Corsair MP600 Pro LPX 2 TB

[W1zzard]

The Corsair MP600 Pro LPX is optimized for gaming on Sony PlayStation 5 and the PC. In our review's real-world game load tests we saw a clear lead in load times compared to other flagship SSDs like the Samsung 980 Pro and WD Black NVMe. Thanks to the preinstalled heatsink there's no thermal throttling either.

Show full review

 

[Selaya]

priced to the moon. fucking PCIe 4.0 tax.
at the very least the drive's good, great real-world and synthetic performance so there's that

 

[Chaitanya]

While most other M.2 NVMe SSDs transfer data over the PCI-Express 3.0 x4 interface, the Corsair MP600 Pro LPX connects to the host system over a PCI-Express 4.0 x4 interface, which doubles the theoretical bandwidth.

1st 3rd is not needed to mention its a PCI-e 4x4 drive nor is last 3rd.

 

[nguyen]

Still waiting for MS DirectStorage to justify buying a PCIe gen4 NVMe drive, sadly MS does nothing interesting beside buying game studio these days.

 

[WhoDecidedThat]

We are finally getting 3600 MB/sec sustained writes. Almost 2x scaling when going from 1 TB SSDs to 2 TB SSDs.

 

[Makaveli]

Since the SLC cache is double the size of the (PC version) this drive is actually looks better.

 

[W1zzard]

Still waiting for MS DirectStorage to justify buying a PCIe gen4 NVMe drive

Highly unlikely. They have to make sure it works well on all drives or developers will ignore the technology

 

[Chomiq]

Hold up, is that a €240 controller?

 

[Wooden Law - Black]

We are finally getting 3600 MB/sec sustained writes. Almost 2x scaling when going from 1 TB SSDs to 2 TB SSDs.

Well, the flash is set at 1200 MT/s, not 1600 MT/s as controller's channels support, so imagine how better it could be using the flash at 1600 MT/s...

 

[WhoDecidedThat]

Well, the flash is set at 1200 MT/s, not 1600 MT/s as controller's channels support, so imagine how better it could be using the flash at 1600 MT/s...

What is your source regarding 1200MT/sec?

 

[Wooden Law - Black]

What is your source regarding 1200MT/sec?

In the datasheet it isn't listed as 1600 MT/s but 1200 MT/s, even though it can runs at 1600 MT/s, like B27B (96L TLC).

 

[WhoDecidedThat]

Well, the flash is set at 1200 MT/s, not 1600 MT/s as controller's channels support, so imagine how better it could be using the flash at 1600 MT/s...

Also one more question, how does the 1200 MT/sec have any effect on TLC write speeds? I mean, consider this... there are 8 channels operating at 1200 MT/sec. This is sufficient to write to the NAND dies at 5-6 GB/se (when it is in SLC mode). But in TLC mode, we are getting 3600 MB/sec. Don't these 2 facts imply that the reason sustained write speed is limited to 3600 MB/sec is not because the connection between controller and NAND is slow? Because if that connection was slow, we would also see slower SLC write speed. But we don't see that. So we logically conclude that it is the NAND dies which are the limiting factor in sustained write speed and not the connection speed.

In case you weren't aware. 1200 MT/sec is the connection speed between the NAND chip and the controller. It says absolutely nothing about how fast the controller is at processing the data. It says nothing about the connection between the controller and CPU (PCIe 4.0 x4 in this case). And it certainly says nothing about whether the NAND chip is fast enough to saturate the 1200 MT/sec connection speed.

 

[Wooden Law - Black]

Don't these 2 facts imply that the reason sustained write speed is limited to 3600 MB/sec is not because the connection between controller and NAND is slow? Because if that connection was slow, we would also see slower SLC write speed. But we don't see that. So we logically conclude that it is the NAND dies which are the limiting factor in sustained write speed and not the connection speed.

The theoretical maximum speed of the Phison E18 in sustained write is 3840 MB/s (8 * 1600 MT/s = 12800 MB/s * .9 = 11520 MB/s / 3), not 3600. Keep in mind that the die's speed (I guess around 160 MB/s) is the same, even if the flash runs as 1200 MT/s or 1600 MT/s.

In case you weren't aware. 1200 MT/sec is the connection speed between the NAND chip and the controller. It says absolutely nothing about how fast the controller is at processing the data. It says nothing about the connection between the controller and CPU (PCIe 4.0 x4 in this case). And it certainly says nothing about whether the NAND chip is fast enough to saturate the 1200 MT/sec connection speed.

I know, higher T/s (which refer to I/O speed) can lead less latency between commands and such.

 

[jesdals]

Why testing with a ZEN2 processor?

 

[Soup`]

and WD Black SN850 ($330 without heatsink, $350 without)

I'm guessing this is a typo? It's in the conclusion.

