Amazon - SK hynix Platinum P41 PCIe NVMe Gen4 M.2 2280 Internal SSD 1TB ($102) and 2TB ($170)

check for the amazon shipped and sold one - they're backordered til mid-Dec or so (yet still purchasable), so that's why it defaults to the other sellers
Does it have memory modules on both sides? Most motherboard heatsinks only cool one side? How would the backside be cooled? Any consideration about this?
 
Does it have memory modules on both sides? Most motherboard heatsinks only cool one side? How would the backside be cooled? Any consideration about this?
I checked some reviews, and it does appear that the 4TB version has modules on both sides. I'd guess that the back side would be cooled by any bit of air flowing through and any heat that dissipates through to the other side and the heatsink.

I'm not the expert on it, I just responded to your query first xD
 
Great catch I'd kinda forgotten about MP600, it's probably the best consumer drive in post-cache speed for heavy sustained write scenarios, at least before jumping to enterprise class.

Not to derail here because most peoples' daily general use is well within the boundaries of what most SSD cache's can accomodate, and aren't pushing around multiple hundreds of GBs or TBs per day. I'm just a stickler for knowing how every layer of a SSD performs.
 
Last edited:
Great catch I'd kinda forgotten about MP600, it's probably the best consumer drive for post-cache sustained write scenarios before jumping to enterprise class.

Not to derail here because most peoples' daily general use is well within the boundaries of what most SSD cache's can accomodate, and aren't pushing around multiple hundreds of GBs or TBs per day. I'm just a stickler for knowing how every layer of a SSD performs.
What do you mean by post cache? Is the the issue with the 980pro people are talking about? Something about the cache slowdown?
 
What do you mean by post cache? Is the the issue with the 980pro people are talking about? Something about the cache slowdown?
980 Pro's appear to be failing simply from bad blocks.
The write cache is a memory buffer used to help sustain write speeds. "Post-cache" I assume refers to the write speeds after the cache is full.

"Most SSDs implement a write cache, which is a fast area of (usually) pseudo-SLC programmed flash that absorbs incoming data. Sustained write speeds can suffer tremendously once the workload spills outside of the cache and into the "native" TLC or QLC flash. We use Iometer to hammer the SSD with sequential writes for 15 minutes to measure both the size of the write cache and performance after the cache is saturated. We also monitor cache recovery via multiple idle rounds."

J654fY3Dd8mcjx5D3rjkZg-970-80.png.webp
 
Last edited:
I sent Hynix this message asking about their plans for DirectStorage support:

Hello,

The Hynix Platinum P41 is a fast m.2 SSD, but a test for its performance with Microsoft's DirectStorage shows it performing poorly compared to several other m.2 SSDs, notably m.2 SSDs with the Phison IO+ firmware, and also the WD SN850 and WD SN850X, which get very consistent performance with DirectStorage. Is there a plan to address the DirectStorage performance of the Hynix Platinum P41 with a firmware update or by some other means, to make it more competitive for DirectStorage uses, such as future gaming?

The Hynix Platinum P41's DirectStorage performance can be seen in this link:

https://www.tomshardware.com/features/the-directstorage-advantage-phison-io-ssd-firmware-preview/2
I hope they'll respond with a mention of a plan.

This review of the P41 says:
The Platinum P41 is also competitive in Crystal Disk Mark’s random performance tests. We again see a bit of disparity when looking at the very high QD256 results, as the drive dominates with reads but falls behind a bit with writes. Most users look at low queue depth, random results to get an idea of general performance, which is valid, but the Platinum P41’s strengths here do imply some interesting things. For one, it really can hit 1.4M+ IOPS with reads, but the QD1 sequential read test is also quite high. These metrics may become more important once the DirectStorage API matures. As such, this is a future-looking drive.
 
I had to cancel the P41 order because WD dropped the price of the SN850X and I found out that Amazon can do free shipping direct to an overseas address while with the P41, it would have had to go through a shipping forwarding service. I was planning to use the new 2TB NVMe for a dedicated game drive so I guess I made the right move?

That said, let's see how this DirectStorage thing plays out.
 
980 Pro's appear to be failing simply from bad blocks.
The write cache is a memory buffer used to help sustain write speeds. "Post-cache" I assume refers to the write speeds after the cache is full.

