Date: May 5, 2020
Author(s): Rob Williams
Intel’s NUC family has evolved a lot over the years, but the core goals remain the same: to deliver a lot of performance in a small form-factor. The latest NUC, built around the Quartz Canyon platform, is bigger than most others, but it also packs a real performance punch – and even offers a bit of modularity.
Over the past year, we’ve taken a look at a handful of PCs from a workstation perspective, including the mobile-bound Razer Blade 15 Advanced notebook, and the Corsair ONE i164 compact tower. Both of those machines were impressive in their own right, and while we don’t want to spoil things too soon, we’re about to dive into another that can join the club.
It may be modest in appearance, but the Quartz Canyon NUCs have got some powerful hardware inside. Our sample includes an eight-core Xeon processor, 32GB of ECC memory, and an NVIDIA Quadro graphics card. Being a Xeon, there’s also full support for vPro, making this an intriguing platform for a managed work environment.
We’re going to provide plenty of photos later so that you can better understand how this NUC comes together, but despite its compact size, it’s not complicated inside. The motherboard and CPU are built into the same “Compute Element”, which can later be replaced with a faster model. Even if you don’t have plans to upgrade the Compute Element, the nature of this NUC will allow you to continue upgrading your discrete GPU – as long as it fits!
Over the past month, some publications posted looks at “Ghost Canyon” NUCs, focused on the gamer. Those specific builds included a GeForce RTX 2070 graphics card, and Core i9 processor in lieu of Xeon. Our “Quartz Canyon” model shifts everything to the pro side, where we move from Core to Xeon, GeForce to Quadro, and see the introduction of error correction memory.
The Ghost Canyon NUCs are targeted at the gamer who want a simple all-in-one system that still allows them to upgrade their memory, storage, or graphics solution when necessary. Once future Compute Elements release with updated processors and technologies, users will be able to upgrade and swap what’s effectively the heart of the PC themselves.
Quartz Canyon, by contrast, skews its focus towards the professional workstation market. As mentioned before, that involves introducing ECC memory and opting for a workstation CPU and GPU . With vPro support baked in, this NUC becomes an attractive option for end-users looking for a modest-sized but still upgradeable machine, or production houses that want to provide team members with a capable, reliable machine that fits neatly behind or next to their monitor.
As of the time of writing, Intel has launched five versions of its NUC 9 “Compute Element”, each of which has been made available as a complete kit. Of these five, three belong to the gamer-focused Extreme series, while the other two are branded as “Pro”. The Pro kit we’re looking at includes a Xeon E-2286M eight-core processor, which can reach a top Turbo clock of 5GHz. The other option is a six-core Core i7-9850H, which pulls back to a 4.6GHz peak clock.
While none of the Extreme models support Intel’s vPro features, both of the Pro models do – even the Core i7 (which ordinarily wouldn’t). All of these NUCs include both the i210-AT and i219-LM network adapters, but only the latter can be used with vPro’s management features. The -LM port is not specified on the unit itself, but on our tested rig, that reflected the bottom port.
Intel sells both Compute Elements and their full kit counterparts, so vendors can either integrate the elements into their own designs, a la Razer’s Tomahawk, or sell varied configurations of Intel’s “reference” design, a la SimplyNUC. If you actually want to use this NUC as a proper workstation, you will need a discrete GPU. You can either configure for one with a provider like SimplyNUC, or buy the standalone kit and put your own in. Just be warned: this platform has notable GPU size and power limitations.
