Intel Acquires Rivet Networks: Killer Networking is all in for Team Blueby Dr. Ian Cutress on May 20, 2020 4:00 PM EST
- Posted in
- Rivet Networks
- Wi-Fi 6
- Gigabit Ethernet
News hot off the wire is that Rivet Networks, the company behind the Killer range of accelerated networking products and analysis tools, is being acquired by Intel. The two companies have been working very closely of late, using a unified silicon strategy for the latest gigabit Ethernet networking silicon and also Wi-Fi 6 add-in cards and CNVi CRF modules for laptops. This new acquisition for Intel will enable an element of Ethernet traffic monitoring and optimization the portfolio has not had before, but it will be interesting to see how Intel hands the acquisition compared to when Qualcomm Atheros acquired Rivet Networks some years ago.
A lot of technology savvy users that have been around a while know about the Killer brand of networking products. The company originally burst onto the scene with an FPGA and a big heavy K heatsink looking to offer lower PC-to-Internet latencies, especially in games. Over time that FPGA became its own ASIC and gigabit Ethernet controller, and the company moved more into the ability to transparently detect and shape networking traffic, allowing high-priority traffic to pass through with the lowest latency, but downloads and streaming to get the lowest latency. Users were able to configure their network, as well as direct traffic through different networking interfaces if two Killer products were supported.
The original Bigfoot Killer NIC in 2006
The company originally started as Bigfoot Networks, and came to market with the Killer NIC in 2006. Qualcomm’s Atheros division, focused on networking, acquired the company in September 2011. The acquisition with Qualcomm gave a lot of access to Qualcomm’s ASIC building capabilities, bringing the power of the NIC down from an FPGA but also increasing the capability of the hardware and software. However, after several years of no product development or generational iterations, the original founders and engineers of the company spun back out of Qualcomm to form Rivet Networks, in an effort to build the Killer branding once again. Originally working with Qualcomm’s Atheros silicon, Rivet Networks started partnering with Intel and Realtek on various parts offering a standard version under the normal brand or the Killer version with additional network detection and shaping capabilities. This led to a resurgence in the capabilities of the hardware, with Dell, MSI, GIGABYTE, ASRock, and other OEMs becoming customers.
The Rivet Networks Killer AX1650, already built on Intel AX200 Silicon
When we saw a Killer NIC in the Dell XPS, the company had truly made it. Dell’s business machines also got access to SmartByte, a special app detection algorithm for Dell end-users and business customers. Rivet Networks have also developed a number of technologies to its portfolio, including supporting switch-like mechanics for multi-controller systems, or Wi-Fi extension services through time-muxing the Wi-Fi modem.
All these technologies will now fall under the Intel umbrella. The Rivet Networks team will join Intel’s Wireless Solutions Group within the Client Computing Group. Given that the two groups have been working very closely with the AX201 and Killer AX1650 networking chips recently, which underneath both use Intel silicon, it will be interesting to see where it all goes from here. I know of a number of plans that the Rivet team were working towards, some of them would be very beneficial to the consumer market, so I hope that Intel keeps the same passion alive.
This news is still breaking, we will update as we get more information
I had an on-the-record call with the Rivet Networks team and Intel, with lots of interesting information. While the value of the acquisition is not being disclosed, talks started in earnest at the end of last year about the right time and the level of synergy between the two companies. There is no mention of personnel, however every person that Intel offered a position too at Rivet took that offer. Rivet's CEO Mike Cubbage will now be Intel's Senior Director of Connectivity Innovations.
Intel is set to keep the Killer brand and integrate it into its portfolio of products. I asked if there were any particular brands that Intel was keen on or not keen on - Intel's Eric McLaughlin, VP and GM of the Wireless Group stated that Intel is interested in all of them, especially in how they've been deployed so far and how Intel can scale them in more places and different ways.
I did ask a question about the integration, given how when Rivet/Bigfoot Networks was acquired by Qualcomm and then had to spin out again in order to drive the product, I was worried Intel might do the same. Mike told me that Rivet's Killer brand strengths back then, and even today, are in the PC and Gaming space, which perfectly aligns with what Intel is focused on. This is different to the previous acquisition, where is was more of a business portfolio play, but this time around Intel looks set on developing the Killer technology into a wide variety of products at scale, something which Rivet wasn't able to do previously.
Post Your CommentPlease log in or sign up to comment.
