Crucial P310 2TB Review

Introduction It never ceases to amaze us the directions the consumer storage marketplace sometimes randomly go in. Take for example the latest craze… 22x30mm M.2 drives that promise to deliver performance levels that leave many a M.2 22x80mm premium offerings from just a few years ago in the dust. While that sounds like a tall tale told around the campfire, it actually is what the new breed of ultra-compact M.2 NVMe (PCIe 4.0) drives like the Crucial P310 series promise to deliver. Furthermore it promises to not only offer sequential speed in excess of 7Gbps, and 1 to 1.2 Million (r/w) IOPS, but also offer up to 2TB of capacity.

For many those promises of capacity and performance and size certainly are eyebrow raising considering one of the best, for its day, Seagate FireCuda 530 2TB’er cannot match that level of performance… and that was a great 2022 SSD we personally used in a lot of builds. Even by 2024 standards fitting all that performance into a 2230 form-factor is noteworthy… as typically 80x60mm was about as small as you could go without sacrificing performance. Yet, thanks in no small part to the combination of Micron’s 2400MT/s Replacement Gate NAND and Phison E27 controller, that is what is now widely available for purchase… albeit with one small catch: the asking price. $210 (USD) and an under a 11 cents per GB of capacity is rather decent, but these days PCIe 4.0 drives (including Crucial’s very own T500 series) go for well under 8 cents. Making the P310 one of the most (relatively) expensive P models to be released to date.

The other potential deal-breaker is, much like the M.2 2242 form-factor only to an even greater extent, the M.2 2230 form-factor is a compromise design. One that prioritizes compactness over everything else. Including performance. Including robustness. Even including price. Put bluntly, when dealing with a PCB that is only 22 millimeters wide and 30 millimeters long there is just not much room for components. This issues are further compounded when these new mighty mites are intended for Steam Deck and RoG Ally sized devices where single sided PCBs are not just a de rigueur but a basic requirement . To be precise, to hit 2TB and the 2230-S dimensions companies like Crucial have had to make some hard choices. First up is there is only room for 2 ICs. One for the controller and one for all the NAND ICs. Thus the use of the DRAM cache-less E27 controller instead of a E25… and the use of QLC instead of TLC NAND.

On the positive side, the new E27 4-channel controller promises to be unlike 99 percent of dram-less controllers that have come and gone over the years. What this means it may not actually be a(n entirely) crippled controller. Especially when paired with (albeit the QLC variant of the) ‘full speed’ FortisFlash 232-Layer Micron Replacement Gate CuA NAND (thus the “N” instead of “B” proceeding the vaunted 58R name). Needless to say, this combination raises a veritable ton of questions (e.g. how good is the performance in the real-world), and a few concerns (e.g. oof that NAND endurance), most of which we will endeavor to answer. Furthermore, thanks to Crucial, we will be able to show you what not just one but two of these little bad boys can do. So let’s dig in.

Features and Specifications

Micron Replacement Gate NAND Primer

For those that have read our previous Solid-State Drive reviews the easiest way to imagine what Micron has done in order to create their ‘Replacement Gate’ NAND is to imagine the melding of typical Charge Trap design with earlier Micron CuA NAND logic layout. Then imagine it bigger, stronger and better than other NAND designs available on the market today. Simply put, it is a new spin on Charge Trap NAND theory and design… and not an entirely new approach to NAND design.

For those that are new(er) to SSD technology and want to know exactly what ‘Replacement Gate Technology’ is… a bit more theory and information is needed. The foundation of all Not-AND (aka ‘NAND’) Solid State storage is a tiny little electrical gate, and a tiny little electrical ‘battery’ cell called a transistor. To change the stored data bit from a 0 to 1, or vice versa, the controller tells the gate to either let the charge ‘out’ (aka ‘discharge the cell’) or store some in the cell. This is called a program operation as it is literally setting the charge state of cell or ‘programming’ it.

To read the NAND cell the controller simply pushes a tiny trickle of power across the gate and reads that cell’s charge. For example, in SLC or single level cell NAND the cell has two states: full and empty. Ironically, since the controller is reading a difference in voltage across the cell, the default ‘empty’ state is usually considered 1 and not a 0. So, an ‘empty’ or discharged state is the same as storing a 1 in that cell, and fully charged is a 0 bit. For old timers, this is why in AS Cleaner or ‘Tony Trim’ type programs (before TRIM was a sure thing) you had to use the ‘FF’ write option… as ‘FF’ is the same as all ones being written to the cells. Since a NAND transistor can hold its charge even when the system is powered off, NAND storage is considered ‘non-volatile storage’. However, just like ‘bit rot’ on a Hard Disk Drive this state is not permanent . Instead, and just like the battery in your turned off phone, if left long enough the power in those tiny transistors will leak out… and turn a 0 into a 1. This is a bad thing from a long-term storage point of view.

With SLC NAND, a cell that is left alone will not lose all that much voltage and is fairly stable for months between power on states. However, as each NAND cell can only store 1 bit it is rather expensive to create say a 1TB Solid State Drive, and scaling up beyond 1TB requires a lot of NAND chips. Even layering NAND cells on top of NAND cells on top of NAND cells the resulting footprint would be rather large. This combination is why Multi-Level Cell NAND was created and doubled a NAND cell’s storage capacity from 1 bit to 2 bits. It did this by adding more voltage resistance points between ‘full’ and ‘empty’. So instead of two states there was now four states: 0, 0.25, 0.5, 0.75, and 1. This certainly increased storage density, and decreased cost of manufacture to hit a given storage capacity, but at the same time made the self-discharging of NAND a greater issue – as lot less of a voltage change can indeed impact two bits of data. This is why MLC is not as stable as SLC and must be refreshed / rewritten more often (aka ‘internal housecleaning’).

