Phison E31 PCIe 5.0 SSD Engineering Sample Preview

[W1zzard]

The Phison E31T controller brings fresh wind to the PCIe 5.0 SSD space. It's a value-oriented solution that boasts impressive energy efficiency. This enables 10 GB/s+ solid-state drives that can run without a fan-cooled thermal solution, which was one of the biggest issues people had with existing designs.

Show full review

 

[Onasi]

Oof. Okay, this IS significantly better than the E26, but isn’t convincing me that Gen 5 drives are a good idea for anything other than niche cases. Still needs a heatsink to avoid throttling, still has questionable idle power usage on desktop, still not really practically better in day to day usage compared to Gen 4 drives.
I mean, theoretically, this cooooould be improved with firmware updates? Maybe? After all, this is just an ES. But the problem remains that where you want Gen 5 speeds is actually the area where all the weaknesses rear their ugly heads.

 

[Ferrum Master]

Size limited to 2TB? It is pointless...

EDIT.
It does theoretically support larger sizes. False alarm.

 

[bug]

Throttling while reading is always a big no-no for a SSD. But breaking the 100MB/s for random 4k reads is something you don't see every day.

 

[Chaitanya]

Even with improvment to power consumption its still a hot controller.

 

[Asni]

Even with improvment to power consumption its still a hot controller.

We already experience that on cpus.
7nm is the first dense node that confronts us with an inherent limit: the thermal conductivity of silicon. The denser the node, the slower the heat is dissipated and the chip overheats despite being efficient (5w peak is half of the old E26 12nm based SSDs).

And this SSD does not even reach full 14GB/s potential (power consumption isn't linear).
SM2508 is a TMSC 6nm based controller which is drawing 3.5w, 7w for the entire disk.

Silicon Motion Demos Low-Power PCie 5.0 SSD Controller: SM2508

www.anandtech.com

 

[bug]

We already experience that on cpus.
7nm is the first dense node that confronts us with an inherent limit: the thermal conductivity of silicon. The denser the node, the slower the heat is dissipated and the chip overheats despite being efficient (5w peak is half of the old E26 12nm based SSDs).

I'm pretty sure density has zero impact on how fast heat is dissipated. It does have an impact on how much heat is generated by a given area.

The silver lining is, fi you don't absolutely must have the sequential speeds PCIe5 offers, you're absolutely fine with a PCIe4 drive.

 

[Asni]

I'm pretty sure density has zero impact on how fast heat is dissipated. It does have an impact on how much heat is generated by a given area.

The silver lining is, fi you don't absolutely must have the sequential speeds PCIe5 offers, you're absolutely fine with a PCIe4 drive.

Let me rephrase it: "the increasing amount of heat generated by a given area", due to the increasing node density, exposes the thermal conductivity limit of the material.
The thermal equilibrium is higher because the chip cannot transfer the heat quickly enough to the cooler compared to the past.

 

[TristanX]

great drive for prices just above gen4 drives. Hopefully finished variants will get lower idle power consumption. I think retail SSD will come with much smaller and manageable radiators, the same as with gen4 drives, so RW speed will be high.
'In the past weeks we've been testing new controller designs from Innogrit and Silicon Motion', where is SM2508 your review ?

 

[Chrispy_]

Yay, another lower-power Gen5 controller.

Honestly, consumers don't need Gen5 - we barely even need Gen4 for most things - but one of the biggest hurdles to Gen5 adoption has been a lack of sensibly-priced drives that run at a sensible temperature in a sensible form-factor.

The old E26 is probably the biggest culprit for Gen5's problems - it was the first Gen5 controller and it's simply too hot for the intended form factor. IMO it was a mistake to launch it, but I can understand why Phison did - partly because there are always some people with deep pockets willing to early-adopt high-end stuff whether it's good or bad, and partly because of the old chicken-and-egg problem that in order to refine a design you need prior experience. The E26 was necessary to highlight the problems most in need of solving.

 

[W1zzard]

SM2508

wait for my review before getting excited

 

[Chrispy_]

wait for my review before getting excited

FORESHADOWING!

Even with improvment to power consumption its still a hot controller.

I think transfer rate per Watt in these reviews is a fantastic metric though, it's only hot if you get blistering speeds out of it.

Another way to look at it is that faster drives getting hotter is okay if they finish faster and then cool down earlier. Think of it like Ryzen's "rush to idle" mentality - better to boost to the moon to get the workload finished ASAP so you can go back to sleep.

 

[Asni]

wait for my review before getting excited

MX500 is still my pick nowadays. So, i don't really care!

 

[bug]

MX500 is still my pick nowadays. So, i don't really care!

Arguably the premier workhorse since the advent of SSDs.

 

[AnarchoPrimitiv]

Throttling while reading is always a big no-no for a SSD. But breaking the 100MB/s for random 4k reads is something you don't see every day.

I think we'd be better served with an All SLC 4.0x4 drive instead of a 5.0 drive, especially for random, low queue depth performance.

I'm a storage enthusiast, but have very little interest in the 5 0 drives (I currently have an Intel 905p optane as my OS drive, bought it on ebay a few months ago), but I would jump all over the chance to buy a consumer all SLC drive, even if it was only 500GB just to use as an OS drive (would love to even throw some in my home server/nas)....but I guess these companies have probably looked at the Financials and have decided it's not profitable.

