RISC-V SSD controller provides 14GB/s transfers without a cooling solution

[Alfonso Maruccia]

A hot potato: Many PCIe 5.0 SSDs currently on sale provide extremely high transfer rates compared to previous generations, however they all appear to require an often-cumbersome cooling solution to avoid throttling. The RISC-V open instruction set architecture (ISA) could arrive to help and end this hotly debated issue.

Major tech companies have invested in promoting RISC-V, a royalty-free ISA that could theoretically power everything from automotive solutions to smartwatches, or even next-generation phones. Some storage hardware manufacturers already employ RISC-V in fast SSDs. However, China-based Yingren Technology is one of the first to use the open ISA exclusively to power its latest SSD controller.

The manufacturer recently introduced the YRS820 chip, a controller for PCIe 5.0 SSDs based on an all-RISC-V design. The chip provides a read speed of up to 14 GB/s and a write speed of up to 12 GB/s. The YRS820 doesn't need an active cooling solution and is the consumer-grade version of Yingren's controller designed for data center applications (YRS900).

Both versions have the same read/write speeds, so it's unclear how they differ besides on how they are marketed. The transfer rate for a typical, four-lane PCIe 5 solid-state drive is around 16 GB/s, which means that the new controller comes close to the theoretical limit of the latest NVMe storage tech.

Yingren's demo is impressive, considering that the controller is almost on par with the leading solutions already on the market. The Phison E26 controller uses four Arm Cortex-R5 cores and a single RISC-V AndesCore N25F core, reaching 14.5 GB/s and 12.7 GB/s in reading and writing speeds (available on the Crucial T705 SSD).

However, Phison E26-powered SSDs need passive or even active, fan-based cooling solutions to achieve those speeds, while the YRS820 chip can seemingly come pretty close to the same data transfer values within a "fanless" SSD design.

Even better, the unit can provide higher transfer rate levels (2,667MT/s) compared to Phison's E26 (2,400MT/s), with larger storage capacities (8TB vs 4TB). The RISC V-based YRS820 controller could be an enticing alternative for consumer SSD manufacturers, but Yingren and other smaller companies are arriving somewhat late to the PCIe 5 game.

Challenging Phison's leadership in the market will not be easy, as the company is already developing a new, low-power controller (E31T) that can reach read and write speeds of up to 10.8 GB/s.

Permalink to story.

https://www.techspot.com/news/102413-latest-risc-v-controller-provides-14gbs-transfers-without.html

 

[maxxcool7421]

BS. Show me a live demo copying 10,000 12MB photos from a 20GB/s source to this unit in a 90f case. 100% bet it throttles and drops speed after like 40 seconds

Im not against Risc-V taking over as a ''open source'' (it is not 1000% OSrc) for things. but everything silicon based under load for long periods gets hot.

 

[WhiteLeaff]

BS. Show me a live demo copying 10,000 12MB photos from a 20GB/s source to this unit in a 90f case. 100% bet it throttles and drops speed after like 40 seconds

Im not against Risc-V taking over as a ''open source'' (it is not 1000% OSrc) for things. but everything silicon based under load for long periods gets hot.

Most controllers are manufactured at 12nm or worse, there is a lot of scope to bring efficiency to the table, especially with a design purely optimized for low power.

 

[bviktor]

90% of users won't ever notice. Most NAND is garbage. They specify the peak speed, but that's only the cache - once you fill it up, it's over. They like to blame it on throttling, but no, most of the time it has nothing to with throttling, and everything with the fact that 99% of the NAND on the SSD is pure junk, and they only give you like 1 or 2 gigs of very fast, quality NAND that acts purely as a write cache. But they don't mention that in the specs, do they.

I just bought an ADATA "2800 MB/s" (max) SSD, then cloned another SSD onto it, so it was like a 400 GB copy. After the first couple of seconds it went down to almost precisely 100 MB/s, and stood there during the whole process. And this is an SSD with a heat sink, I might add. The funny thing is, they even brag about the whole cache thing: "The LEGEND 800 is equipped with SLC caching". The average Joe won't understand anyway, so why not.

But it's not specific to ADATA, everyone's doing it. Just awful. And if you actually manage to get a premium SSD with proper NAND, then you get to fix your thermal issues, of course. Put a huge heat sink on it, or even a fan, because 200 IQ product development.

There, enjoy your 10 GB/s+ SSDs. It's a CRIME that SSD manufacturers aren't obligated to specify SUSTAINED write speeds, because at this point, it's really just a big scam.