As time went by MLC was further supplanted by TLC (Tri-Level Cell) NAND which can store three bits of data per cell. In TLC NAND there are not two, not four, but eight voltage states (for comparison’s sake to hit 4 bits per cell, QLC NAND requires sixteen voltage points). This in turn increases the necessary precision in properly writing and reading a given state of a NAND cell. It also significantly decreases how long a TLC NAND cell will hold this more precise voltage state. This increased precision plays a large role in why TLC NAND is ‘slower’ at write operations than MLC (which is in turn slower than SLC)… and why QLC (and its sixteen votlage points) is slower still.

In original NAND design how a NAND transistor worked was via a 'floating' gate (FG for short). With an FG transistor the gate itself is the conductor of electricity. Thus, in order to change a NAND cell’s state electricity must be pushed through the gate itself. Over time this sudden shock of current running through the floating gate causes miniscule fractures in the crystal lattice that makes up its foundation.

The size of these fractures and their likelihood of occurrence is directly tied to how hot the transistor is; however even a ‘cold’ NAND transistor will do slight damage to its gate every time a charge state change is required. On their own a single fracture will not cause voltage related issues, but over time the culminative effective does create a significant problem. When this happens, the end result is a gate which is incapable of keeping voltage levels inside the transistor stable . This is what a ‘dead’ NAND cell really is – a NAND cell that can no longer be trusted to hold a precise charge.

Worse still, with every reduction in fabrication node size the cell itself not only becomes smaller but the gate itself also becomes smaller. A smaller gate not only has less tolerance for heat but means fewer defects are required before the gate fails. Also, the more precise the voltage requirements the faster accumulative defects become an issue.

In laymen’s terms, a floating gate literally ‘eats itself’ every time it is called upon to change the charge state of its transistor(s). The smaller the gate, or the more precision required, the faster it finishes its meal… and the faster the cells die. NAND manufactures are aware of this issue and include free NAND cells to replace ‘dead’ cells. They are also fully aware that this is only a band-aid solution as eventually there will be no more replacement cells left.

Controller and storage companies are also aware of this issue and include even more ‘replacement cells’ (aka ‘over-provisioning’ and why your 256GB drive may be seen as a 240GB drive) but also use slower / gentler write algorithms to reduce this damage (typically using 8 voltage steps or more). In simplistic terms this means they write to the NAND using less voltage but over a longer period of time. This does reduce cell damage, but at the cost of write performance. To overcome this side effect, the controller will pretend a given chunk of TLC NAND is SLC NAND and only store two charge states instead of eight (or 16 in the case of QLC) per cell. This allows those cells to continue being used long after the amount of damage to the gate would result in random bit changes if eight (or more) states were needed (some estimate the difference at 40K program erasure cycles vs. 5K p/e cycles). This pseudo-SLC is then used for real time writes as a buffer and is called different things by different manufactures. For example, Crucial calls it “Dynamic Write Acceleration”.

Sadly, including more backup NAND cells and / or lowering the voltage applied does nothing for charge coupling. This rare event happens when the fracture in the lattice is close enough to a neighboring NAND cell that the electrons can jump from one cell to another creating a quasi-bridge between the cells. This in turn can randomly impact multiple cells worth of data and do so silently – with the controller only knowing about it when it checks the state of the cells vs. what they were supposed to be. To visualize this imagine sorting your spare change into cups that are touching one another… and a quarter bounces off the lip and ends up in the dime cup. Both cups total amount will be off, and you will not know they are off until you count both cups.

These are all well-known issues. Issues that have been known for decades. While Samsung and Toshiba opted to change the very foundation of how their NAND transistors work… Micron and Intel opted to increase the size and rearrange their cell layout to help reduce them to manageable levels. Put another way, they decided to kick the can down the road. The upside is the CMOS under Array (CuA) design was created. One where the interconnects are much like ‘elevators’ in an apartment building (at the ends of the ‘hallways’) and the logic hardware is under the NAND transistor array (think ‘basement’ in an apartment building). This layout helps (to somewhat) alleviate heat buildup by moving the potential hotter running interconnects closer to the edges of the NAND block.

The downside to typical CuA NAND is twofold. Firstly, it was still FG tech based. Secondly, CuA is not exactly the most efficient layout. It does have a tendency to bottleneck slightly faster than some of the competition… and the deeper the NAND layer depth the more it is likely to occur under sustained extended loads. To help minimize this last issue, Micron has opted for a rather fast ONFI “NV-DDR3 IO” rev 5 bus which can handle 2400 MegaTransfers per second. Thus CuA “Replacement Gate” 3.0 has ‘hallways’ and ‘elevators’ which are 50 percent wider than the previous gen 2.0 RG NAND’s 1600MT/s, which in turn was 50 percent faster than the previous CuA NAND’s 1200MT/s bus.

This in conjunction with increasing the number of hallways each deck of NAND has (aka “planes” aaka logic pathways that allow parallel programming/reading of NAND cells) from four to six means that each sub-block of NAND actually contains fewer cells per sub-block than in previous 176-layer RG NAND. Fewer NAND cells not only means higher performance, but faster multi-block erasures – the bane of NAND SSD performance. This one-two combination of a 50 percent faster interconnect with 50 percent more parallelism means that RG 3.0 NAND is even more capable of alleviating the pressure modern controllers place on the NAND. All while at the same time keeping performance in tip top shape… even under extreme conditions that will become common place with PCIe 5.0 x 4 controllers. After all, the more parallelism, and the faster each interconnect is, the faster the controller can interact with the NAND cells. Which in turns means greater IOPS.

Will that still be the case in the future when PCIe 6.0 doubles the bus and thus doubles the pressure being placed on the NAND? No one knows. What Micron is betting on is that ONFI will to continue to increase the spec’s maximum allowable frequency faster than their CuA’s low-level bottlenecking issue transforms from a theoretical into a real world issue. In the meantime, this new Replacement Gate 3.0 NAND tech is relying upon an incredibly fast bus (by today’s standards) that is more than up to the demands placed upon it.

This is where the other half of the changes that come along with Replacement Gate technology come into play… as Micron literally replaced their NAND cell’s gate technology. Instead of the Float Gate that was used before, and all the issues outlined above that go along with it, Micron has finally admitted that Charge Trap (CT) design is superior. Since Intel is essentially out of the consumer marketplace this makes Float Gate design a dead tree branch tech. One that only CompSci university students will learn about in their history books.