"Most SSDs implement a write cache, which is a fast area of (usually) pseudo-SLC programmed flash that absorbs incoming data. Sustained write speeds can suffer tremendously once the workload spills outside of the cache and into the "native" TLC or QLC flash. We use Iometer to hammer the SSD with sequential writes for 15 minutes to measure both the size of the write cache and performance after the cache is saturated. We also monitor cache recovery via multiple idle rounds."

J654fY3Dd8mcjx5D3rjkZg-970-80.png.webp
So according to this chart the sabrent rocket is the only drive that doesn't suffer as much of a slow down as the rest of the drives?
 
I put in a last-minute order for the WD SN850X. It's backordered so it won't ship under later in December, and I could cancel it before then if I change my mind - and if I don't, I'd have 30 days to return it after that. And I think I won't be charged until it ships. I also have 30 days to return a Hynix P41, if I choose to keep the SN850X. I'm hoping to find out more about these drives' DirectStorage performance in that time, so I can make a decision.

So according to this chart the sabrent rocket is the only drive that doesn't suffer as much of a slow down as the rest of the drives?
Yeah, that chart shows the where the write speeds drop-off for various drives. The Sabrent Rocket drops from its peak write-performance pretty quickly, but might make up for it if you're doing lots of large copying / writing. For most common usage, I think the 50 seconds of sustained maximum write-speed that the P41 gets is pretty good.
 
I put in a last-minute order for the WD SN850X. It's backordered so it won't ship under later in December, and I could cancel it before then if I change my mind - and if I don't, I'd have 30 days to return it after that. And I think I won't be charged until it ships. I also have 30 days to return a Hynix P41, if I choose to keep the SN850X. I'm hoping to find out more about these drives' DirectStorage performance in that time, so I can make a decision.


Yeah, that chart shows the where the write speeds drop-off for various drives. The Sabrent Rocket drops from its peak write-performance pretty quickly, but might make up for it if you're doing lots of large copying / writing. For most common usage, I think the 50 seconds of sustained maximum write-speed that the P41 gets is pretty good.
Amazon/Newegg both Jan 31 return for everything right now. And you won't be able to tell any diff between those drives outside of benches as they're so close.

In general, gotta take a step back from these almost neurotic SSD spec and purchasing deliberations we get sucked into, and realize you'd be hard pressed to find a significant diff between high end Gen4 NVme and even a SATA500 SSD.

The real problem is we need an OS level "DirectStorage" - a ground-up rewrite and rethinking of storage in the OS that's designed for modern Gen4/Gen5 era SSDs as well as NVDIMM (NV storage and RAM will eventually merge and no longer be separate entities). We're still running an OS and filesystem on archaic spaghetti code 3-4 decades old whose primary storage layer interactions happen in 4K chunks.

Basically we're buying faster and faster racecars just to drive them on a highway with a gated stopsign every 2 inches. But because starting and stopping is near instant with SSDs, they're not that apparent.
 
Last edited:
If Amazon, returns are until Jan 31 on everything right now. And you won't be able to tell any diff between those drives outside of benches as they're so close. They're both very good.

In general, gotta take a step back from these almost neurotic SSD spec and purchasing deliberations we get sucked into, and realize you'd be hard pressed to find a noticeable diff between high end Gen4 NVme and even a SATA500 SSD.

The real problem is we need an OS level "DirectStorage", so to speak, if not a ground up rewrite built for NVDIMM eventuality (NV storage and RAM will eventually merge and no longer be separate entities). We're still running an OS and filesystem built on archaic spaghetti code 3-4 decades old whose primary storage layer interactions happen in 4K chunks.

In other words we're buying faster and faster racecars just to drive them on a highway with a gated stopsign every 2 inches. But because starting and stopping near instant with SSDs, they're not that apparent.
So when installing say a massive game in the magnitude of 200gb for example that definitely takes longer than 50 seconds or whatever the drop off is for most NVMe drives, would the slowdown rear it's head or no? When would the slowdown rear it's head actually? A massive single video file if you are a videographer? But not in a game because a game has multiple different sized files? Or a game would suffer also? I still haven't grasped when the slowdown occurs EXACTLY?
 