For a professional workstation, this NUC has got the right looks. It’s unassuming, and compared to a mid-tower desktop, it takes up almost no space. Yet, despite its modest form-factor, the amount of power inside is significant. The reason the hardware doesn’t get too toasty owes some thanks to the breathable side panels, and blower fans for exhausting air efficiently out the back.
|Intel Quartz Canyon Workstation NUC (NUC9VXQNX)|
|Processor||Intel Xeon E-2286M 8-core/16-thread @ 2.4GHz|
5.0GHz Turbo; 16MB L3 Cache; HyperThreading
|Motherboard||Intel Quartz Canyon|
32GB ECC (16GBx2, 19-19-19 @ DDR4-2666)
|Graphics||Intel UHD Graphics P630|
NVIDIA Quadro P2200
|Storage||Intel Optane H10 1TB (HBRPEKNX0203A)|
3x M.2 Slots Total
|Audio||Realtek HD Audio|
|Network/Wireless||Intel I219-LM 1Gbps Ethernet|
Intel I210 1Gbps Ethernet
Intel Wi-Fi 6 AX200
|Connectivity||2x Thunderbolt 3|
4x USB 3.1 Type-A (Back)
2x USB 3.1 Type-A (Front)
4x DisplayPort (from discrete GPU)
SDXC Reader (UHS-II) (Front)
Audio In/Out (inc. TOSLINK) (Back)
Audio Out (Front)
|Dimensions||9.4″ x 8.5″ x 3.8″|
|Price||~$2,993.00 USD (As Tested)|
Alright, let’s get dirty and tear this thing apart:
The ease of getting inside of this NUC surprised us, since with most small form-factor PCs, significant effort will be required. On this design, two screws is all that stands between you and getting simple access to the hardware, and another two screws to swap either the GPU or the Compute Element.
The two screws in the fan enclosure come with an added bonus of not being removable, so you don’t need to be wary of losing them. This top piece houses two exhaust fans, which use a proprietary connector that connects with the chassis to power them. Ease-of-use is the name of the game here. That extends to the panels, which slip off easily, without the need for more screws.
Our NUC sample shipped with an NVIDIA Quadro P2200 workstation graphics card by PNY, which takes up just a single slot. Despite how complex CAD suites are, they’re generally not of the sort that demand a lot of GPU horsepower. For that reason, users of suites like SolidWorks, CATIA, or Siemens NX should stick to a Quadro here, and then take advantage of the freed-up 4x PCIe slot for either expanding M.2 storage, or to add in a discrete network card (if one is needed, of course). The lack of 2.5G or 10GbE network is slightly disappointing, but not entirely necessary for this form factor, but it’s simple enough to add additional ports through the spare 4x PCIe port if you stick with a single slot GPU.
Given how small this machine is, we expected some hassle when swapping out the GPU, but it turned out to be a simple affair. The first step in removing the graphics card is to remove two screws holding a guard in place (near the top), and then unscrew the graphics card to then lift it out. Naturally, given the lack of space here, you will have to finagle a bit to get a GPU in or out.
Removing the Compute Element is a bit trickier than the GPU, since there are some connections running straight to it from the chassis (the wireless and BT antennae). This is not really a complex setup though, so chances are you will be fine plugging everything back in once reinstalled. That said, it wouldn’t hurt to take pictures beforehand, because it is easy to overlook something.
The Compute Element is cooled by a single blower fan that plugs straight into a small motherboard header. That header can be seen to the right side of the board, along with a blue USB header for the front panel, and also an 8-pin power connector. At the top of the unit are two gold connectors that route the wireless antennae. These will likely be the final connectors you plug back in before putting the chassis back together.
This unit doesn’t look like much with its cooler installed, but once removed, we get a glimpse at the goods:
What we see here is a (nearly) complete system. Without a discrete GPU, the integrated graphics can be driven from the CPU, at the expense of performance. Two slots are available for the memory, which in our case hold two 16GB Kingston ECC DDR4-2666 SODIMMs. The 1TB Optane M.2 hybrid drive can be seen in blue, sitting beside an empty M.2 slot awaiting your additional storage. The included M.2 drive isn’t strictly an Optane drive, but a hybrid of 1 TB of NAND flash and 32GB of Optane cache.
We’re not sure how easy it is to remove the Compute Element’s heatsink, but we didn’t want to take a chance, since it is an embedded design. If you are not afraid to dig in a bit, you may want to remove this Compute Element and the discrete GPU once in a while just to blow air out – although after three weeks of use in an annoyingly dust-prone room, we haven’t seen any real buildup thus far.
