Microsemi Announces PCIe 4.0 Switches And NVMe SSD Controllerby Billy Tallis on August 3, 2018 8:00 AM EST
- Posted in
- PCIe Switches
- PCIe 4.0
Microsemi is starting their transition to PCIe 4.0 with updates to their Switchtec PCIe switch family and Flashtec NVMe SSD controllers. The new PCIe standard doubles the per-lane throughput from approximately 1GB/s to almost 2GB/s.
The new Flashtec NVMe 3016 SSD controller features a 16-channel NAND interface and all the features of the existing 2016 controller, but the faster host interface will allow drives to exceed 8GB/s and 2 million IOPS when using a PCIe 4.0 x8 link. The 3016 controller is designed to allow for reuse of firmware developed for the previous generation of controllers, so new SSDs can be brought to market quickly. Microsemi's Flashtec controllers are used in many of the fastest enterprise SSDs, and the update to support PCIe 4.0 speeds should help Microsemi stay on top of this market segment.
The first Switchtec PCIe switches to support PCIe 4.0 are intended to support GPGPU applications that rely on low-latency peer to peer DMA. These switches will be available with up to 100 PCIe lanes, but specific models and configurations have not yet been announced.
Both the PCIe switches and the new NVMe controller are currently sampling to select customers. Microsemi's primary competition in the PCIe switch space comes from Broadcom's PLX switch family, which has not yet been updated to support PCIe 4.0 speeds. The Flashtec NVMe 3016 likewise doesn't have any direct competition.
Adoption of the new switches and controllers may be slow due to the dearth of other PCIe 4.0 hardware. IBM's POWER9 processors and some of Mellanox's networking interfaces and processors are shipping with PCIe 4.0 support, but GPUs and x86 CPUs have not officially started the transition.
Microsemi will be exhibiting at Flash Memory Summit next week, where we hope to see a working demo of their PCIe 4.0 solutions.
Post Your CommentPlease log in or sign up to comment.
View All Comments
smilingcrow - Friday, August 3, 2018 - link"Allow drives to exceed 8GB/s when using a PCIe 4.0 x8 link."
That seems very low when you look at what can be achieved with a PCIe 3.0 x4 link which is around 3.5GB/s.
DanNeely - Friday, August 3, 2018 - linkMight not be the clearest wording, but I think the intent was to say it'd be faster than the theoretical note quite 8GB/s from a PCIe3 x8 line.
FullmetalTitan - Friday, August 3, 2018 - linkIt also appears to be related to the compatibility built in here. They claim that controllers on market will work with these switches, perhaps that figure for bandwidth is only for the specific use case of a controller designed for PCIe 3.0 used on these new 4.0 chipsets.
FullmetalTitan - Friday, August 3, 2018 - linkOn second reading I think Dan had it right, that quoted speed is for the 3016 controller. It may be just a quick and dirty port of existing functionality to support 4.0 interface, but it's a new controller nonetheless
Xajel - Monday, August 6, 2018 - linkIt's might be a mistype in the text, AFAIK PCIe 4.0 will double the speed of PCIe 3.0, so PCIe 3.0 8 lanes can already reach 8GB/s in paper (a little less in real world) so PCIe 4.0 8 lanes could reach 16GB/s
Most NVMe's are based on M.2 or U.2 PCIe 3.0 4 lanes, they maxes out at 3.x GB/s which is plausible as the bus theoretical limit is 4GB/s, so with PCIe 4.0, those 4 lanes could reach 8GB/s.
Triklops - Friday, August 3, 2018 - linkIs it just me or does the font used in the 'Microsemi' logo look strikingly similar to that of 'Microsoft'?!
bubblyboo - Friday, August 3, 2018 - linkMicrosoft doesn't use that font anymore though. Should be back in pre-XP era. Logo colors are also the same, just swapped.
FXi - Monday, August 6, 2018 - linkThe move to 4.0 will double Thunderbolt bandwidth without increasing the lanes. "When" we finally get this on the x86 side of things the increase will greatly help move things forward. GPU's probably won't see any benefit but connected devices and possibly an increase to DMI 4.0 (possibly) would boost areas that could use it like interconnects. I think this is more an interesting sign that vendors know where they need to be come the end of this year and are beginning to head there.