A BOXX Full of Workstations: The BOXX FLEXX Server System

BOXX Technologies has sponsored this post.

In a world of remote work, remote workstations make a world of difference. To prove it, workstation manufacturer BOXX Technologies sent us their latest product: FLEXX. The BOXX FLEXX is a modular rack system that can accommodate up to ten modules, each providing the performance of a standard BOXX workstation, and each of which is remotely accessible to a team of users with the right software enablement.

In this review, sponsored by BOXX, we take a hands-on look at the FLEXX system to see how it works.

Video Overview

For a visual introduction to the BOXX FLEXX, check out our video review. For all the details of our experience with the FLEXX, continue reading the article below.

What is BOXX FLEXX?

The BOXX FLEXX. (Image courtesy of BOXX.)

You may know BOXX best for their high performance desktop and mobile workstations, such as the BOXX APEXX Enigma S3 or GoBOXX SLM 15. With FLEXX, BOXX puts that high performance into a blade form factor.

The rack-mountable FLEXX enclosure has space for up to 10 compute nodes, a.k.a. modules, measured in what BOXX calls a vertical unit, a.k.a. VU. Whereas a typical rack unit (denoted RU or simply U) measures height up a rack, a VU measures width across a rack. This is one of those times when a picture is worth well over a thousand words:

(Image courtesy of BOXX).

FLEXX nodes are either 1 VU or 2 VU in size, and nodes can be mixed and matched across the FLEXX enclosure. With 10 VU of total space, up to 10 nodes can be inserted side by side for ten separate workstations. Nodes are hot-swappable, so they can be added or removed without disrupting neighboring nodes. There are four series of FLEXX nodes currently available:

Overview of FLEXX nodes. (Image courtesy of BOXX).

For our review, BOXX shipped us both a FLEXX S1G (Intel-based workstation node) and FLEXX A1G (AMD-based workstation node) with the following specs:

Other than the processors, the A1G and S1G each have slightly different shapes and port options due to their different motherboards. The A1G is flush against the back of the FLEXX enclosure, while the S1G is partially recessed (see graphic above). The A1G has three RJ-45 ports, two USB Type-A and a VGA connector, and supports IPMI for server management. The S1G has three DisplayPorts, one Thunderbolt 3, two USB-A and one RJ-45. Both nodes can be powered with a single thick 1625W cable, though a fully-loaded FLEXX requires four. The power adaptors are built in to the FLEXX enclosure to provide a common power backplane.

For the purposes of this review, we conducted all benchmark tests with the A1G node.

Connecting to FLEXX

The FLEXX would just be a big metal box stuffed with transistors collecting dust in your datacenter if there wasn’t a way to connect to the workstations inside. Users can connect to their FLEXX nodes with the Teradici PC-over-IP (PCoIP) protocol, which streams pixels from the node to a client device. That client can be extremely lightweight. Users can tap into a powerful 12-core workstation from nothing more than an inexpensive Chromebook.

Connecting to the FLEXX on an internal network is extremely easy. There are two pieces of software involved: the Teradici Graphics Agent installed on the FLEXX node, and the Teradici PCoIP Client installed on the client. Connecting the two is as simple as entering the IP address of the node along with its Windows username and password. You will, however, need to license the Teradici software, which costs around $240 per year (for five users, the minimum amount).

The Teradici PCoIP client.

More useful, but also more involved, is connecting to the FLEXX on an external network. This process requires you to set up two dedicated servers: one for the Teradici PCoIP Connection Manager and Security Gateway, and one for the third-party Leostream Connection broker, which together work to authenticate users and establish PCoIP sessions without a VPN connection. These components must also be licensed. I configured each of these on a separate virtual machine running on a couple of old computers in our office, but to a networking novice this process was not just cumbersome, but cumberlots. My advice: leave the IT work to the IT pros.

IT guys and gals, keep up the good work. And have you tried turning it off and on again? (Image courtesy of Teradici.)

The FLEXX Experience

Once the PCoIP connection is configured, using a FLEXX workstation is as simple as opening the Teradici client and logging on. You can’t log on to more than one node at a time from the same client device, but you can quickly disconnect from one node and connect to another if needed. For instance, you could set up two rendering jobs on two nodes and halve your wait time.

One of the biggest concerns of a remote workstation is latency. It’s one thing to have a bit of lag if you’re editing text, but when you’re working in highly visual 3D applications, you need a certain level of responsiveness. The PCoIP protocol streams only pixels, and only the delta from one frame to the next, so there is no exchange of whole data packets. The pixel stream is also compressed and secured with AES 256 encryption. In theory, this adds up to “a high-performance user experience for knowledge workers and power users using even the most graphics-intensive applications,” according to Teradici.

In practice, of course, the network itself plays a large role in the PCoIP protocol. Case in point: if the network is down, the workstation is down. However, this is the same problem faced by every cloud service, and is something of the digital equivalent of electric vehicle range anxiety. How often are you really going to take that 500-mile road trip with no way to stop and charge? How often does your internet really go down long enough to meaningfully impact your productivity? Of the many users of cloud services who I’ve asked about this issue, none have found it a problem in practice.