In CT based NAND, the gate is not a conductor , instead it is the insulator . So instead of being responsible for getting power into the gate, it is responsible for keeping it there. This means during write or erase operations less stress is being placed on the gate itself. Since the method of charging/discharging is different the operation can not only be faster (typically about half the time vs FG) but gentler (typically 6 voltage states). The end result is those deadly fractures are less likely to occur, and when they do happen they are usually measured at the electron level. Since they are smaller, they can only drain off the electrons that are 'touching' the fracture. Leaving the rest of the electrons in the cell to control the voltage adequately enough that the controller can still accurately tell what the cell state is.

While yes eventually fractures in the gate’s lattice matrix will multiply to the point that they too also cause the cell to be marked as ‘dead’, the likelihood and severity means a CT based NAND transistor will last noticeably longer than float trap design. Equally important, this in turn not only increases longevity of the NAND cell but also allows for faster – and yet ironically enough still gentler- programming cycles to be implemented on the NAND. Which in turn increases sustained write performance. Charge Trap gates also gain longevity and performance from the switch from planar (aka 2D) NAND designs to 3D designs. Arguably CT NAND gain even more durability as they do not ‘eat themselves’ nearly as fast as FG NAND - so the increased mass means the NAND cells last even longer. When paired with pseudo-SLC algorithms the end result is higher sustained performance, increased durability, and generally a better end-user experience.

Where Micron was starting with both a blank slate and over a decade worth of real-world data provided to them by their competition their design team has been able to focus in fixing a weakness that existing CT designs have – their layout. In classical Charge Trap NAND each cell is basically laid out just as it would be in FG NAND design, with each cell being a sperate and distinct ‘apartment’… with one ‘family’ (or 1/2/3/4 bits of storage depending on NAND type) per ‘apartment’. While this is indeed an improvement in both the durability and performance characteristics; charge coupling is still an issue, performance is still not optimal, and neither is data density.

To understand these issues, imagine the typical high end condo building. While all the condos on a given floor share the hallway, each condo has its own door and rather thick “firewalls” separating one condo from another. In Replacement Gate technology… Micron has clustered cells into groups. With a nonconductive layer of silicon nitride surrounding the insulator gates. This not only allows Micron to consistently increase performance (100% write / 75% read RG 3.0 improvement over RG 2.0 which in turn was 33-35% faster than their previous non-RG CuA NAND) and scale up the number of layers per deck each generation of Replacement Gate NAND offers (232 vs 176), but also reduce the footprint each TB of NAND requires on a drive’s PCB. For this generation it is a 28% reduction in room required to hit a given drives specified capacity. It also promises to be even more resilient to charge coupling issues – further making it even more reliable – when compared to standard CT NAND designs.

Without going into white paper levels of detail all the above is the simplified basics on what makes Replacement Gate Technology ‘tick’ and why Micron has indeed not only achieved parity but arguably surpassed the competition with the release of their next generation 232 Layer CuA RG NAND.

Closer Look

In a very… very “interesting” move the P310 series is not using the new (gray+blue) Crucial Pro shipping container. Instead it is using the same ‘old school’ shipping container found on the P5 Plus… and basically every Crucial SSD from the pre-Pro era. We actually do not have much issue with this older design making a comeback. The older white + blue combination was aesthetically pleasing enough and there was very little wrong with it (for our rather conservative nature). Furthermore there is pretty much no way one can mistake this P310 for a T700, a T05, or even a T500. It just is not possible for anyone who is not color blind. This is a good thing. This is a M.2 SSD that will not securely fit in 90+ percent of systems without an adapter. Even for those that do, the performance is offers is not going to be the same as what a T705 can offer… and yes. You will be extremely angry at yourself if you had purchased what you thought was T705 2TB just to find out it is a P310 2TB drive. Put bluntly, this radical departure, or reversion depending on ones point of view, to this shipping earlier container design is a good thing. It still gets the job done, and does so without needlessly adding to the confusion that was the existing Crucial 2024 lineup.

Moving on. As you can see the interior shipping protection is also pretty much what one would expect from 99 percent of M.2 SSDs. That is to say a two-piece plastic ‘shell’ type affair that snuggly holds the drive and can take a veritable ton of shock, and surprisingly high amount of cutting damage before any of it is transferred on to the P310 itself. Put another way, this is a shipping container that you can trust to get your newly purchased P310 safely home. Regardless if the trip is across town, across the country, or even across an ocean or three. Considering that this is a P and not T branded model we are glad to see Crucial is still taking customer satisfaction so seriously.

Moving on.

Let us rewind and revisit our earlier claim on compatibility… as this is going to be the weak link in any 2230 model – and the P310 is no exception. Especially considering the fact that they have not included an adapter to negate this issue… like Crucial used to do with their ‘short’ 7mm z-height 2.5-inch SATA drives. The issue is simple. 22mm wide by 30mm long, and low-profile single sided (aka M.2 2230-S1) standard is an incredibly small standard. So much so the vast, vast majority of motherboards (desktop and laptop) as well as external USB/Thunderbolt adapters can not secure it. Instead the smallest the designers of said components ever envisioned anyone ever willingly using is the 2242 form-factor. That is because while a mere 12 millimeter difference does not sound like much it is a whopping 28.5 percent reduction is surface area available on the PCB for the SSD team to work with.

In practical terms this means that instead of deciding between either a controller + RAM IC + one NAND IC or a ram-less controller + 2 NAND ICs… designers have to forgo RAM and the second NAND IC. Yes. With only room for a moderate sized controller and one NAND IC there is really very little room to work with. This is why no one foresaw the rise of the 2230 form-factor for anything beyond WiFi + BT controller cards… which the 2230 specification was originally built for. Not storage. Certainly not high performance NVMe storage. Not even SATA storage. Just a piddly litter WiFi+BT “all in one” controller chip like that is found on the rear I/O of most modern motherboards.