So when installing say a massive game in the magnitude of 200gb for example that definitely takes longer than 50 seconds or whatever the drop off is for most NVMe drives, would the slowdown rear it's head or no? When would the slowdown rear it's head actually? A massive single video file if you are a videographer? But not in a game because a game has multiple different sized files? Or a game would suffer also? I still haven't grasped when the slowdown occurs EXACTLY?
One scenario: upgrading your Steam drive to larger NVMe SSD, and your existing already has 500GB-2000GB of games installed. You drag and drop steam folder to new drive, and after a few hundred GB you notice copy speed drop significantly, and might wonder why.

But a SN850X writing to its underlying TLC at 900MB/s, or Hynix P41 at 1.4GB/s is still very fast. Will just take a bit longer than lets say Corsair MP600 Pro 2TB writing at 3.5GB/s (assuming source drive is fast enough to saturate it, and no other bottlenecks created by your PC's configuration)

And also consider how often you're actually going to be copying an entire full drive, or restoring a big backup image that's 300GB-1000GB. Not often for most users.
 
Last edited:
One scenario: upgrading your Steam drive to larger NVMe SSD, and your existing already has 500GB-2000GB of games installed. You drag and drop steam folder to new drive, and after a few hundred GB you notice copy speed drop significantly, and might wonder why.

But a SN850X writing to its underlying TLC at 900MB/s, or Hynix P41 at 1.4GB/s is still very fast. Will just take a bit longer than lets say Corsair MP600 Pro 2TB writing at 3.5GB/s (assuming source drive is fast enough to saturate it, and no other bottlenecks created by your PC's configuration)

And also consider how often you're actually going to be copying an entire full drive, or restoring a big backup image that's 300GB-1000GB. Not often for most users.

Seems like outside of reliability the most important factor is price. WD Black drives, Corsair P5 Plus, SK Hynix, Samsung are generally reliable. For most people it is best to find the best pricing.

Yesterday the Crucial P5 Plus 2TB was $160 from newegg. It is back up to $200. If you're looking for a name brand SSD that is probably very reliable, that is a good choice if the price goes back down and you need something as cheap as possible.
 
Last edited:
  • Like
Reactions: DPI
like this
Just a thought:

With all this talk about NVMe SSDs and firmware optimized for Direct Storage, in the case of the PS5 where the user is allowed to install non proprietary NVMe SSDs in the provided slot on the console, the loading times do not seem to differ much regardless of SSD used and also in comparison to the internal SSD. I have always assumed that the PS5 has a custom optimized storage architecture that's somewhat analogous to Direct Storage.

I seem to recall a video from Digital Foundry looking into this and they found that the add-in SSD was actually faster than the internal one in loading times for most of the games they tested.
 
Last edited:
Just a thought:

With all this talk about NVMe SSDs and firmware optimized for Direct Storage, in the case of the PS5 where the user is allowed to install non proprietary NVMe SSDs in the provided slot on the console, the loading times do not seem to differ much regardless of SSD used and also in comparison to the internal SSD. I have always assumed that the PS5 has a custom optimized storage architecture that's somewhat analogous to Direct Storage.

I seem to recall a video from Digital Foundry looking into this and they found that the add-in SSD was actually faster than the internal one in loading times for most of the games they tested.
A. The PS5 has a chip dedicated to the advanced storage features. So, any PCIe 4.0 drive (which has its own DRAM) which meets minimum speed requirements, can be used.

B. The internal drive for the PS5 is, indeed, not the fastest out there. It emphasizes read performance, but even that part isn't the fastest. Its not a 7,000MBps drive. Other drives, like the SN850, MP600 Pro LPX, P41----- also have better sustained writes. And maybe even a bit lower latency. The better read performance and slightly lower latency, is probably why those drives load games a little bit faster.
 
In these benchmarks for texture streaming, which chart would likely apply to DirectStorage gaming? One is for 64 Outstanding, one is for 256 Outstanding, and one is for 1024 Outstanding - and I have no clue what any of those mean.

eperformance4.JPG.8afc6735d2cfc32049b1df69c439be98.jpg


eperformance5.JPG.a1fd813ed63521d2223b612353e4bd4d.jpg


eperformance6.JPG.d49843c791a272acb2106046c6db4784.jpg
 
Back
Top