• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

Next Gen Phison NVMe SSD Controllers comparisons - Upto 7 GB/s

Status
Not open for further replies.

Bernkastel

Ask me about my fanboy energy!
We know that the Xbox Series X SSD is based on the Phison PS5019-E19T. But how does it compare to other upcoming or existing Phison Controllers ?
At FMS 2019, Phison shared the specs for its next generation of client SSD controllers. The PS5018-E18 high-end controller will be made using a FinFET process technology and promises to boost sequential read and write performance of drives to 7 GB/s. The PS5019-E19T mainstream controller will be a quick follow-up to the E13T that is currently in production but has not yet shipped in retail products.
With its PS5016-E16 chip, Phison was the first to introduce client SSD controllers with a PCIe 4.0 x4 interface and is currently enjoying its unique position on the market. The PS5016-E16 is essentially an update of the PS5012-E12 — still featuring two Arm Cortex-R5 cores, eight NAND channels rated at 800 MT/s and made using TSMC’s 28 nm process technology, but now featuring a PCIe 4.0 x4 PHY and enhanced error correction capabilities. In the coming months, Phison's entry-level NVMe controller will get a similar successor: the new PS5019-E19T will be based on the PS5013-E13T (one Arm Cortex-R5 core, four NAND channels, 28nm technology), but featuring a new PCIe 4.0 x4 PHY and thus enabling cost-effective yet fast SSDs.
Phison’s next-generation PS5018-E18 high-end client SSD controller with a PCIe 4.0 x4 interface will be significantly different from the current E16. The chip gains a third Arm Cortex-R5 core, NVMe 1.4 compliance, improved hardware security algorithms, and will be made using TSMC’s 12 FFC fabrication process.
The most important improvement in the new controllers is support for faster NAND interface speeds of up to 1200 MT/s. This is necessary to get full performance out of 96L and newer 3D NAND. For the high-end E18 this allows the controller to actually deliver twice the throughput of any PCIe 3.0 x4 drive, where the E16 can only use a fraction of the extra bandwidth provided by the PCIe 4.0 upgrade. The faster NAND interface also allows the E19T to hit sequential speeds that are slightly faster than a PCIe 3.0 x4 link can handle, despite the E19T having only four NAND channels instead of the eight channels used by high-end NVMe controllers.
Peak power consumption for both the E19T and E18 will be higher than for their respective predecessors, but that's because they are so much faster; the newer controllers will actually be more efficient.
Phison's booth at FMS included live demos of the E13T, E19T and E16 controllers, but the E18 controller was not in attendance. The low-end E19T controller will be sampling by the end of the year and is likely to show up in retail drives by the end of Q1 2020. This means that the E13T is likely to be very short-lived in the retail market. The high-end E18 will be arriving a bit later, sampling in Q1 2020 and hitting the shelves in Q3, a year after the E16 it replaces.
Phison NVMe SSD Controller Comparsion
E8/E8TE13TE19TE12E16E18
Market SegmentMainstream ConsumerMainstream ConsumerMainstream ConsumerHigh-End ConsumerHigh-End ConsumerHigh-End Consumer
Manufacturing
Process
40 nm28nm28 nm28 nm12 nm FFC
CPU Cores2x Cortex R51x Cortex R51x Cortex R52x Cortex R53x Cortex R5
Error CorrectionStrongECC4th Gen LDPC3rd Gen LDPC4th Gen LDPC
DRAME8: DDR3
E8T: No
NoNoDDR3/4DDR4DDR4, LPDDR4
Host InterfacePCIe 3.0 x2PCIe 3.0 x4PCIe 4.0 x4PCIe 3.0 x4PCIe 4.0 x4
NVMe VersionNVMe 1.2NVMe 1.3NVMe 1.4NVMe 1.3NVMe 1.4
NAND Channels, Interface Speed4 ch,
533 MT/s
4 ch,
800 MT/s
4 ch,
1200 MT/s
8 ch,
667 MT/s
8 ch,
800 MT/s
8 ch,
1200 MT/s
Max Capacity2 TB2 TB2 TB16 TB16 TB16 TB
Sequential Read1.6 GB/s2.5 GB/s3.75 GB/s3.4 GB/s5.0 GB/s7.0 GB/s
Sequential Write1.3 GB/s2.1 GB/s3.75 GB/s3.2 GB/s4.4 GB/s7.0 GB/s
4KB Random Read IOPSE8: 240k
E8T: 120k
350k440k700k750k1M IOPS
4KB Random Write IOPSE8: 220k
E8T: 130k
450k500k600k750k1M IOPS
Controller Power1.8 W1.2 W1.6 W2.1 W2.6 W3.0 W
SamplingQ2 2017Q2 2019Q4 2019Q2 2018Q1 2019Q1 2020
Retail SSD
Availability
Q4 2017Q4 2019Q1 2020Q4 2018Q3 2019Q3 2020

