easy for people to say having a large cache is better, but that's not always true. I won't get into that here though.

(the drive will always try to retain some SLC cache even when completely full, say 12GB, so at 2/3 full you might have like 18GB left)
 
Maxx said:
Inland had the wrong sequential write speed listed for that SKU. It's as fast as any other E12 drive in reality.

The speed you're seeing is the native TLC speed, which is roughly 1200 MB/s at 1TB (amount of dies varies with capacity, and thus does speed). The 3000 MB/s speed is only for the SLC cache. E12 drives have about 30GB of dynamic cache when empty but it gets smaller as the drive is filled (the SLC is just the TLC in single-bit mode, so it takes up three times the capacity). When this cache is exhausted it drops into direct-to-TLC mode. It juggles the SLC cache in the background at this point, so actually speed is closer to 1050 MB/s or so if it's flat-lined (that is, if you are looking at any given second's average). You can see that here with a drive that has the same hardware.

This is unfortunately one of the things people don't really grasp about TLC-based drives. I don't blame them, it's complicated. It's further confused by the fact that SLC cache design is not uniform: the WD SN750 has only a small, static cache, the SX8200 Pro has a large, dynamic cache, and the 660p has both! And it's easy for people to say having a large cache is better, but that's not always true. I won't get into that here though.

(the drive will always try to retain some SLC cache even when completely full, say 12GB, so at 2/3 full you might have like 18GB left)[/QUOTE

I appreciate you detailing the whys of the drop off, the thing I found odd was that for a week or so after and zeroing the drive and restoring the backup to the same % used give or take, I was able to test at the higher write speeds. then a couple of weeks later after testing, it reverted to the slower speeds. Any ideas on why?
Click to expand...
 
Doing a secure erase or sanitize - you NEVER want to "zero" a NAND-based drive, by the way, or anything like you would do to secure a HDD - resets the drive to its factory condition so its SLC cache works as if you just pulled the drive out of the box. It's not until you write the drive once and fill it a bit that you reach a realistic performance state. Some people call this "steady state" but that's a bit misleading - when I say "steady state" I mean once you hit the drive's long-term performance after exceeding the SLC cache. In any case, these two performance areas may be very similar for you with that drive because the SLC cache is relatively small.

Why is this? Because the drive might not empty the SLC cache immediately or even in the background. It might retain data in the SLC cache. It might not want to shrink the cache unless it has to, etc. This is partially due to predictive algorithms in the controller: it's trying to maintain the best balance of performance and response time. Data in the SLC cache can be folded out later, beneficial because folding writes sequentially (less wear) and if data is not permanent these writes can be deferred (also improving endurance). Data that's in transition (being folded) has a read latency penalty which is undesirable. Etc. It's a relatively complicated process - Malventano (who now works for Intel research) shows the potential inconsistency of a drive here.

Now that's not to say there isn't something wrong with your drive. I've seen some drives get stuck in direct-to-TLC mode, specifically small batches of the SX8200 and SX8200 Pro. I can't rule out it happening to other TLC-based drives. However it is a minority issue (very small batches) when it exists and it's firmware-based (should be fixed with overwritten firmware) generally. It's up to the controller to manage the SLC-capable portion of the NAND dies. I can't say that 100% of drives work properly because they don't, but it is like 99%. I guess that doesn't ease your concerns though.
 
Last edited:
  • Like
Reactions: Nenu
like this
You must log in or register to reply here.
Back
Top