Over the summer, PC Labs had a opportunity to test and review the Intel NUC 11 Extreme Kit, dubbed during development with the evocative codename “Beast Canyon.” The new top-end offering in Intel’s Next Unit of Computing (NUC) family of DIY-minded small desktops, we awarded the PC our Editors’ Choice honors, and the product has finally come to market this month.
Intel’s NUC desktops have long been known for their compact designs and pace-setting innovation among small PCs. With Beast Canyon and 2020’s NUC 9 Extreme “Ghost Canyon” models, however, Intel has nurtured that Extreme side branch of the NUC family tree, set apart from the trunk in two big ways: chassis designs that incorporate discrete graphics cards, and with their core computing built on the Intel Compute Element architecture.
We first got wind of Compute Element at Computex 2019. At the time, Intel demonstrated it as a small, modular implementation of a PC’s central compute silicon: the processor, the core system logic/chipset, the memory, and in some implementations, even the storage, all in one small SSD-like enclosure. It evolved from that first sighting (the design was meant for OEM laptops) into its appearance in the Ghost Canyon NUC desktop: a graphics-card-like module, containing the main guts of the system, that plugs into a small PCB—a mini-motherboard or “baseboard”—in the NUC chassis. The Compute Element concept has intrigued us ever since, as a model of modular computing with the potential for easy desktop upgrades down the road: Swap a card, and gain a new CPU, chipset, and more.
The Beast Canyon NUC uses the latest rev of Compute Element, based on 11th Generation Intel Core processors. In our wide-ranging discussion that follows around this nifty new bare-bones PC, we touched on the nuances of Compute Element, lots of detail on the inner design of the new NUC, and what happens when NUCs Get Big. (The interview has been edited slightly for brevity and clarity.)
PCMAG: I’m John Burek, the executive editor covering the hardware beat. A colleague and I tested the latest NUC, the NUC 11 Extreme, and I’ve been reviewing (or directing others in reviews of) the Next Unit of Computing platform for many years. So I’m pretty familiar with the product line. Could you tell me your name and role in the project?
FAISAL HABIB: Sure, Faisal Habib. Simplest way to pronounce my name is “vessel” with an F. I’m the segment director within the Systems Product Group, which is essentially the NUC team. And within my team, obviously, there’s a group of people who focus on the gaming segment, which is our shining star as this overall segment continues to grow spectacularly.
PCMAG: Good to meet you. So you were involved in the initial design and initiation of NUC 11 Beast Canyon, correct? I recall the previous generation, which was the first large, Extreme, video-card-upgradable NUC that we looked at. I’m probably going to get my Canyons wrong, but that was…Ghost Canyon?
FAISAL HABIB: Ghost.
PCMAG: Ghost Canyon, okay. I always get Ghost and Phantom mixed up! So that generation brought in support for the Compute Element as well, right? Some of my initial questions about this would be around Compute Element itself and how Compute Element works within the “NUC-o-sphere,” if that’s a term we can use. With upcoming generations of CPU and with a different CPU architecture coming up soon, what does that portend for, say, being able to upgrade Compute Elements? How much of that stuff is handled on the Compute Element itself, and how much of it is on the baseboard underneath? And does Alder Lake’s [12th Generation Core] coming architecture change have a bearing on that?
The first one you saw from a gaming perspective or in the workstation space was Ghost Canyon. And we expect to continue to have something similar as we go into future CPU generations like Alder Lake. And the value proposition here is that all of it is within the Compute Element module, right? So ultimately what it connects to, at the baseboard level, is a PCI Express x16 slot—it’s standards-based, we’ve separated the CPU stuff onto that module so that you can, from a customer perspective, just plug it in and it would just work. That allows for a longer generational upgrade path than you would have otherwise gotten. You couldn’t potentially have upgraded a “Tiger Lake” product or an “Alder Lake” product without such modularity.
Talking the Modular Approach
PCMAG: That makes sense. So the chipset-level stuff is all on the module itself? It’s happening [outside] of the baseboard, in other words? [The “baseboard” is the equivalent of the motherboard in the Beast Canyon NUC, a PCB which the CPU Compute Element module and GPU plug into.]
FAISAL HABIB: Exactly. There are some feature differences you would have seen from a Ghost to Beast [Canyon PC], like PCI Express Gen 3 versus PCIe Gen 4, which are CPU architectural differences, which don’t translate as easily because that is a baseboard conversation. But we are enabling the market with providing those baseboards, as well to say, hey, you know, if you really want to go in for a full upgrade, or if you want to work with Cooler Master and other chassis vendors who are playing in the ecosystem, they will provide those upgrade paths as well. So our ultimate goal is to create a vibrant ecosystem, which as you know, like in the gaming space, you’re spoiled for choice on what kind of chassis, what kind of cooling solutions, what else you could provide. So as we go down the journey of modularity, that’s where we envision we end up.