Ofcourse, Microsoft added many of there own customizations on top of it and its also possible that some of the features of E18 are available on the Series X SSD. It should also be noted that both Microsoft and Phison have different metrics to rate their SSD
 
Last edited:

Dodkrake

Banned
What is the purpose for this thread? Just trying to understand. We all know that the XBSX SSD has a throughput of 2.4GB/s, that's it. It can't go faster than that. Now, the decompressor will be able to yield an average of 4.8GB/s of compressed data, which can have peaks above 6GB/s.

Also, based on your table, it's much more likely it uses the E13T and not E19T, as the E19T has around 56% higher throughput than what you have in the XBSX.

Edit to add quick clarification:

The SSD can feed the decompressor a maximum of 2.4GB/s of compressed data. This is its peak throughput. Then, it's the decompressor's job to decompress that data so that the GPU can use it. There's nothing fancy, or miraculous.
 
Last edited:

Bernkastel

Ask me about my fanboy energy!
What is the purpose for this thread? Just trying to understand. We all know that the XBSX SSD has a Throughput of 2.4GB/s, that's it. It can't go faster than that. Now, the decompressor will be able to yield an average of 4.8GB/s of compressed data, which can have peaks above 6GB/s.

Also, based on your table, it's much more likely it uses the E13T and not E19T, as the E19T has around 56% higher throughput than what you have in the XBSX.
XSX SSD is based on E19T, its known for a while even before the Digital Foundry reveal. 3.7 GB/s raw speed is the max E19T can go, but as shown in that tweet Microsoft does not use max speed to rate, but a more conservative number.
SR/SW (MBps)Max: 3750/3750
 
Last edited:

MoreJRPG

Suffers from extreme PDS
XSX SSD is based on E19T, its known for a while even before the Digital Foundry reveal. 3.7 GB/s raw speed is the max E19T can go, but as shown in that tweet Microsoft does not use max speed to rate, but a more conservative number.

Its an interesting philosophy. Sony markets their SSD with the maximum speed that in practice will never reach that and Microsoft markets the more realistic speed, but knows they’ll take a beating in the press because no one can critically think anymore it seems.

PS5 is still obviously faster but it’s neat to watch unfold.
 

Degree

Banned
XSX SSD is based on E19T, its known for a while even before the Digital Foundry reveal. 3.7 GB/s raw speed is the max E19T can go, but as shown in that tweet Microsoft does not use max speed to rate, but a more conservative number.
SR/SW (MBps)Max: 3750/3750

which is a good thing.
Underpromise and overdeliver. And be transparent. Great approach.
Diversity is good. Every 10th or 20th "PS5 SSD for president" threads 1 "Xbox SSD does not completely suck" thread does not harm.

can you contribute to the thread or do you only want to shitpost?
 
Last edited:

Dodkrake

Banned
XSX SSD is based on E19T, its known for a while even before the Digital Foundry reveal. 3.7 GB/s raw speed is the max E19T can go, but as shown in that tweet Microsoft does not use max speed to rate, but a more conservative number.

Stop twisting what they said. The tweet clearly refers to the decompressor block, not the SSD raw throughput (which is 2.4GB/s).



And as other users mentioned Sony, which I wasn't going to, this is exactly the same narrative Cerny passed:

- RAW throughput of 5.5GB/s
- Decompressor handles 8 to 9GB/s typically, and can go up to 22GB/s

As for it being based on the E19T, I don't want to doubt you, but please provide proof for that. None of your articles provide said proof.

Diversity is good. Every 10th or 20th "PS5 SSD for president" threads 1 "Xbox SSD does not completely suck" thread does not harm.

That's not my point.
 

