Phison Unveils E26, the Company's First PCIe Gen5 Controller for High-end Desktop Gaming

[WhoDecidedThat]

An interesting video for the highly technical crowd.

Thanks for sharing that.

Real world usage with consumer workloads is 4KB QD1-4 random reads. The memory bus speed (flash to controller) has a direct impact on that in real world application performance.

So you're saying that a PCIe 5.0 SSD is equal to a PCIe 3.0 (three point zero) SSD as long as both share the same speed flash.

 

[Chris_Ramseyer]

Thanks for sharing that.


So you're saying that a PCIe 5.0 SSD is equal to a PCIe 3.0 (three point zero) SSD as long as both share the same speed flash.

A PCIe Gen3 SSD wouldn't have the interface to run memory released during the PCIe Gen5 era at full speed. All of the parts increase in capabilities together over time. It's not any different from the motherboard, DRAM, and CPU relationship. Would you expect to use an Alder Lake processor and motherboard with DDR2 and have the same results?

With higher speed NAND and lithography shrinks you also need new controller technology that can manage the error correction required. SSDs are very complex at the controller level.

 

[Mussels]

You may want to read up on what is coming to PC gaming in DirectX 12 Ultimate

The media will see the demo in a few days. I think your thoughts about PC gaming and storage will change then.

As long as we actually get some news on that, DirectStorage and RTX IO feel dead in the water

Obviously I can't tell you everything I know about it due to NDAs but I can say the entire industry is in meetings every week over this topic. If it wasn't coming, we wouldn't have anything to talk about.

Real world usage with consumer workloads is 4KB QD1-4 random reads. The memory bus speed (flash to controller) has a direct impact on that in real world application performance. That said, I did explain "why - esp. in real world usage." The reason why is because the new flash increases the speed between the memory and the controller and that reduces latency.

The last one is my favorite. Phison has publicly stated we spent 30 million Dollars to bring the first PCIe Gen4 SSD to market. I would say that is putting our money where our mouth is. We haven't released a number on Gen5 development but from 2018 to 2020 we added an additional 400+ people to our R&D department (engineers).

Also, it's good to have reps posting and discussing products with people. Don't mind the grumps.

Thanks for sharing that.


So you're saying that a PCIe 5.0 SSD is equal to a PCIe 3.0 (three point zero) SSD as long as both share the same speed flash.

They can be, yes.

Look at any SSD review and you'll find that certain tests give really slow results still - the biggest changes we see are when flash and/or controllers improve, allowing those results to change.
Raw read and write speed are the main improvement with new PCI-E generations, not important results like 4K random read and write

 

[Tek-Check]

No one wants to wait and I'm right there with you. The industry is a machine and we just make one piece. My job is just to make sure our piece is the fastest you can buy.

E26 supports memory bus speeds up to 2400MT/s. The fastest shipping today is 1600MT/s. The increase in memory bus speed will decrease latency so you will see the performance.

One thing I want to note about what was said above. Many of the SATA vs. NVMe tests were ran on first generation NVMe drives that used the same flash (and same memory bus speeds) as the SATA products shipping at the time. That is why the results were so close together. Run a SATA SSD against something like the Kingston KC3000 or Seagate FireCuda 530 using modern flash and a high-performance E18 controller and you will be surprised at the difference.

Not correct. Please see below. There were several high-en PCIe 3.0 and three PCIe 4.0 drives tested too. Those drives have faster and better controllers than SATA SSDs, such as SM2262EN and E16. Gain is there, but it is negligible, only 2-3 seconds. E18 could add another second or two and E26 another second or two. All in all, change from PCIe 3.0 to 5.0 could be 5-6 seconds, for gamers not optimised to take advantage of faster DRAM on SSD controller, which is vast majority of them at the moment.

Storage Game Loading Test: PCIe 4.0 SSD vs. PCIe 3.0 vs. SATA vs. HDD

What sort of storage do you need to play today's games and provide the best loading times? Do you need a PCIe drive? Do you need something...

www.techspot.com

No one is disputing better overall performance of incoming PCIe 5.0 NVMe drives and no one is disputing some better loading times for games. That is true. However, gains for games are negligible, in the same way as adding more cores than 6-8 does not scale up with frames per second boyond that point. Hardware has become more capable than game engines.

This is how the market looks like for PCIe 5.0 NVMe drives in 2022. I am afraid it will largely not exist on client gaming platforms until mid 2023 the earliest, if not later:
- no support whatsoever on new 2022 laptops - both AMD 6000 and Intel 12th gen platforms announced today come with PCIe 4.0 M.2 slots only - so, no one will buy any new PCIe 5.0 drive
- on desktops, only a few high-mid and high tier Z690 support PCIe 5.0 on GPU slots - bifurcation is necessary and another spending is needed on super-expensive PCIe 5.0 NVMe adapter
https://www.tomshardware.com/uk/news/msi-gen-5-nvme-card-adapter-for-z690-alder-lake
- Seagate FireCuda 520 1TB (PCIe 3.0) - €150, Seagate FireCuda 530 1TB (PCIe 4.0) - €200, Seagate FireCuda 540 1TB (PCIe 5.0) - €300? - another €100 for 2-3 second gain in 2023?
- CPUs will not natively support PCIe 5.0 NVMe drives until the next gen, at least

PCIe 5.0 NVMe drives are coming into client market too early and they are going to be too expensive for promised gains. Simple as that.