 

[WhoDecidedThat]

3840 MB/s (8 * 1600 MT/s = 12800 MB/s * .9 = 11520 MB/s / 3)

why divide 11520 by 3?

 

[W1zzard]

Why testing with a ZEN2 processor?

Because it's what was released at the time I setup this test suite. It shouldn't make any difference.

For 2022 the plan is to update my tests, and will also use new hardware of course. Still waiting for DirectStorage to come out, so I can design one or two tests around that

I'm guessing this is a typo? It's in the conclusion.

Not a typo, that was the pricing on Newegg yesterday, no idea why the version with heatsink is cheaper

 

[Wooden Law - Black]

why divide 11520 by 3?

TLC = 3-bit (per pages).

 

[WhoDecidedThat]

TLC = 3-bit (per pages).

While it is true that TLC triples the capacity of a NAND chip. It does not 1/3rd the write speed. Or the read speed. It depends on how the NAND chip is built. You cannot just say that 8 * 1600 is the write speed of SLC chip so 8 * 1600 / 3 is the write speed of TLC flash.

I said it before, I'll say it again. 8 * 1600 is the speed of the connection between NAND chips and the controller. It is not the speed of SLC chip or TLC chip or whatever. It is like.. Windows shows your Ethernet connection operates at 1000 Mbps. That is not the speed of your internet because the speed of your internet is determined by your Internet Service Provider.

If you want an even simpler example, when your open the faucet of your shower in your bathroom, how much water comes out of it is determined by how much water your water supply is pumping to your house. If they are facing problems, the water will come slowly. If everything is working fine on their end, you get water at full speed. Your pipes are the same. But the speed/amount of water you get depends on the supply.

Similarly, 1600 MT/sec is the speed of the connection. Eight 1600 MT/sec connections are the 8 pipes. The actual speed (amount of water you get) is determined by the NAND chips (the supply) and how they are made. And going from TLC to SLC does not scale write speeds proportionally. It does not scale read speeds proportionally. There is not a single performance metric that is scaled proportionally.

Here the 970 Evo goes from 2400 MB/sec to 600 MB/sec. In my previous examples, the 980 Pro goes from 4000 MB/sec to 2000 MB/sec. WD's SN850 goes from 5000 MB/sec to 1000 MB/sec. They all scale from SLC to TLC differently because the NAND chips in them are all built differently.

 

[Makaveli]

Why testing with a ZEN2 processor?

Probably because you are going to see very little if any difference at all in I/O test between Zen 2 and Zen 3. And probably the most important reason all of there test data is on that zen 2 system from previous drives. Making any changes to the test platform may invalidate all previous results. And I doubt they want to go back and test everything because of a cpu change.

 

[Wooden Law - Black]

While it is true that TLC triples the capacity of a NAND chip. It does not 1/3rd the write speed. Or the read speed. It depends on how the NAND chip is built. You cannot just say that 8 * 1600 is the write speed of SLC chip so 8 * 1600 / 3 is the write speed of TLC flash.

I never said that my mathematical calculation refers to the speed in SLC, that calculation is purely theoretical and serves to understand the maximum speed that the controller is able to "support" in TLC mode. The calculation of the E18 is correct, but you can do it also with others, for example with the IG5220: 4 * 1600 MT/s = 6400 MB/s * . 9 = 5760 MB/s / 3 = 1920 MB/s, in fact it coincides with the Atom 50, which before does "data folding" does little less than 2000 MB/s.

Similarly, 1600 MT/sec is the speed of the connection. Eight 1600 MT/sec connections are the 8 pipes. The actual speed (amount of water you get) is determined by the NAND chips (the supply) and how they are made. And going from TLC to SLC does not scale write speeds proportionally. It does not scale read speeds proportionally. There is not a single performance metric that is scaled proportionally.

I know, bus speed ("I know, higher T/s (which refer to I/O speed) can lead less latency between commands and such.")...

Furthemore, MT/s' flash = I/O or bus speed while MB/s' flash = program throughput (for example 160 MB/s for Micron B47R).

 

[Deleted member 24505]

Nice drive. I'd like one, got a spare Gen 4 slot. Still, my Sn850 gen 4 is still a pretty good drive for steam, not cheap for 1TB though.

 

[Zareek]

Expensive, and yet it's still cheaper than the M$ version for the Xbox Series X! MSRP $399.99 for the 2TB expansion drive. I will redownload my games as needed until that price is more reasonable or an alternative presents itself.

 

[mama]

How does it compare to the original (non Pro) MP600? Any improvement?

 

[WhoDecidedThat]

serves to understand the maximum speed that the controller is able to "support" in TLC mode

To simplify it, you're saying that if a controller can do a maximum of 300 units/sec write speed in SLC mode, it will be able to do a maximum of 100 units/sec write speed in TLC mode?