Degree

Banned
Stop twisting what they said. The tweet clearly refers to the decompressor block, not the SSD raw throughput (which is 2.4GB/s).



And as other users mentioned Sony, which I wasn't going to, this is exactly the same narrative Cerny passed:

- RAW throughput of 5.5GB/s
- Decompressor handles 8 to 9GB/s typically, and can go up to 22GB/s

As for it being based on the E19T, I don't want to doubt you, but please provide proof for that. None of your articles provide said proof.



That's not my point.


Maybe you can stop twisting?

The news comes from Phison software engineer Jerry Lou's LinkedIn page, which specifically outs the Xbox Series X's solution:

"Designed flagship PS5016-E16, world's first PCIe Gen4x4 NVMe SSD solution, and PS5019-E19, DRAM-less PCIe Gen4x4 NVMe SSD installed in Xbox Scarlett by hosting scrum ceremonies as scrum master."

source:
 
Last edited:

Bernkastel

Ask me about my fanboy energy!
Stop twisting what they said. The tweet clearly refers to the decompressor block, not the SSD raw throughput (which is 2.4GB/s).



And as other users mentioned Sony, which I wasn't going to, this is exactly the same narrative Cerny passed:

- RAW throughput of 5.5GB/s
- Decompressor handles 8 to 9GB/s typically, and can go up to 22GB/s

As for it being based on the E19T, I don't want to doubt you, but please provide proof for that. None of your articles provide said proof.



That's not my point.

It was also in a Phison engineers LinkedIn page, but he removed it
Its true both for the RAW speed and compressed speed. Phison and other companies use max speed to rate their SSDs, Microsoft does not.
 

Dodkrake

Banned
Maybe you can stop twisting?

source:

The onus for providing proof is not on me, as these are not my claims. That page is not in the original post, so my point still stands. Still, thank you.

It was also in a Phison engineers LinkedIn page, but he removed it
Its true both for the RAW speed and compressed speed. Phison and other companies use max speed to rate their SSDs, Microsoft does not.

Conveniently, the user removed it. Still, thank you.

Now, where does it say that MS doesn't use the "max speed" for their SSD throughput measurements? Your post does not provide proof of this, and instead focuses on the decompressor block.

Not being confrontational, but I would like to see the actual proof behind these statements.
 

Bernkastel

Ask me about my fanboy energy!
The onus for providing proof is not on me, as these are not my claims. That page is not in the original post, so my point still stands. Still, thank you.



Conveniently, the user removed it. Still, thank you.

Now, where does it say that MS doesn't use the "max speed" for their SSD throughput measurements? Your post does not provide proof of this, and instead focuses on the decompressor block.

Not being confrontational, but I would like to see the actual proof behind these statements.
Well Phison rated it at 3.75 GB/s. With all those customizations like SFS, XSX SSD should be faster than E19T, yet they rated it 2.4 Gb/s. Also, because of BCPack, compressed speed should be around double the raw speed, so they used the same metric to measure both.


Why would they even use a different metric ?
 
Last edited:

Dodkrake

Banned
Well Phison rated it at 3.75 GB/s. With all those customizations like SFS, XSX SSD should be faster than E19T, yet they rated it 2.4 Gb/s. Also, because of BCPack, compressed speed should be around double the raw speed, so they used the same metric to measure both.


Why would they even use a different metric ?


That's a big assumption. Let's see that claim:

  • Phison claims 3.75GB/s (let's assume that's the maximum and not the average)
  • MS claims 2.4GB/s on their page
  • You claim MS is using a "conservative number" and should be faster than the E19T, rated at 3.75GB/s
This leads me to believe that they are either massively underselling their SSD tech or that your claims have no substantiation. As requested before, please provide proof that 2.4GB/s is an average and / or a massively conservative number.

Additionally, I feel like I need to clarify that you seem to be (willingly or unwillingly) mixing decompression with RAW data. This is actually fairly simple.
  • Not all data can be compressed the same. Some data can be compressed as high as 90%, some data only 10%
  • Your decompressor must be able to handle 2.4GB/s (X) of Compressed data and then be able to output >6GB/s (Y) RAW data
  • In this case, X is your SSD throughput speed, and Y is your decompressor block speed.
  • You cannot go above neither X or Y, however, you can compress your data better so that the constant X is then turned into Y + 10%, for example
I fail to see how the BCPack comment is relevant

Edit: Added bolds and a note / strike-through regarding Bernkastel Bernkastel correction on the Phison SSD speed being the maximum, so that I'm not misleading people that may get the post.
 