 

[Wirko]

Arguably the premier workhorse since the advent of SSDs.

In horse speak, a coldblood.

 

[W1zzard]

MX500 is still my pick nowadays

I have like a dozen of those in various machines, too

 

[_roman_]

Such speeds is just unacceptable.
Comparing with existing PCIE 4.0 drives I lost interest in this product.

Once 1.1 TB have been written, there's another drop, to around 700 MB/s.

 

[Chrispy_]

MX500 is still my pick nowadays. So, I don't really care!

I have like a dozen of those in various machines, too

I must have installed over a hundred MX500 SSDs, (most of them 2.5") but they're always SATA, right?

Don't get me wrong, they're decent SATA drives but aren't we at the point now where trying to buy a SATA drive instead of an NVMe drive is a fool's errand? IIRC they were completely outclassed by the SN550 and at a lower price something like half a decade ago?

 

[mechtech]

The NAND are 1GB per chip...........should be able to get 8 on there. Where's my 8TB NVMe for the cost of four 2TB NVMe minus 3 ram chips, minus 3 controllers, minus 3 pcbs etc. etc. so $500CAD.

 

[Asni]

I must have installed over a hundred MX500 SSDs, (most of them 2.5") but they're always SATA, right?

Don't get me wrong, they're decent SATA drives but aren't we at the point now where trying to buy a SATA drive instead of an NVMe drive is a fool's errand? IIRC they were completely outclassed by the SN550 and at a lower price something like half a decade ago?

Out of 1000 consumers buying pci-ex drives, 990 are not performing workloads where it makes any meaningful difference.

Recently many manufacturers are replacing high-end components with chinese cheap stuff keeping the same name and price. Aside from some reliable products (Crucial T500, SK Hynix P41, Samsung 980 and 990Pro and a few others disks, mainly because these brands manufacture their own components), buying a Crucial MX500 is always a safe and smart move (it sounds like a promo, i'm sorry).

 

[Wirko]

Still needs a heatsink to avoid throttling

Throttling while reading is always a big no-no for a SSD.

You overlooked something, and @W1zzard didn't point it out either: this SSD is as bare as it gets, without even a graphene sticker. It's a prototype after all, and it can't be considered as fit for regular use in a desktop (where you have many options to cool it) or a notebook (where a sticker is the least, and possibly the most, that you can apply).

 

[Chrispy_]

Out of 1000 consumers buying pci-ex drives, 990 are not performing workloads where it makes any meaningful difference.

Recently many manufacturers are replacing high-end components with chinese cheap stuff keeping the same name and price. Aside from some reliable products (Crucial T500, SK Hynix P41, Samsung 980 and 990Pro and a few others disks, mainly because these brands manufacture their own components), buying a Crucial MX500 is always a safe and smart move (it sounds like a promo, i'm sorry).

Oh I know that 99% of consumers will be served just fine by SATA SSD performance. My raised eyebrow was more that the MX500 is super old and significantly more expensive than newer, better NVMe SSDs today.

WD/Sandisk are like Samsung and Micron in that they are vertically intergrated and make the controller, the firmware, and the NAND themselves. The SN550, 570, and 580 have all been award-winning, fantastic value and even the old SN550 runs circles around the MX500. Today, an SN580 is about 30% cheaper than an MX500 and around 10x faster in sequential throughput, whilst also being around 50% faster in real-world application performance, which is significant enough to be noticeable even by non-technical buyers.

I honestly thought the MX500 was discontinued a couple of years ago, but if I'm wrong on that, then the higher price is likely either the economies of scale (fewer people are buying SATA, it's likely that orders of millions of units from big OEMs are all NVMe now) or the fact that SATA controllers are more complex than NVMe ones, because they have to do everything an NVMe controller does, but also they have to emulate a mechanical hard drive and map requests to sectors, cylinders, tracks etc and support/leverage queue management features like NCQ which are for mechanical disk heads primarily. Yes, the MX500 does all this so well that it's often indistinguishable from a decent modern NVMe drive, but all that extra complexity has a very real cost overhead.

The other issue of SATA is transfer rates. Right now there's no killer feature that requires the sequential speeds of NVMe drives. Video editing with large RAW footage is certainly one of those features, but it's not something your average joe will do. Average Joe is currently unfazed by transfer rates because most of what Joe does is small transfers where SATA's transfer rate disadvantage only results in a fraction of a second here and there.

 

[Minus Infinity]

Sad that the old 970 Evo Plus has much faster sustained writes than most of the newer ssd's at ~1900MB/s and PCI-E 3.0. Still this is a prototype and I'm sure plenty of new products will use this controller and we'll see 4 and 8TB models next year.

Maybe Phison has hit the wall and we see more using that SM controller in the future.

 

[Chaitanya]

FORESHADOWING!


I think transfer rate per Watt in these reviews is a fantastic metric though, it's only hot if you get blistering speeds out of it.

Another way to look at it is that faster drives getting hotter is okay if they finish faster and then cool down earlier. Think of it like Ryzen's "rush to idle" mentality - better to boost to the moon to get the workload finished ASAP so you can go back to sleep.

On desktops you can tame these SSDs with massive heatsinks(many motherboards these days come with decent sized heatsinks on primary slot) and airflow. Problem is going to be in laptops(someone is going to blink on that side of things and put these hot gen5 drives in there) where there is limited cooling potential available.