Then “Saint Gabe” of the Holy Valve brought out the Steam Deck and made the terrible decision to shave off a couple millimeters from the overall dimensions by using a standard that was never intended for storage… for… you guessed it storage . Yes. Instead of giving us Half-Life 3 they made a modern day GameBo… errr… “Totally Not a GameBoy as Nintendo Would Sue Us So It Is A Hand Held Gaming Device™”. Then ASUS did what they do best and improved upon it with their RoG Ally X.

Some will argue that this new breed of NVMe based M.2 devise was not intended solely for these devices and will point to Raspberry Pi form-factored ‘computer’ systems. After all, the OG M.2 was designed by Intel for their Next Unit of Computing systems and a Rasp’y makes a NUC look downright chunky by comparison. The problem with that is… the HAT+ can accept 2242 drives, and is limited to PCIe 2.0 (3.0 is possible but not recommended as it will corrupt the drive’s data at some point). Thus making the PCIe4.0 devices like the P310 series the new definition of ‘overkill’.

In that vein, PCIe 4.0 devices are also overkill in Gabe’s Gamble as the Steam Deck is PCIe 3.0 only. Thankfully, ASUS at least makes their Ally X a PCIe 4.0 compliant device so the P310 is not technically overkill on all the intended platforms it was made for. Just most. Saying that is rather ironic as PHISON when combined with MICRONs Replacement Gate NAND is pretty much overkill for entry level NUCs. Let alone this generations GameBoy and toy like systems. Yes. With sequential read speed of 7GBps it literally is the fastest one can get on the PCIe 4.0 bus… as such it can actually whistle the tune of being a secondary D drive / “game drive” inside a desktop (or laptop) gaming system; alternatively with its max power draw rating well below that of even the Crucial T500 series it also can potentially make for a wonderful external storage solution when paired with a proper USB (or ThunderBolt) based adapter.

So before we dig into the nitty gritty of what the P310 is, lets go over the easiest ways to adapt this tiny drive to work with M.2 2242, 2260, 2280 slots… as you probably will have to do so if you are buying for anything other than those tiny systems. The cheapest, easiest, but least effective way is to pick up a package of ‘Cupboard Stoppers’ or similar clear rubber bumpers with integrated sticky tape. As long as they are 2.25mm to 2.5mm tall you are pretty much good to go (2mm is the shortest we would go, 2.75 is about the tallest). Place the 310 in your free M.2 slot and eyeball where the bumper needs to go. Stick. Screw in the M.2 screw (it will take a bit of effort but they can be considered ‘self tapping’ when dealing with rubber) to secure the drive and enjoy. This method is best for those who have a motherboard with integrated M.2 heatsinks… as the heatsink will keep the P310 secured long term.

The next step up from the ghetto chic option is to 3D print an adapter. Head on over to ThingVerse (or your preferred STL sharing site of choice) type in 2230 or NGFF or 2230 to 2242 adapter and pick the one you like. They are all basically the same. Print, use a spare M.2 screw to secure it to the P310… and off you go. If you don’t own a 3D printer, 12 Gauge (Hot and Cold Roll) Mild Steel is a nominal 2.656mm thick… but a more optimal choice is either 11 Gauge Aluminum or Brass sheeting which is a nominal 2.305mm thick. In a pinch 10 Gauge Copper is a nominal 2.588mm and is also ‘good enough’. Avoid Stainless, and Galvanized options as they will either well be over (2.77) or under (1.98) the nominal thickness you want. Just remember all metal gauge standards have a baked in allowance of plus or minus 5% thickness deviation… thus the ‘nominal thickness’ in their specifications. So “when in doubt, mic it out” before you buy anything.

Either way, cut a piece 22mm wide by 55mm long of whatever you find in your spare metal bin(s). Drill a hole in the center (11mm from edge) at each end (albeit drill bit size plus ~1mm in from edge) using a 5.5 to 6mm (or 7/32 to 15/64 in freedom units) drill bit on one end (aka the ‘2280’ socket end) and a 1.6mm (or a #52 if you like freedom units more than euro units) drill bit on the other (aka the 2230 adapter end), tap the 1.6mm hole end with a CM2 / M2-0.4 tap (in a pinch use a bog standard M2 screw to ‘tap’ the hole… as it is thin, rather malleable metal we are talking about and even a Chinesium steel screw will cut threads into it). Place the P310 overlapping the threaded end and using a spare M2 screw… use your fancy homemade adapter. If you were off on the 80 end… redrill. All you care about, is it fitting “close enough” for the screw to grip it and secure it to the M2 2280 mounting bracket length.

If all that sounds like too much hassle you have a couple decent ‘off the shelf’ options. The first is the Lenovo M.2 adapter that is made from stamped metal. Head on over to FleaBay and drop the cash needed for one. If you are not made of money, or do not have a Traumatic Brain Injury (you will understand once you check the prices of the Leno’ adapter), go to either Fast AliExpress (AKA Amazon) or Slow AliExpress (aka AliExpress… or for some “Temu”). Type in “NGFF NVME M2 Adapter” and pick one that does 30 to 42/60/80. They come in two main types. The ‘cut away’ PCB material style shown above… and the 32 to 80 only ones made with “best quality” Chinesium alloy. Either work fine. Fast AliExpress will have more of the PCB style, and AliExpress and/or Temu will have both in darn near any configuration and material you want. Expect to spend anywhere from two Canuck bucks to ten bucks (depending on where you get them) per adapter (plus shipping).

So with that problem solved let’s peel the sticker off the P310 and see what makes it tick.

Let us make something very, very clear. Crucial is up front about what components each model use and if it changes… the model name changes. This is rare these days as some others do not change the name, just the components. As such, if you are thinking about a 1TB’er P310 the only differences will be in the capacity of the NAND IC and the size of its pSLC buffer. Nothing else.

