Phison Showcases 12 GB/s Speeds for PCIe 5.0 SSDs Through Its New E26 Controller

[Raevenlord]

Phison has showcased the expected performance of its upcoming PS5026-E26 controller, built to usher NVMe SSDs into the PCIe 5.0 realm. The company showcased its new controller's prowess by building a reference SSD design based on 1 TB of Micron's TLC NAND. Phison's new controller has been built from the ground-up to accelerate next-generation SSD workloads - including direct access technologies based on Microsoft's DirectStorage API, accelerated by two ARM Cortex-R5 cores and three proprietary CoXProcessor 2.0 accelerators built on TSMC's 12 nm process.

Phison's internal testing shows its reference SSD achieving sequential read speeds of over 12 GB/s in CrystalDiskMark, with sequential writes going as high as 10 GB/s - a 70% performance increase compared to the world's fastest PCIe 4.0 SSDs, which currently top out at around 7 GB/s sequential speeds. As to 4K performance, one of the most tangible metrics for user experience, random reads are set at around 16.000 IOPS, showcasing room for improvement with further firmware optimizations for actual shipping products.





Interestingly, Phison opted for the M.2 2580 form-factor for its proof-of-concept SSD, which features a slightly wider PCB and connector footprint that's not backwards compatible with M.2 2280 slots. Expect SSDs based on Phison's PS5026-E26 controller to hit the market later this year - closer to AMD's release of its 600-series chipsets for its next-generation AM5 platform.

View at TechPowerUp Main Site | Source

 

[dj-electric]

Prepare for a wave of drives that can flex its sequentials, but barely top randoms.

 

[Chaitanya]

Prepare for a wave of drives that can flex its sequentials, but barely top randoms.

That is so common in gyms here, people just work on upper body while having chicken legs.

 

[mechtech]

And the real world difference will be next to nothing.

HP FX900 1 TB M.2 NVMe SSD Review

Priced at just $105 for the 1 TB variant, the HP FX900 1 TB is one of the most affordable PCI-Express 4.0 SSDs. It's not only affordable, but also runs very fast, beating every single PCIe 3.0 drive we've ever tested. A preinstalled heatsink is included, too—there's no thermal throttling at all.

www.techpowerup.com

 

[DeathtoGnomes]

70% that should load windows in 5 seconds...

 

[Mysteoa]

70% that should load windows in 5 seconds...

Like that would help.

 

[zlobby]

* While DRAM cache lasts.

 

[MikeGR7]

No random 4kQD1 performance increase= No performance increase.

Intel Optane will still be the performance king for years to come.

 

[DeathtoGnomes]

Like that would help.

damn, where is that /sarcastic emote?

 

[aQi]

Nothing that can flip the world. Those high speeds are useless in real life senerios until you companies do something about randoms.

 

[DrCR]

Any insight as to how a company would go about building a drive differently to significantly improve randoms? Any why don’t they?

 

[dj-electric]

Any insight as to how a company would go about building a drive differently to significantly improve randoms? Any why don’t they?

Intel Optane drives were random monsters at the time. Consumers didn't care, and Optane Cache (a completely different product) had a bad name to it that might have done colleteral damage to its hype and potential sales. It was also dummy expensive to make, since the controller was a beast of a component in complexity and power.

-Optane 905p user, from then until it goes bad, which might take a while, since it's rated at multiple dozen times higher endurance than your avg drive

 

[Fierce Guppy]

Prepare for a wave of drives that can flex its sequentials, but barely top randoms.

Please post this to Phison's Facebook page where they will see it.

Phison Technology

Phison Technology. 602 likes · 11 talking about this. Phison Electronics Corp. (TPEX:8299) is a global leader in NAND Flash controller IC and storage solutions.

www.facebook.com

 

[Oberon]

Any insight as to how a company would go about building a drive differently to significantly improve randoms? Any why don’t they?

A combination of faster NAND and a faster (not wider) controller. Realistically, you're going to be waiting for a successor to NAND flash before you see real improvements in low-QD random performance.

 

[ZetZet]

And the real world difference will be next to nothing.