 

[Theinsanegamer]

90% of users won't ever notice. Most NAND is garbage. They specify the peak speed, but that's only the cache - once you fill it up, it's over. They like to blame it on throttling, but no, most of the time it has nothing to with throttling, and everything with the fact that 99% of the NAND on the SSD is pure junk, and they only give you like 1 or 2 gigs of very fast, quality NAND that acts purely as a write cache. But they don't mention that in the specs, do they.

I just bought an ADATA "2800 MB/s" (max) SSD, then cloned another SSD onto it, so it was like a 400 GB copy. After the first couple of seconds it went down to almost precisely 100 MB/s, and stood there during the whole process. And this is an SSD with a heat sink, I might add. The funny thing is, they even brag about the whole cache thing: "The LEGEND 800 is equipped with SLC caching". The average Joe won't understand anyway, so why not.

But it's not specific to ADATA, everyone's doing it. Just awful. And if you actually manage to get a premium SSD with proper NAND, then you get to fix your thermal issues, of course. Put a huge heat sink on it, or even a fan, because 200 IQ product development.

There, enjoy your 10 GB/s+ SSDs. It's a CRIME that SSD manufacturers aren't obligated to specify SUSTAINED write speeds, because at this point, it's really just a big scam.

It wouldnt matter if they did. Despite all the evidence online from a variety of sources on how slow QLC SSDs are once the cache is full, and the fact the LEGEND 800 series uses QLC flash comes up if you google search the model name, you STILL went and bought one, then got surprised by how slow it was. If you dont read reviews or testimonials, then what good does publishing the spec do?

You can lead a horse to water, but you cant make it drink.

 

[Watzupken]

90% of users won't ever notice. Most NAND is garbage. They specify the peak speed, but that's only the cache - once you fill it up, it's over. They like to blame it on throttling, but no, most of the time it has nothing to with throttling, and everything with the fact that 99% of the NAND on the SSD is pure junk, and they only give you like 1 or 2 gigs of very fast, quality NAND that acts purely as a write cache. But they don't mention that in the specs, do they.

I just bought an ADATA "2800 MB/s" (max) SSD, then cloned another SSD onto it, so it was like a 400 GB copy. After the first couple of seconds it went down to almost precisely 100 MB/s, and stood there during the whole process. And this is an SSD with a heat sink, I might add. The funny thing is, they even brag about the whole cache thing: "The LEGEND 800 is equipped with SLC caching". The average Joe won't understand anyway, so why not.

But it's not specific to ADATA, everyone's doing it. Just awful. And if you actually manage to get a premium SSD with proper NAND, then you get to fix your thermal issues, of course. Put a huge heat sink on it, or even a fan, because 200 IQ product development.

There, enjoy your 10 GB/s+ SSDs. It's a CRIME that SSD manufacturers aren't obligated to specify SUSTAINED write speeds, because at this point, it's really just a big scam.

I think it depends. If there is no heatsink, a PCI-E 5.0 SSD will quickly overheat and throttle. If there's sufficient cooling, then the next thing to slow it down will be when the SLC cache is used up. Its been a long time since TLC NAND have became mainstream, and perhaps people some would have forgotten that TLC is very slow to begin with. Most people won't hammer the SSD with constant write that will easily exhaust the SLC cache, and so the issue is not apparent to us. The other thing to point out is that regardless of SLC, MLC or TLC NAND, none of them will be able to sustain in various types of load. If you want something that sustains well, then you got to get an Intel Optane drive.

With regards to the marketing, I think they will normally cover themselves by mentioning "up to". In short, the speed is not guaranteed. To most running a short benchmark, the results will be close.

 

[bviktor]

It wouldnt matter if they did. Despite all the evidence online from a variety of sources on how slow QLC SSDs are once the cache is full, and the fact the LEGEND 800 series uses QLC flash comes up if you google search the model name, you STILL went and bought one, then got surprised by how slow it was. If you dont read reviews or testimonials, then what good does publishing the spec do?

You can lead a horse to water, but you cant make it drink.

You're imagining things. I never said anything close to what you're assuming. I wasn't surprised how slow it was. I am perfectly aware how slow they are. I brought it up to prove my point, not for the audience to feel sorry for me.

If you don't understand the difference between the manufacturer's official specs and the hands-on testing with some random methodology by some random youtuber, I don't know what else to say to you.