If an additional M.2 SSD is installed inside of the Compute Element itself, then some blue protective film will need to be removed from inside the cooler, as it has a pre-applied thermal pad ready. A third M.2 slot can be seen in the final photo in the slider above. It’s directly beneath the Compute Element and graphics card, and is pictured with its heatsink installed.
For such a small unit, it’s nice to see three full M.2 slots available, but be warned that the chassis slot shares its bandwidth with whichever devices are installed in the other PCIe slots (using the 4x port will drop the 16x port to 8x mode). We really can’t imagine that being a problem even with a GPU like (up to) the RTX 2070 installed, but it’s important to bear in mind that the best M.2 performance will be from the Compute Element itself.
For those interested, the included Flex ATX power supply is an FSP FSP500-30AS 500W which carries an impressive 80 Plus Platinum rating. FSP has been making power supplies adhering to this form-factor for some time, so we’re hopeful that if you ever encounter yours burning itself out, you will be able to get a replacement. The fact that the unit carries a Platinum rating should mean chances of that will be very low. However, it is worth noting that despite the 10-pin ATX plug, it is not an ATX12VO power supply, as it still uses a 5V standby rail, so make sure the PSU you pick up to replace it adheres to this.
This NUC is challenging to drum up comparison benchmarks for, because it’s not as though we always have a bunch of prebuilds kicking around to retest. That said, we still wanted to test this NUC as thoroughly as we could, and compare to a couple of desktop CPUs, to both show its strengths and weaknesses versus full-blown desktop PCs.
It’s important to note that out-of-the-gate, these NUCs are not meant to go toe-to-toe with full-sized desktop PCs. The point of this NUC series is to offer as much performance as possible in as modest a form-factor as possible. So, it’s not going to reach the other chips we tested, but it does show how close we can come with a form-factor that’s at least 1/8th the volume of our regular test PC.
Note that each one of these tested CPUs hovers around ~$500 SRP.
In addition to CPU tests, we’ve also run a number of GPU tests, using the integrated graphics, the included NVIDIA Quadro P2200 workstation graphics card, as well as our own EVGA GeForce RTX 2060 KO that happened to fit in there no problem.
We’ll talk more about important results as we go along, but for now, let’s first analyze our CPU test results:
|Xeon E-2286M||Core i9-9900K||Ryzen 3900X|
|Adobe Lightroom||224 s||170 s||81 s|
|Arnold E-Type Render||434 s||370 s||238 s|
|Blender Classroom Render||421 s||291 s||201 s|
|Cinema 4D Animation Render||1,566 s||1,118 s||705 s|
|Cinebench R20 MT||3,427||4,777||7,030|
|Cinebench R20 ST||487||518||525|
|Corona Livingroom Render||270 s||205 s||140 s|
|HandBrake x264 Encode||278 s||200 s||137 s|
|LameXP Music Encode||379 s||279 s||219 s|
|LuxMark Food Scene Render||1,046||1,421||1,752|
|Premiere Pro 1080p Encode||696 s||481 s||375 s|
|Premiere Pro 4K Encode||734 s||524 s||385 s|
|Premiere Pro 8K RED Encode||358 s||256 s||169 s|
|Premiere Pro 8K ProRes Encode||172 s||122 s||81 s|
|Vegas Pro 17 Encode||627 s||423 s||344 s|
|V-Ray Flowers Render||476 s||346 s||245 s|
|Notes||Lower is better for all results listed in seconds.|
Higher is better for all other results.
Because it’s such an easy-to-understand number, we can compare the E-2286M’s Cinebench score to the 9900KS’, and see that Intel’s current top-end mainstream chip is 39% faster, putting the E-2286 chip in a similar score bracket as the Core i7-8700K eight-core (not listed). When it comes to full desktop performance, AMD’s latest ~$500 chip gives Intel’s equivalent a rough fight.
It’s also important to note that while this E-2286M expectedly lags behind the much more power-hungry socketable desktop CPUs, neither of those other options offer enterprise-level security features, as vPro is not found on those chips, and AMD ties its similar features to the Ryzen Pro line. Similarly, those other platforms do not include error correction memory, whereas our tested NUC platform does.