HP FX900 1 TB M.2 NVMe SSD Review

Priced at just $105 for the 1 TB variant, the HP FX900 1 TB is one of the most affordable PCI-Express 4.0 SSDs. It's not only affordable, but also runs very fast, beating every single PCIe 3.0 drive we've ever tested. A preinstalled heatsink is included, too—there's no thermal throttling at all.

www.techpowerup.com

Well hopefully when DirectStorage starts getting implemented and improved we will see the difference in games.

 

[regs]

Intel Optane will still be the performance king for years to come.

Sadly Intel and Micron ceased developments of 3D XPoint.

 

[Space Lynx]

No random 4kQD1 performance increase= No performance increase.

Intel Optane will still be the performance king for years to come.

I wonder if physics has reached its limits on the random 4kQD1 performance

if so that is really sad...

 

[zlobby]

Well hopefully when DirectStorage starts getting implemented and improved we will see the difference in games.

I for one prefer NV RAM at the speeds of current generation DDR, and the size of current generation platters.

 

[MentalAcetylide]

That is so common in gyms here, people just work on upper body while having chicken legs.

I think Mick Hart ripped on these types frequently over in the UK. I call those guys "bicep boys" or "chest brahs". When it comes to strength & functionality, your legs, hips, and core(abs) are more important. Neglect any of those, and it won't matter what you do with your upper body. Sure, the upper body will look nice, but the overall appearance will be a golf tee that has much less strength unless you bury the individual in the ground from the feet up to their waist.

 

[Wirko]

I wonder if physics has reached its limits on the random 4kQD1 performance

if so that is really sad...

It's economics rather than physics. New technologies are develeped for the datacenter first, that's where the money is. But server applications care little for QD1 because they are properly multithreaded where it matters, and longer queues are more common. Look at these three datacenter SSDs, they only achieve 38-50 MB/s on random read at Q1T1:

Kingston DC1000M SSD Review: Big, Fast NVMe Storage - Page 3

Kingston's DC1000M is the company's latest, high-performance NVMe SSD targeting enterprise and data center workloads. - Page 3

hothardware.com

A combination of faster NAND and a faster (not wider) controller. Realistically, you're going to be waiting for a successor to NAND flash before you see real improvements in low-QD random performance.

That's probably true. Dynamic RAM has the same issue, although on a different scale: true random access still takes about 60 ns, which is a modest improvement since DRAM was invented. Ways to get around this problem are similar, too - more parallelization, more queueing (DRAM controllers do that), larger caches, more levels of cache. And optimization of software to use random access less often.

 

[MarsM4N]

Look at those speeds, it's going to blow our socks off!

Well hopefully when DirectStorage starts getting implemented and improved we will see the difference in games.

Exactly. Just look at the file sizes of some AAA games, 100GB+ becomming the standard.

 

[Tomorrow]

A combination of faster NAND and a faster (not wider) controller. Realistically, you're going to be waiting for a successor to NAND flash before you see real improvements in low-QD random performance.

Didn't some companies experiment with Z-NAND and XL-NAND?

I've never heard of these so im guessing these attempts were canned when Optane was pushed to the sidelines?

It's sad really because we have the tech to push for nearly unlimited endurance and randoms/low latency courtesy of Optane but Intel in their infinite wisdom decided to keep it proprietary and there it died. With economies of scale and opening it up to AIB's at least the cost could have been brought down while allowing them to keep making Optane themselves.

 

[zlobby]

Didn't some companies experiment with Z-NAND and XL-NAND?

I've never heard of these so im guessing these attempts were canned when Optane was pushed to the sidelines?

It's sad really because we have the tech to push for nearly unlimited endurance and randoms/low latency courtesy of Optane but Intel in their infinite wisdom decided to keep it proprietary and there it died. With economies of scale and opening it up to AIB's at least the cost could have been brought down while allowing them to keep making Optane themselves.

Are you talking about 3DXpoint?

 

[Mysteoa]

damn, where is that /sarcastic emote?

I like to go in people sarcasm at times.

 

[kapone32]

Intel Optane drives were random monsters at the time. Consumers didn't care, and Optane Cache (a completely different product) had a bad name to it that might have done colleteral damage to its hype and potential sales. It was also dummy expensive to make, since the controller was a beast of a component in complexity and power.

-Optane 905p user, from then until it goes bad, which might take a while, since it's rated at multiple dozen times higher endurance than your avg drive

It's not that people did not want OPtane as much as it was much too expensive for the Capacity.