 

[W1zzard]

PCIe 5.0 NVMe drives are coming into client market too early and they are going to be too expensive for promised gains.

Everyone in the industry that I talked to knows that, there's still tremendous marketing value and profit from being the one to capture this niche

 

[looniam]

Don't mind the grumps.

this grump is willing to bet $1,000 that next year will be (2122) just be more marketing bullcrap.

any takers?

 

[Assimilator]

A PCIe Gen3 SSD wouldn't have the interface to run memory released during the PCIe Gen5 era at full speed. All of the parts increase in capabilities together over time. It's not any different from the motherboard, DRAM, and CPU relationship. Would you expect to use an Alder Lake processor and motherboard with DDR2 and have the same results?

With higher speed NAND and lithography shrinks you also need new controller technology that can manage the error correction required. SSDs are very complex at the controller level.

All I want to know is whether your PCIe 5.0 NVMe controllers are going to be power gluttons like the current E18 is; if so, I won't be buying drives using your controllers. Doubling bandwidth shouldn't require doubling power consumption!

SK hynix has shown that they can produce NVMe SSD controllers that are both performant and power-efficient with the Gold P31 (granted it's only PCIe 3.0); are you guys going to do the same? Or are you just going to go the Intel route and throw efficiency out of the window in pursuit of raw speed and headlines about it - speed that is going to be wholly irrelevant to the vast majority of users?

 

[Chris_Ramseyer]

All I want to know is whether your PCIe 5.0 NVMe controllers are going to be power gluttons like the current E18 is; if so, I won't be buying drives using your controllers. Doubling bandwidth shouldn't require doubling power consumption!

SK hynix has shown that they can produce NVMe SSD controllers that are both performant and power-efficient with the Gold P31 (granted it's only PCIe 3.0); are you guys going to do the same? Or are you just going to go the Intel route and throw efficiency out of the window in pursuit of raw speed and headlines about it - speed that is going to be wholly irrelevant to the vast majority of users?

We, along with everyone else, have to play in the power budgets set by PCI-SIG and what the board makers give us. It's not the wild-wild west out here.

It's not really fair to compared E18 with Gen4 to a Gen3 drive from a competitor. That drive was supposed to be a Gen4 product but they made an error in the lithography that couldn't be changed in software to fix it. Oops! So they essentially used the latest in lithography to make a previous generation product. If we wanted to go backwards and make an amazing Gen2 or Gen3 SSD using the latest in materials offered today we could, but what fun would that be?

Running the industry standard test for battery life (BAPCo MobileMark), E18 is one of the top performers for Gen4 SSDs available today. In a notebook with a factory battery time of 5.5 hours, the very best time scored, and the E18, are within 20 minutes of each other.

 

[Tek-Check]

Everyone in the industry that I talked to knows that, there's still tremendous marketing value and profit from being the one to capture this niche

Perhaps in server market once the feature is available with new generation of CPUs.

In client market, "tremendous" marketing value is currently zero and there is no profit to be made for at least two years. It's a pie in the sky at the moment.
There is no "niche" and nothing to capture in client market, as no single desktop or mobile platform support M.2 PCIe 5.0 slots. Perhaps next year if CPUs get x4 PCIe 5.0 support, in addition to x16 for GPUs.

Currently, the only way to install NVMe PCIe 5.0 drive on desktop is through expensive PCIe 5.0 adapter on expensive Z690 board with two PCIe 5.0 slots and run it in x8 mode, alongside GPU in x8 mode. This is far away from being a "niche". Literally a handful of enthusiastic users with a lot of money to spend would ever do this this year. And, of course, a few youtubers will show us how this works. That's it.

That is the reality of PCIe 5.0 this year.

PS5013-E13T - Phison's BGA for Mobile Gaming
Xiaomi chose Phison's E13T BGA SSD and its superior performance and efficiency for the Black Shark 4 gaming phone series. Xiaomi credits the E13T BGA for delivering a 69 percent increase in read and write performance showing that NVMe redefines mobile gaming.

I have E13T controller at home in one of my NVMe drives in NAS for caching. It's rock solid.

 

[Tek-Check]

PS5026-E26 - Phison's First PCIe Gen5 SSD Architecture

Are you able to say whether full speed PS5028 E28 controller is already in development for release in 2024?

 

[Prima.Vera]

For gaming, even a SATA drive suffice, just look at the loading times from TPU charts.