Overall, considering the small form-factor it’s smooshed into, the performance level out of this E-2286M is great. We spent a lot of time using this NUC as a standard PC, and importantly, it felt like a desktop. Over three weeks of arduous testing, this platform never let us down, and never behaved unexpectedly. We’ve yet to see a blue-screen-of-death, despite hammering the machine with as much varied testing as we could (even with three different GPUs). That also included many hours of gaming, not just work.
Speaking of gaming, that’s not something we did with the Quadro P2200 that was shipped with this NUC, as it was considered a “low-end” workstation card when it came out three-years-ago. Unfortunately, it’s also the fastest Quadro solution available right now that can fit inside of this NUC, which has led us to hope that the likes of NVIDIA and PNY will work harder to release a Quadro RTX 4000-class GPU in this form-factor down-the-road.
Here are a number of GPU tests done on the integrated graphics, P2200, and RTX 2060:
|UHD 630||P2200||RTX 2060|
|3DMark 1080p Fire Strike||1,466||10,913||18,928|
|Blender Classroom Render||N/A||316 s||140 s|
|BRAW Speed Test 8K 8:1||14 FPS||84 FPS||111 FPS|
|LuxMark Food Scene Render||747||777||NEED|
|Premiere Pro 1080p Encode||433 s||163 s||95 s|
|Premiere Pro 4K Encode||518 s||206 s||158 s|
|Premiere Pro 8K RED Encode||244 s||67 s||32 s|
|Premiere Pro 8K ProRes Encode||181 s||32 s||18 s|
|Vegas Pro 17 LUT Playback||31 FPS||41 FPS||43 FPS|
|V-Ray Flowers Render||N/A||647 s||343 s|
|VRMark Blue Room||116||1,044||2,322|
|Notes||Lower is better for all results listed in seconds.|
Higher is better for all other results.
Intel’s integrated graphics are suitable for regular desktop use, but not actual work that requires a GPU for acceleration. Interestingly, the IGP did come close to the P2200’s LuxMark score, perhaps as a result of that being a test built on top of Intel’s own Embree ray tracer. In all of the other tests, it’s clear that the RTX 2060 is the one to beat, as it simply demolishes the small Quadro in most tests.
If it seems obvious that the Quadro P2200 is not the GPU to include in this machine, there are a couple of things to bear in mind. First, with Quadro, you’re wrapping up an appropriate workstation PC for use where uptime and optimization for CAD-like workloads are of more importance than encoding or rendering. Quadros carry better customer support than GeForce, which is one reason why they cost more out-of-the-gate.
The next couple of tables will highlight the differences between Quadro and GeForce well in viewport use, such as those with the top-end CAD suites we mentioned before:
|1080p Viewport Performance|
|Intel UHD 630||Quadro P2200||GeForce RTX 2060|
|Blender 2.8 LookDev||N/A||17 FPS||46 FPS|
|Notes||Higher is better. All reflect “scores”, except Blender.|
|2160p Viewport Performance|
|Intel UHD 630||Quadro P2200||GeForce RTX 2060|
|Blender 2.82 Solid||16 FPS||79 FPS||101 FPS|
|Blender 2.82 LookDev||N/A||6 FPS||18 FPS|
|Notes||Higher is better. All reflect “scores”, except Blender.|
In the match-up between the NVIDIA cards, it’s clear that some performance uncapping is happening on the Quadro in select tests, since it manages to beat out the much more powerful RTX 2060. In more neutral workloads, like 3ds Max and Maya, GeForce comes ahead, but in CATIA, SolidWorks, and especially Siemens NX, Quadro use is hugely recommended.
Blender’s LookDev shading mode that was introduced with 2.80 won’t take advantage of Intel’s integrated graphics. If you change a scene to LookDev, it’s effectively going to lag not just the application, but the entire PC, and then not be usable at any point – at least with our tested scene. Without a GPU in this machine, it means Blender will always use the CPU for rendering, since the OpenCL for graphics option was hidden when we tried to test.