Last edited:

Panajev2001a

GAF's Pleasant Genius
XSX SSD is based on E19T, its known for a while even before the Digital Foundry reveal. 3.7 GB/s raw speed is the max E19T can go, but as shown in that tweet Microsoft does not use max speed to rate, but a more conservative number.
SR/SW (MBps)Max: 3750/3750

I think you are misinterpreting these stats and the XSX setup: that tweet is talking about the BCPack + Zlib decompression block MS added between SSD and RAM essentially, not the SSD raw throughout.

The max SSD raw speed for the XSX SSD is 2.4 GB/s (that is not a conservative estimate, that is what MS configured it as regardless of the controller specs on paper), the “conservative/realistic” number is the 4.8 GB/s equivalent one once texture and data decompression is factored into things.
 
Last edited:

Bernkastel

Ask me about my fanboy energy!
(let's assume that's the maximum and not the average)
They literally call it max speed. I am requoting my own post for you.
XSX SSD is based on E19T, its known for a while even before the Digital Foundry reveal. 3.7 GB/s raw speed is the max E19T can go, but as shown in that tweet Microsoft does not use max speed to rate, but a more conservative number.
SR/SW (MBps)Max: 3750/3750
 
Last edited:

Bernkastel

Ask me about my fanboy energy!
Great, I stand corrected. My whole post still stands, so please address everything else. I see what you tried to do here, but you failed.
The form factor is cute, the 2.4GB/s of guaranteed throughput is impressive, but it's the software APIs and custom hardware built into the SoC that deliver what Microsoft believes to be a revolution - a new way of using storage to augment memory (an area where no platform holder will be able to deliver a more traditional generational leap).
 

Panajev2001a

GAF's Pleasant Genius
They literally call it max speed. I am requoting my own post for you.

The controller, if configured to do so, could take data at that speed from the NAND and serve it to the I/O decompressing unit for data amplification. The latter makes sense to tak about conservative estimates, not really the former.

You are mixing the equivalent raw bandwidth with decompression in place (based on average/realistic/etc... compression ratios) with the speed of the I/O solution MS uses which is 2.4 GB/s max.

I find it odd that for some in the XSX side of things the SSD bandwidth is both overrated and yet all these kinds of arguments come out of the woodwork to make it look like it has tons and tons more bandwidth than people realise or MS is willing to advertise 🤷‍♂️...
 
Last edited:

Panajev2001a

GAF's Pleasant Genius

... that... is that your evidence to say that the actual raw bandwidth before taking compression into account is almost double that amount? That actually the XSX SSD’s raw I/O is more like 3.5-3.7 GB/s and this 7+ GB/s when factoring compression? ... and MS is just hurting themselves by being too humble and forcing fans to educate others for them?

This is as if Sony fans started thread upon thread to theorise about scenarios Sony is shy and avoids bragging about where actually PS5 is way more than 10.x TFLOPS and actually really is 13 TFLOPS and I dunno held on to an earlier 2020 claim where the GPU frequency is capped to 2.23 GHz in conservative scenario but by launch this can even be unlocked and reach 2.6 GHz or more, etc... etc...

It feels like one one side SSD speed is overrrated and only TFLOPS matter, but just to be safe (and because the victories that XSX has are “just” victories where the loser is really close behind, not being outclassed and dominated perhaps?)... we need to try to make the SSD speed gap look smaller to appear stronger overall.
 

ToadMan

Member
Because everyone already knows that.

I don't think everybody is as misinformed as you are. You could've posted your evidence - if it existed, which it doesn't.

But this post is an admission that you're posting known inaccuracies to.... stir up shit? Make an irrelevant point?
 

Bernkastel

Ask me about my fanboy energy!
I don't think everybody is as misinformed as you are. You could've posted your evidence - if it existed, which it doesn't.

But this post is an admission that you're posting known inaccuracies to.... stir up shit? Make an irrelevant point?
Are you being dumb on purpose ? XSX using a customised E19T is an old news.
 
Last edited:
Status
Not open for further replies.
Top Bottom