So let’s start with the NAND. We personally are not big fans of QLC. Even in external storage ‘thumb drives’ we try and avoid them for 99 out of 100 cases. Microns latest and greatest 58R series of ‘FortisFlash’ Replacement Gate NAND is pretty much the exception to our personal rule. Based upon Microns latest generation Replacement Gate NAND technology these 4-bits per cell NAND is actually rather decent. It is obviously is going to be much, much more fragile in QLC mode than its brother B58R TLC NAND variants. It is going to be slower at writes – in QLC mode - than its brother B58R TLC NAND variants. It is going to be slower on average at reads than its brother B58R TLC NAND.

Most of this is not going to matter for the average P310 buyer. First and foremost is the 2TB variant may indeed only have a single NAND IC on it. However, this is a 16 die pack NAND IC. Meaning 16 ‘blocks’ of 232-layer NAND piled one on top of another. All of which can be somewhat accessed at the same time. We say somewhat as the CMOS Under Array (CuA) controller is going to be the bottleneck and when the IO requests get hot and heavy… 2/4/8/16 separate NAND ICs worth of this 58R will be faster. In typical home user scenarios its actually not that big of a deal… as the controller is only 4 channels and each channel is rated for 3600MT/s. Meaning that two layers per channel is all that it takes to satisfy the controller… as this B58R is trucking along at 2400MT/s. Not 22000. Certainly not 1600. This speed boost over the last gen P model does help overall performance, and does minimize the bottlenecking from 1 NAND IC. This in turn means the 8 die pack 1TB P310 variant is also going to fully stock the controllers channels with NAND. Making this one of the few times where the 1TB vs 2TB performance differences is not going to be all that big in the real world.

With all that said… we still would opt for TLC over QLC 99 times out of 100. Even if its N58R QLC RG NAND versus last generation Kioxia/Samsung/etc. NAND. It is just that this QLC NAND is showing how good the underlying 232-layer Replacement Gate technology is. In a few more generations it might actually be good enough… to be good enough for more than edge cases. Which the P310 and its design envelope (somewhat) is.

We can say this is one of the few edge cases where QLC NAND is (somewhat) acceptable because of two things. The first is Crucial is using a massive dynamic/floating pseudo-SLC buffer. One that slurps up a whopping eighty percent of the empty space on the drive for its pSLC write buffer. This large floating buffer size means that when empty the drive has a 400GiB pSLC write buffer. When half full it still has 200GiB pSLC write buffer. Even when at 75 percent capacity that pSCL write buffer is a bigboi 100GiB. In simplistic terms, in pSLC write mode (and read for that matter) the difference between QLC and TLC is minimal. Hell the difference between QLC and SLC in negligible as those NAND cells are being used to hold only one bit (thus the pseudo-SLC moniker applied to this write buffer configuration).

Quite literally it is only when the pSLC buffer is saturated that performance falls off a cliff. Thankfully PHISON is well… well aware of that and for a couple generations now has implemented what they call SmartFlush. In a nutshell, the firmware algorithms monitor the write requests and before the pSLC buffer is full and it predicts the writes are going to even come close to saturation it starts the emergency cleaning procedure to try and keep ahead of the write demands. In theory, this means the average home user will never see anything but pSLC write performance. That is the theory.

In reality, things are not so rosy. Yes, in a typical 4 to 8 NAND ICs device the controller can spread out the emergency write requests enough so as to not noticeably ‘rob’ real-time write performance from the drive. When talking about a drive with only NAND IC, like all 2230 form-factor SSDs, all the I/O requests go through the same CuA controller baked into the lone NAND IC. This bottlenecking of said NAND ICs communication with the controller will cause the P310 to recover slower than if there was two (which in turn would recover slower than 4, and this trend continues until the NAND ICs equals the number of channels at which point the differences become academic). This means that this E27t based drive has to slow things down earlier. So when close to saturation expect performance to drop, but not as badly as if it was saturated. Thus giving the P310 three separate and distinct write performance levels. As such keep the sequential write requests to well under the pSLC limit (75 percent of it would be a safe rule of thumb) and buyers will always be fine and never even see the second, let alone third, write performance tier.

For most drives, this hand waving away a rather large issue would be a disingenuous. Yes. It is highly unlikely the average home user will push more than 75GiB at a time (i.e. 75% of the pSLC capacity when the drive is 75 percent full). It however is possible and should be accounted for in a given design. Which it is in the P310. Just poorly. Thankfully… the P310 is not really meant to be written too all that often. It is meant to be mostly an uber fast read storage device. One where you push the games too it once… and then use it to push the data to the systems iGPU’s RAM buffer every time you use the system. So the highly rare slow down will be just that. Highly rare. With that said, this potential issue due to the nature of QLC NAND is what makes the P310 even more niche than just its use of a tiny form-factor. Which is a shame as the controller is rather good, and this NAND is actually better than a lot of what it being used even in 2240 drives.

This segues nicely into the controller. First and foremost is that some like to consider the Phison E27 to be nothing more than the continuation / next generation of the E21t. The E21T was also a DRAM-less, PCIe 4.0 controller. It too was a 4-channel design that could handle TLC and QLC NAND. It too also used an ARM based controller built on the older 12nm fab process (and not newer 7nm like the E31). This is all true. However it is being a bit too harsh on the E27T. This is why others will argue that it is a cut down E25. We like to fall somewhere between these two extremes.

Like the E21t and E25 the new E27t is part of PHISON’s “Single CPU Architecture” modern design. One that does away with big.Little complexity of earlier generation Phison controllers. Instead it uses only ‘big’ ARM 32-bit Cortex-R5 cores with some running faster than others. To be precise the single core (instead of 2 like in the E25) responsible for real time IO requests clocks in at an “up to” 1.2GHz whereas the two R5s handling the internal housecleaning / ECC / emergency flash cleaning /etc. run at “up to” 600Mhz (instead of 800 to possibly 1GHz like the in the E25). Contrast this 3-core performance with the E21t’s 1GHz on the main plus typically under 600MHz on the secondary cores and there is merit to both sides arguments for which design the E27T mainly takes after.