There are a couple of other things to bear in mind here, as well. Quadro P2200 is of the older Pascal generation, which means it does not have RTX features which will actually speed up rendering workloads quite significantly (where supported). Solutions like Octane, Redshift, Arnold, and V-Ray are all using RTX’s ray tracing acceleration, so if 3D design is important to you, and you opt for this Quadro P2200, you are doing yourself a major disservice.
Again, we hope to see NVIDIA release a future high-performance workstation card that can actually fit in this form-factor. PNY is obviously keen on this platform, as it’s the one that supplied the P2200 here, so it’d be nice to see some better options down-the-road. While we tested RTX 2060 in our NUC sample, there also exist RTX 2070s that can fit in, as well. The Quadro RTX 4000 shares similar specs of the RTX 2070, so… enough said.
|Power & Temperatures|
|RTX 2060 Temperature||32°C||80°C|
|CPU Video Encode Power||95W|
|CPU+GPU Video Encode Power||160W|
|SolidWorks Viewport Power||120W|
|Notes||Power measured from wall with Kill-A-Watt.|
Temperatures monitored with AIDA64.
To stress-test this NUC, we decided to swap the Quadro P2200 for the RTX 2060, in order to give it a proper push. The Quadro P2200 is low-end to the point where it’s not going to heat up the inside of the chassis too much, so us testing that as-is (when we wouldn’t even recommend such a modest GPU to most people) would feel like a bit of a disservice.
Over the course of a couple of hours, we ran 3DMark and various CPU workloads to see how toasty things would get. In the end, our GPU topped out at around 80°C, while the CPU hit 93°C. During regular use, the CPU is not going to run too warm; it’s only when it’s truly being taxed that things will heat up. What we see here is effectively on par with what we’d expect out of a notebook using the same hardware.
We didn’t test for noise, as we don’t have the proper test equipment here, but we can say that we were surprised overall with the modest amount of sound the machine sends towards your ears. You can absolutely hear it when it’s being pushed hard, but its peak is not as loud as we expected it to be. Of the Compute Element and GPU, it was the latter that was loudest when the going got tough.
Ultimately, if you have this NUC on your desk, you’re going to hear it hum away while you’re doing some heavy work, but it will be pretty quiet otherwise. If you’re a headphone user, the noise generated from the NUC is not going to be noticeable. If you’re a speakers user, you may want to keep the machine under your desk to muffle whatever noise there is a bit better. Depending on your tolerance level, that may not be necessary.
Throughout our testing, we plugged every port and tested every bit of connectivity that we could. The one exception is that we don’t have any proper TB3 storage devices here, but we did test our USB-C devices in the same port. Because virtualization was on by default in the EFI, we decided to install VMware Workstation 15.5, and see how things fared:
We know that this hardware is capable, but there’s still something about running virtual machines at good speed on a really small PC that is really cool. We copied over a barebones Windows installation and installed an MMORPG in there, and the performance was excellent. As the screenshot above shows, even the network speed inside of the virtual machine effectively matched what we’d see from the native OS.
On a powerful 18-core desktop, we usually allocate 4 threads and 4GB of memory to such a virtual machine, and we had to do nothing different on the NUC, since there are 16 threads available along with 32GB of memory. So, if you configure this with a healthy amount of memory, you can definitely take advantage of your virtual machines on this NUC.
As time has gone on, the “software” component of system reviews seems to get smaller and smaller, because nowadays, many companies have seemed to catch on that users prefer to have clean desktops, sans bloatware. Our NUC followed that philosophy to the point where it’s pointless to include a desktop screenshot, as we all know what a factory fresh Windows 10 looks like.
Some vendors include at least one software component for managing settings or monitoring temperatures, but there’s no such thing here. That can be considered either a good or bad thing based on your preferences, but because there’s not much to really “control” on the hardware front, there isn’t a huge a need for Windows software.
It is worth taking a look at the EFI, however:
Given that this NUC effectively sports similar hardware found in notebooks, we expected the EFI options to be limited. That really couldn’t be further from the truth, as there’s a bit of everything to tweak for power users. You even have limited fan control by way of telling it when to shut the fan off if temperatures go low enough (the default is 50°C).