My experience connecting to the FLEXX nodes revealed very little lag, as promised by Teradici, who suggested that round-trip network latency under 100ms should provide for a smooth experience. I could tell the difference from a local workstation, but only if I really tried. Even when running highly dynamic 3D benchmarks, I never found lag to be a problem.

With the PCoIP protocol, each FLEXX node is capable of supporting up to four 4K monitors on the client device. Although adding monitors theoretically increases the amount of pixels to stream, in practice—since PCoIP only transmits the changes from one frame to the next—adding extra monitors won’t necessarily cause any extra bandwidth problems. Speaking of which, you will want at least 10Mbps to ensure a smooth connection.

Using the Teradici PCoIP client to access the FLEXX A1G node.

You can also pass USB devices through your client to the FLEXX node. Common interface devices such as mice and keyboards are automatically passed through, but certain devices such as 3D mice and graphics tablets must be configured each time a PCoIP connection is established. If these devices are connected to the FLEXX node, they will be unavailable to the client device. USB flash drives can be added in the same way, though users will have to explicitly authorize each device.

Testing the FLEXX A1G

Like BOXX’s standard desktop workstations, the FLEXX nodes cater to design professionals. CAD users, for instance, need high single-core CPU performance, as most CAD applications are single-threaded. For those doing CPU rendering, a high core count is crucial.

The FLEXX A1G module we tested hits the mark, performing just as well as we’ve come to expect from BOXX in our standard benchmark tests. We present the results below. For reference, the results are compared to the BOXX APEXX Enigma S3, a BOXX workstation we reviewed in 2019 with the following specs:

Though this review is sponsored by BOXX, all benchmark results were obtained directly by engineering.com (unless otherwise noted).

SPECviewperf 13

SPECviewperf 13 is a benchmark of viewport performance in a number of popular 3D modeling applications including 3ds Max, CATIA, Creo, Maya, NX, SOLIDWORKS and more. It is primarily a GPU test, but even with less than half the raw GPU power of the APEXX, the FLEXX A1G achieves comparable scores. For users who want an extra graphics boost, the P2G FLEXX node accommodates dual graphics cards.

SPECworkstation 3

SPECworkstation 3 is a full-system benchmark which comprises the viewport tests of SPECviewperf 13 as well as dozens more tests of CPU, GPU and storage capability. The results are organized by industry segments including Product Development and Life Sciences. The FLEXX A1G emphatically proves its workstation mettle. Again, the area that is most ripe for improvement is GPU compute, which would see a definite boost from the dual-GPU FLEXX P2G node.

*Estimated value. See below for details.

*Estimated value. While six out of seven SPECworkstation Energy subtests were successful, one failure (Poisson) prevented the calculation of an aggregate industry score. A rough estimate of Poisson and aggregate scores was obtained by comparing successful subtests to those of the APEXX. SPEC does not disclose official aggregate formulas.

SOLIDWORKS Benchmark

The SOLIDWORKS benchmark measures the time required for several categories of SOLIDWORKS operations, lower time being the better outcome. The FLEXX A1G’s results demonstrate that SOLIDWORKS users should feel right at home with a workstation that is not right at home with them.

Cinebench R20

Cinebench R20 is a CPU rendering benchmark that, in general, awards higher scores to higher CPU core counts (though tests are run for both single core and multi core rendering). As such, it’s no surprise that the 12-core Ryzen performs exceptionally well in this benchmark.

*Scores from cpu-monkey.com and not independently verified by engineering.com.

Conclusion

Going from a local workstation to a remote workstation is a bit of an adjustment. For one thing, you will have to log on to two computers before starting your work, the client and then the FLEXX node. For another, you will have to configure certain peripherals (such as 3D mice, graphics tablets and flash drives) on a session-by-session basis. You may have to put up with a spurt of lag now and again depending on your network connection, and you won’t be able to work offline at all.

But there are many advantages to remote workstations such as the BOXX FLEXX. You are not tied to a single location to do your work, since you can tap into powerful hardware from anywhere and with whatever minimal device you have at hand. To anyone who was alive in 2020 up to the present day, this benefit needs no further elaboration. You can also easily swap between different workstations (i.e. nodes) to use whatever befits your current workload, giving users even more flexibility with FLEXX (I’m starting to think that may have inspired the name).

Another advantage of a remote solution like FLEXX is that it offers centralized IT administration, making upkeep a lot easier than for a distributed fleet of local workstations. But if you’re not interested in managing your own datacenter, BOXX also offers a service called BOXX Cloud that provides the exact same FLEXX nodes in the form of a managed service. Unlike many prominent cloud desktop services, each BOXX Cloud node is dedicated to a single customer, so there are no performance drawbacks from portioning and sharing compute resources.

Finally, as our benchmarks show, the FLEXX nodes are every bit as capable as their desktop counterparts. If you need true desktop-level performance that you can tap into from anywhere, a remote workstation solution like FLEXX is your best bet.

To learn more about the BOXX FLEXX, visit BOXX.com.