The reason we fall in-between is because the E25 four channels are rated for 3600MT/s and not 1600MT/s like the E21t, but like the E21t it is NVMe 1.4 and not NVMe 2.0 compliant… which the E25 is . Regardless of which way you decide, beyond those major differences, the E27t is a Phison controller through and through. Insanely robust, fifth generation LDPC ECC engine; internal RAID configuration to the NAND to further make uncorrectable read errors a thing of rarity… even AES 128 and 256bit encryption capabilities. All of which add up to one surprisingly peppy drive. Even if it relies upon the system bus for its caching capabilities.

Also in the win column, the P310 is a downright cool running drive. PCIe 4.0 NVMe M.2 SSDs are when cooling really started to become a concern as they did run noticeably hotter than PCIe 3.0 models. In this regards the P310 runs rather cool. Cool enough that even in nekkid configurations the internal thermal limiting protocols are able to keep up(!) and keep the drive from suddenly shutting down (aka a SCRAM event where the drive has to immediately shutdown lest it start killing the NAND). Which is ironic as all the write requests are being pushed to a single 16 tall stack of NAND IC goodness… which are notorious for running hotter than 8 dies packs (which in turn run hotter than 4 dies pack ICs). The reason for this is the entire board max power draw is set to 5.1watts and in our testing the average was well under 4 watts with torture tests pushing this average up and typical ‘day to day’ tasks pushing it down. For the average home user expect about 3 to 3.5watts of power (and thus heat creation) when active IO requests are happening. Compare and contrast that with even the T500’s 6.2 max and high 4’s average… and the P310 runs cooler. Noticeably cooler. Cool enough you do not have to worry about it shake’n’baking itself. Overall, while we would be much happier to have seen a B58R TLC RG NAND and not N58R QLC RG NAND IC the P310 is a marvel of engineering. One that does expand what is possible these days, and one that is pretty much the very definition of overkill for a Rog Alley X, Steam Deck… or any other ultra-ultra-compact computer system.

Testing Methodology The reasons to purchase a high-performance storage device are as varied as the consumers who purchase them. For some application load time is the deciding factor in their purchasing decision. For others it is Virtualized Operating System performance. For others it is game load times. For others still, it is to shave a .1 of second off their synthetic test scores so as to climb ever higher in the online rankings that various sites host.

With such a wide variety of criteria it behooves us to use as wide and varied a list of testing protocols as possible. As such we have used a blend of synthetic and real-world benchmarks, as well as custom recorded real-world game benchmarking.

For game load times testing we have chosen games which are clearly not Internet, GPU, RAM or CPU bottlenecked. The system was cold restarted in between each load to make sure no data saved in the GPU’s ram could influence load times. As with all tests all games were run four times and an average rounded the nearest second is recorded as the official result.

For the Call of Duty: Ghosts we timed the level load time for the map called Into the Deep in the single player game. This level was chosen as it contains a large map and loads a lot of data from the storage device to the GPU. For Borderlands 2 we have recorded the time it takes to fast travel from Sanctuary map to Oasis. Oasis is the beginning of the Captain Scarlett and Her Pirate's Booty DLC. This was chosen as it not only is a special level not included in the original Borderlands 2 game and thus contains additional data which should slow the load of this special map.

For Saints Row 3 (labelled ‘SR3’ in the chart) we have recorded the load time from a specific save game. The game was saved just at the end of the ‘Shaundi saved’ version of “Three Way Ending” but before the cut scene. This was chosen as the game not only loads the Three Way Ending data but also the next level “Gangsta’s in Space” DLC level.

For synthetic tests we used a combination of the ATTO Disk Benchmark, AS-SSD, Crystal Disk Benchmark, HDTach, HD Tune, and IOMeter.

For real world (non-gaming orientated) testing we have used OS startup, Adobe Photoshop, VMWare, and data transfer times. For data transfer we timed how long a single 100GB rar file took to copy to and then from the devices. We also used 30gb of small files (from 100kb to 200MB) with a total 72,000 files in 2400 subfolders.

All tests were run 4 times and average results are represented.

Sequential Performance Read Bandwidth For this benchmark, HD Tune Pro was used. It shows the potential read speed which you are likely to experience with these storage devices. We don’t put much stock in Burst speed readings and thus we no longer included it. The most important number is the Average Speed number. This number will tell you what to expect from a given drive in normal, day to day operations. The higher the average the faster your entire system will seem.

Write Performance For this benchmark HD Tune Pro was used. To run the write benchmark on a drive, you must first remove all partitions from that drive and then and only then will it allow you to run this test. Unlike some other benchmarking utilities the HD Tune Pro writes across the full area of the drive, thus it easily shows any weakness a drive may have.

ATTO Disk Benchmark ATTO Disk Benchmark The ATTO disk benchmark tests the drives read and write speeds using gradually larger size files. For these tests, the ATTO program was set to run from its smallest to largest value (.5KB to 8192KB) and the total length was set to 256MB. The test program then spits out an extrapolated performance figure in megabytes per second.

Crystal Diskmark & AS-SSD Crystal DiskMark Crystal DiskMark is designed to quickly test the performance of your hard drives. Currently, the program allows to measure sequential and random read/write speeds; and allows you to set the number of tests iterations to run. We left the number of tests at 5 and size at 100MB.

AS-SSD AS-SSD is designed to quickly test the performance of your drives. Currently, the program allows to measure sequential and small 4K read/write speeds as well as 4K file speed at a queue depth of 6. While its primary goal is to accurately test Solid State Drives, it does equally well on all storage mediums it just takes longer to run each test as each test reads or writes 1GB of data.

IOMeter IOMETER: File Server Test To test each drive we ran 6 test runs per device (1,4,16,64,128,256 queue depth) each test having 8 parts, each part lasting 10 min w/ an additional 20 second ramp up. The 6 subparts were set to run 100% random, 75% read 25% write; testing 512b, 4k,8k,16k,32k,64k size chunks of data. When each test is finished IOMeter spits out a report, in that reports each of the 6 subtests are given a score in I/Os per second. We then take these 8 numbers add them together and divide by 6. This gives us an average score for that particular queue depth that is heavily weighted for file server usage.