It seems to be a bit of a trend that motherboards will ship with virtualization support turned off, but Intel has it fully enabled from the get-go here, which is great to see. As mentioned on the previous page, this NUC handles virtual machines really well, something definitely helped by the CPU’s fast clock speeds.
In a final test before wrapping up our testing, we reinstalled Windows 10 fresh to the NUC, to see if any oddity would creep up after we hit a fresh desktop. Fortunately, none did. Windows detected all of the hardware just fine, as Device Manager had no complaints about missing drivers.
When NUCs first hit the scene, it was hard to predict exactly where the series would go, but overall, we’re really glad to see that it’s evolved as it has. From the start, Intel focused on ease-of-use, and with a really small AIO PC that sits neatly next to a monitor, there’s a lot to like. We often hear of businesses that have implemented NUCs, and when we had the chance to visit with some Formula 1 teams a couple of years ago, we were (only sort of) surprised that fleets of NUCs could be seen in the garages.
The great thing is that NUC isn’t only one thing. Intel didn’t stranglehold itself by trying to make sure NUCs were only pint-sized. Even though this Quartz Canyon NUC is much larger than most others, it brings a ton of performance to the table, and offers upgrade paths for both the Compute Element itself, and the graphics card.
It’s worth pointing out again that this Quartz Canyon NUC is not the only potential design. Should system builders want to get in on this platform, they can integrate Compute Elements into their own designs, such as the Razer Tomahawk we mentioned on the first page (although that uses the Ghost Canyon variant).
The truth of the matter is, unless Intel makes major strides in being able to pack more cores into such a small form-factor, it’s going to be the GPU that will be in most need of upgrading over time. As we covered before, it’s really unfortunate that the highest-end Quadro to fit in this NUC is a three-year-old option that was destined for the low-end part of the market from its introduction. We’d love to see a Quadro RTX 4000-equivalent fit in here, so we can cross our fingers that more options will come in time.
It could be that in time, we’ll start recommending Intel’s own graphics cards for a solution like this, as the company is hugely focused right now on developing powerful discrete options. Those are still a ways off, however, and when they do drop, they will almost certainly be tailored to the datacenter before our lowly desktops. As for AMD, its highest-end workstation solutions that fit in here is still lower-end than even this low-end NVIDIA Quadro.
It’s for that reason that we’d suggest most adopters of this platform to opt for a GeForce RTX 2060 or RTX 2070. The exception is where CAD use is concerned. If you use an CAD suite like SolidWorks or CATIA, which are hugely tuned towards workstation hardware, you’ll want to opt for Quadro. There are cases where even the lowbie P2200 outperformed our gaming card, simply because NVIDIA has fully unlocked and optimized performance on Quadro in these suites.
Because of the heat factor, we wouldn’t suggest using the Xeon CPU for rendering, unless you keep to a modest GPU. Nowadays, rendering is hugely sped-up with GPUs, and especially so with NVIDIA’s ray tracing acceleration. Any heavy 3D content in general would also run far better on a GeForce like the one we tested, over the small Quadro P2200.
Ultimately, this is not a platform for everyone, and its higher price-point ensures that. This is an optimized platform that’s designed to be supremely stable, and still allow some upgrade paths. For the enterprise and work environment in general, the vPro features of our NUC 9 Pro in particular will be useful, as will be the error correction memory. If you don’t need vPro but still like this platform, you can also consider the NUC 9 Extreme, which has options as low as the quad-core Core i5-9300H.
This kit is a bit expensive to jump into, but we’re left impressed with the overall package. We found ourselves surprised on multiple occasions by how easy it is to work with this machine, even when needing to tear out the entire Compute Element. Working with small form-factor PCs can sometimes be arduous, but not here. The biggest downsides this platform has are outside of its control – namely with regards to not being able to fit a high-end workstation GPU in here.
As always, if you have any question unanswered, please leave us a comment!
Copyright © 2005-2020 Techgage Networks Inc. - All Rights Reserved.