View All Comments
Samus - Thursday, May 21, 2020 - linkWell, the hardware always was a solution looking for a problem. That's why they just...stopped doing hardware. The magic was always in driver optimization, something Intel was always pretty good at from a reliability perspective but certainly not a performance perspective, especially with WiFi.
Ian Cutress - Thursday, May 21, 2020 - linkThe software isn't driver optimization. It's deep packet inspection, filtering, traffic recognition, and traffic prioritization. Rivet has a whitelist for the most popular apps, sure, but when you have a traffic stream where it doesn't automatically recognize the source, it will analyze it and filter it to an appropriate prioritization path
PeachNCream - Thursday, May 21, 2020 - linkIt's a waste of CPU cycles and RAM to inspect packets and give them priority leaving the PC when the rest of the route over dozens to thousands of miles of cables and optical connections will not do anything to any of that traffic that is different than any other data packets traversing the same links. Why bother wasting the time on the PC to run software that cannot ever actually offer a measurable benefit to the end user given how literally ever other node will treat that data without priority?
Deep packet inspection is fantastic when you are looking for malware or need to properly route data to the right end point in your data center. However doing it from a home computer will only get a priority packet to your ISP-provided junk router which doesn't care whatsoever and will push it up to your ISP that also doesn't care. The ISP will put it out onto the global network that (wait for it) doesn't care either. It'll get to where it ought to go over apathetic wires and apathetic routers, all of which totally disregard what your PC is doing with Killer software and you can enjoy exactly ZERO benefits from having wasted all that effort to begin with.
neuen - Thursday, May 21, 2020 - linkWhile I agree with you theoretically, I do not in practice. Disclaimer: I do not know if what Killer does is analogous to the anecdote I will describe as I have never used Killer however I guess it is in the same system of coordinates.
I had a problem with my ISP. Whenever I started downloading heavily, the websites barely opened, I could literally see 5% packet loss. While I could I understand it on 10 Mbit link, it was kind of strange to see this bad of a situation on the 500 Mbit link I had (I could have expected higher latency but not straight packet loss and websites not loading at all).
I have tried to liaise with them to introduce proper QoS on their side as it should be however it was just time spent in vain when you are dealing with a country-wide ISP.
Then I was finally fed with the problem and took matters in my own hands. I have set up ingress and egress shaping at 496 Mbit on my side and some basic automatic prioritisation (I did not prioritise specific apps, everything was on auto) - lo and behold losing 4 Mbit of bandwidth led to no packet loss and no latency change when I was downloading with maximum speed and thousands(!) of connections at a time - I intentionally made such tests.
Theoretically, as you say, my ISP should have solved this problem and I am wasting CPU cycles. Yet in practice, the theoretical approach was not achievable yet the problem was solved on my side by wasting some CPU cycles.
magreen - Thursday, May 21, 2020 - link@neuen Sounds like the real solution for you is to change ISPs. I doubt Killer is generating solutions for your use case. But do they have any other real use cases?
TheUnhandledException - Wednesday, May 20, 2020 - linkIs it though because a device not having killer ethernet is an added bonus for me. Hell I am still upset Dell removed the option of Intel wireless going with Killer garbage on some of their products.
Reflex - Wednesday, May 20, 2020 - linkMost of us just force installed the Intel driver for the chip those cards are based on and went our merry way.
Drkrieger01 - Wednesday, May 20, 2020 - linkNIC add-in cards aren't a bad idea for extreme enthusiasts. Good server based NIC's offer incredible performance and throughput, as well as higher than normal speeds (for example 10Gb offers 1/10th the latency of 1Gb). If Killer/Intel puts out one of these cards, it would be phenomenal for AMD users - especially Pro-sumers doing dev work with large files over networks. Gamers would benefit from an increase of 1-3ms ping (big whoop?), but that's about it.
Operandi - Wednesday, May 20, 2020 - link"10Gb cards offer 1/10th latency of 1Gb"???
Clearly you have no idea what you are talking about.
Samus - Thursday, May 21, 2020 - linkI lol'd too. Though the serialization delay is in fact a 1/10th shorter duration in microseconds (so Drkrieger01 is technically correct) this doesn't mean 10Gbe is going to have 1/10th the overall latency.
In fact I'm not sure why latency was brought up at all. The latency introduced at the physical layer is measured in microseconds, while standardized network latency is often discussed in milliseconds, so the Gbe or 10Gbe really have negligible impact on real world LAN latency because beyond the speed of the interface, there are a lot of other factors adding tremendously more latency.
Latency depends on many things like distance, queuing, CPU overhead, frame sizes, etc.