Boot time and Data Transfer System Boot Time When it comes to hard drive performance there is one area that even the most oblivious user notices: how long it takes to load the Operating System. While all the other tests were run with a Windows 7 operating system, this particular test uses another older test bed's “day to day” OS (copied over to our new testbed) which has accumulated a lot of crud over the months from installs and removals. We chose the Anti-Virus splash screen as our finish line as it is the last program to be loaded on start up.

Real World Data Transfers No matter how good a synthetic benchmark like IOMeter is, it cannot really tell you how your hard drive will perform in “real world” situations. To this end we will be running timed data transfers to give you a general idea of how its performance relates to real life use. To help replicate worse case scenarios we will transfer a 30.00GB contiguous file, then a folder containing 1200 subfolders with a total 36,000 files varying in length from 200mb to 100kb (15.00 GB total).

Testing will include transfer to and transferring from the devices, using MS RichCopy and logging the performance of the drive. Here is what we found.

Adobe & VMWare ADOBE Photoshop CC LOAD TIME

Photoshop is a notoriously slow loading program under the best of circumstances, and while the latest version is actually pretty decent, when you add in a bunch of extra brushes and the such you get a really great torture test which can bring even the best of the best to their knees. Let’s see how our review unit faired in the Adobe crucible!

VMWare Load Time Performance VMWare is a powerful application which allows users the ability to run a virtualized Operating System from inside their main OS. This program is not only processor and RAM intensive but puts significant load on the storage subsystem with deep, heavy read/write IO requests to the drive. To help give you a general idea on the performance improvements from running a RAM Drive we have timed the time it takes to load XP SP3 from inside VMWare.

Game Load Time Borderlands 2 Load time For Borderlands 2 we have recorded the time it takes to fast travel from Sanctuary map to Oasis. Oasis is the beginning of the Captain Scarlett and Her Pirate's Booty DLC. This was chosen as it not only is a special level not included in the original Borderlands 2 game and thus contains additional data which should slow the load of this special map.

CoD: Ghosts Load time For the Call of Duty: Ghosts we timed the level load time for the map called Into the Deep in the single player game. This level was chosen as it contains a large map and loads a lot of data from the storage device to the GPU.

Saints Row 3 Load time For Saints Row 3 (labelled ‘SR3’ in the chart) we have recorded the load time from a specific save game. The game was saved just at the end of the ‘Shaundi saved’ version of “Three Way Ending” but before the cut scene. This was chosen as the game not only loads the Three Way Ending data but also the next level “Gangsta’s in Space” DLC level.

Partial and Full Drive Performance Partial and Full Drive Performance While it is important to know how a drive will perform under optimal conditions, more realistic scenarios are just as important. Knowing how drive will behave when partially or even nearly full than when it is empty is very important information to know. To quickly and accurately show this crucial information we have first filled the drive to 50% capacity and re-tested using both synthetic and real world tests. After the completion of this we then re-test at 75% and 90% of full capacity.

Synthetic Test Results

For our synthetic testing we have opted for our standard VMWare test.

Real World Results

For a real world application we have opted for our standard Adobe and Saints Row 3 tests.

Score Card and Summary Performance: 34 out of 40

We have to ensure that people do not get confused when we say the P310 offers very good bordering on excellent performance. It does. Relative to what one could expect from its form-factor. A cheaper T500 2TB will outperform it. Even a 1TB T500 will. The P310 simply brings the lagging ‘P’ series into the realm of reasonable for the year of our lord 2024. Nothing more. Nothing less. So do not get it twisted and think this modern PCIe 4, NVMe 1.4 2230 can outperform a similarly priced PCIe 4 NVMe 1.4 2280… let alone compete against a similarly priced PCIe 5.0 NVMe 2.0 M.2 2280. It can’t. Especially when most of those bigger drive will be rocking modern TLC NAND and not modern QLC NAND.

Technological Innovation: 18 out of 20

Is this the first 2230 to use the E27t controller? No. A couple others including PHISON themselves have done it. It however is the first E27t + QLC NAND 2230 that does not out and out suck…eggs. Instead, the P310 2TB variant is extremely good at the task it was designed to do: offer PCIe 4.0 bottlenecking levels of read, and darn near PCIe 4.0 bottlenecking write performance, in a tiny package that veritably sips power. Now if it only had used the TLC variant and not QLC variant and we would be fully satisfied.

Build Quality & Warranty: 15 out of 20

There is not much to complain about here – beyond the NAND – as this is a very competently designed drive with as much care taken to ensure real world performance is top notch and stays top notch. With that said… there is only so much one can do with QLC NAND. You can not make a silk purse from a sow’s ear. So while Crucial has come close, at the end of the day the build quality rating is lowered because of the known fragility, and native performance, of QLC NAND. Even the best QLC NAND we have seen is no exception from this hard and fast rule.

Value: 17 out of 20

As with performance rating, it is all relative. Yes, one can get a lot more bang for their buck opting for a modern M.2 using the standard 2280 formfactor. Good luck fitting even a 2242 SSD into a slot meant for 22x30mm drives. That is the crux of the value issue with this nearly 11 cents per GB SSD. It is insanely fast for its class. Hell its downright overkill for 99 percent of the devices that can only accommodate 2230 M.2 drives. Its just that outside of its niche the performance vs. price disparity become so prevalent as to make it one of the worst value P modes Crucial has ever offered. That is the downside to specialization and razor sharp optimization… as you lose flexibility as you gain in whatever category you have optimized for. As such you can either consider this a 17 out of 20 (as we have nuked 3 full points for the use of QLC)… or you can consider it a 7 out of 20. It really will just depend on your point of view.

Final Score: 84 out of 100

The P310 is easily the most task optimized SSD Crucial has released to date. The upside to this laser like focus is this is one of the better 2230s available today… and it still is the best P series SSD Crucial has released. The downside is the reduction is broad ‘mass’ appeal. For most outside of those who need a 2230 formfactor drive there are veritable ton of better, cheaper, faster options. As such the P310 2TB is the very epitome of the Chinese saying about Dragon Tears. You will either find the P310 to be one sweeeeet deal that offers insane levels of overkill, or you will find it so bitter as to be unpalatable. It really will all come down to precisely what need(s) you are trying to satisfy. Make no mistake if you do want a drive that will rock a Steam Deck / RoG Ally / Rassberry Pi… even entry level NUC’s world the P310 2TB is a very good to great option. It is only when you try and fit it into a more… reasonable sized computing device does its limitations come to the foreground. So chose wisely.

Closing Statement Final Score: 84 out of 100

Let us start with the large elephant in the room. No. The 2230-S form-factor is not going to make large inroads into the desktop or even sensible-sized portable PC market. It is a compromise design that is based on itself a compromise standard. After all. What the M.2 standard was originally called was the Next Generation Form Factor and was made specifically to fit inside the (then upcoming) itty bitty Intel NUC systems’ form-factor. Swap out ‘NUC’ for the even smaller Steam Deck and you begin to realize how many (more) compromises had to be made to even get a PCIe NVMe SSD to fit! This compounding of compromises does impact real-world performance… and when in doubt a 2024 M.2 2242 will outperform a similarly priced M.2 2230; which in turn will be outperformed by a M.2 2060, which in turn will be outperformed by a M.2 2280 sized model.

The other large, and pink… and oh so angry, elephant in the room is the use of QLC NAND. We are not fond lovers of QLC. Sixteen separate and distinct voltage points per cell is pushing the boundaries of not just incredulity but actual sanity . That is why we are much more comfortable with… two(SLC) voltage points. MLC and its four states was a step in the wrong direction, and eight (TLC) is about at the limits of our personal tolerances. It pushes things because each doubling of voltage points dramatically decreases r/w performance, reliability, and overall durability. Furthermore with sixteen die pack, 1Tbit per die, B58R TLC NAND well and truly proven there was no burning need to use a (single sixteen die pack, 1Tbit per die, N68R) QLC NAND IC. Or at least no burning need beyond build cost. The tray price of the 16Tbit variant of “B58R” TLC (e.g. NY266 and NY265) is quite literally over 50% higher than the QLC version used here inside the 2TB P310 SSD. Considering the price of the P310 is already above that of a similarly capacity 2280 form-factor’ed Crucial T500 we understand the reasoning behind making the P310 a showcase model for their QLC variant of their 232-layer R.G. NAND… and not a showcase model for precisely how bloody fast the TLC B58R NAND truly is. We just do not like it when choices are made solely based upon build price and nothing else.

With all that said. We sincerely… truly… deeply wish this controller and this QLC NAND was around back in 2022. It is simply incredible what levels of performance one can now get inside such a small package. It is in fact ‘good enough’ to have changed our opinion on DRAM-less based SSDs from ‘actively hostile to their very existence’ to ‘nearly neutral (in our animosity)’. It has done so because for the first time we can actually see actual use case scenarios for this type of controller. Typical scenarios where its lack of on the PCB RAM cache buffer will not (typically) impact real world performance ehough to justify the added cost of an E25 based model. So much so we generally think that a 2280 form-factored ‘P510’ could easily find its way into Crucial’s existing lineup and give the mainstream “P” options some much overdue love… as the P310 is simply too niche for most mainstream buyers.

With all that taken care of, let us turn our attention to what this P310 is intended to do. It is intended to boost your Steam Deck, any other computing system that relies upon the 2230 form-factor (and even your ultra-ultra-portable external storage device) into the stratosphere. To be perfectly candid, even i3 (or AMD equivalent) ‘NUC like’ form-factored systems could easily do a lot worse than rocking out with one of these little beasties installed in them. Yes. It relies upon QLC NAND. Yes, this means it relies upon its p-SLC buffer to keep write performance above Hard Disk Drive levels. Yes. It lacks a cache buffer causing it to stutter if pushed too hard. None of this matters as it does offer beast mode read performance and decent to excellent write performance in realistic real-world scenarios. It is only in highly unusual and / or synthetic test case scenarios that the QLC NAND and DRAM-Less issues go from theoretical to real. Put bluntly, 7GBps sequential read and over 6GBps sequential write levels are nothing to sneeze at. 1 Million read and 1.2 Million I/O Per Second write performance levels are also nothing to turn your nose up at. While it is true the intended devices of the P310 design will never… ever take full advantage of the P310s performance, overkill is just another way of saying ‘buy once, cry once’. Hell, many a portable ThunderBolt (or USB 4) based external storage device would be green with envy at what this tiny P310 can do. Better still, many a 2-year old mainstream system’s M2 2280 drive could do with an upgrade to that level of performance. All in a package so small that many a mITX system will require an adapter to secure it to the free M.2 port.

This major performance upgrade to, lets face it, low performance computing devices can indeed be even further enhanced via the use of two of these bad boys. Yes, this does exclude the Steam Deck, Raspberry Pi (and their laughably bad, and overpriced, HAT+ add-on board) and similar devices… but there are a lot of “mini-PCs” that do have two and sometimes even more M.2 slots just begging to be used. In these devices the noticeable increase in performance (and durability) RAID offers, combined with their low power and heat output, turns what is basically a great ‘games drive’ model into one that can indeed be used as a dual purpose OS + Games + Everything Else setup. Also something we thought we would never say about a 2242 let alone 2230 device. Just understand that even then compromises are being made, and some of these compromises may not be applicable, let alone appropriate, when it comes to your device. So when in doubt… always get the largest form-factor storage drive you can fit into your system. With all that said, we can see a lot of uses for this P310 series and genuinely look forward to higher numbered P-models. Maybe even a P model than combines this controller with TLC/B58R 2400MT/s NAND. Until that day, the P310 does indeed open up a world of possibilities and turn what was once a minor niche filled with “lowered expectations” into one where actually good performance is now not just a possibility… but a reality. Brave New World indeed.