PCMAG: So what were the changes on the baseboard from NUC 9 Extreme to NUC 11 Extreme?
FAISAL HABIB: The biggest change is obviously PCIe Gen 3 to PCIe Gen 4 support. We took a lot of feedback from our customers, essentially what we do with every generation. When we launched Ghost Canyon, or NUC 9 Extreme, we heard a lot of, “I wish you had done it this way,” or something similar. A good example is the M.2 slot on the baseboard of Ghost Canyon; customers said that was too hard to get to. So if you look at the NUC 11 Extreme, it’s actually at the bottom so you can just open the bottom of the Beast Canyon and get to it. Of course that would not be there on a NUC 9. So if you took a Compute Element from here and plugged it there, will that slot work? Yes. But will it be different? From a feature perspective, those would be the things you would miss more.
Ultimately, what you end up worrying about is not only performance, but also the other customer UX things, right? We ultimately want to drive a modular form factor, which is also super easy for our customers to upgrade. Our vision is a tool-less kind of process. Usually, small-form-factor PCs are very hard for assembly, for integration, which is one of the reasons why most of the market hasn’t gone in that direction. If we can solve that, it will help the market move in that direction.
Compute Element: CPU Choices
PCMAG: Yes, having built a lot of small-form-factor PCs myself, I understand those kinds of limitations, all the cabling and other issues. In terms of the Compute Element itself—as I understand it, and correct me if I’m wrong here—on the NUC 11 Extreme, you have the option for a Core i7 “B” or a Core i9 “KB” system. Could you speak briefly to the choice of the “B” CPUs and what makes those different from the straight desktop chips?
FAISAL HABIB: We’ve traditionally chosen mobile silicon on the NUC roadmap. And we felt like Tiger Lake-H served well in a small form factor. So we were sticking to that lane, and then we chose the highest-performing CPU within that, which is traditionally what we do in the NUC space, right? Because our goal is to show the best of the Intel platforms. So Tiger Lake-H KB is the highest-performing Tiger Lake-H, which is why we chose that CPU. It’s hard to do an equivalence there because there is no Tiger Lake-H desktop CPU to measure against. The decision was basically, we were choosing the CPU, and then we chose the best of.
PCMAG: Got it. I think because of the way the CPUs are named and the way they’re implemented on the Compute Element, it looks like they might be the full desktop versions. But they’re Tiger Lake-H-based?
FAISAL HABIB: Correct.
PCMAG: Understood, okay. It’s just the consumer understanding of the difference between mobile and desktop [chips], and then the differences within mobile, and then the difference when it’s implemented in this unique form factor. It’s something that needs explaining.
FAISAL HABIB: Absolutely. And it is a journey for us as well to communicate that. Honestly, we rely a lot more on publications like yours to help us educate the customers, because a lot of our marketing is primarily through you guys.
PCMAG: In terms of the Compute Element itself, that requires a lot of explanation in terms of the concept, how it works, what the upgrade path is, and what have you. For someone who might be used to looking at, say, a Core i9 in a desktop environment where it’s almost invariably liquid-cooled or has a very robust sort of cooling solution on it, can you talk a little bit about how the cooling evolved in the Compute Element? How is this possible in this small a module?
Thermal Beast: Keeping Things Cool
FAISAL HABIB: Yeah, great question. A lot of that conversation happens within our engineering team, the alignment of our drive to make the smallest possible [PC] while still delivering the kind of cooling, thermal performance, noise vectors that we need to maintain. Our biggest challenge is always, a NUC sits on a table, a gaming desktop sits on the ground. So the noise vectors are very different. We have a higher bar to cross. So we spend a lot more time tuning that.
Our ultimate goal as we see it, and we’ve started to see some of this happening with influencers already taking our current NUC 9 Compute Element and liquid-cooling it and doing other things around it, is that we absolutely expect the enthusiast community to drive there. We are looking at, “How do we do more? How do we accommodate such enthusiast aspirations? Could you enable a Compute Element to be liquid-cooled? Could you enable others to do so?”
We don’t really need to go there [ourselves], but we need to enable the market to be able to do that—because this community loves that kind of differentiation, driving the platform to the extreme. So from our perspective, it’s two vectors. We want to continue to drive the best air-cooled solutions; that is how we show our platforms. But we hear the community. And we’re trying to [figure] how do we make it such that our industry can take advantage of these kinds of technologies.
One of our key vectors on that front: We want to not deviate too much from standards, which is why we chose the PCIe x16 connector on the base rather than designing something custom. We’re trying to do something similar for if we ever get to a water-cooling-capable Compute Element: How would we enable the same keep-out zones, the same holes that they use on Mini-ITX and ATX so that the cooling industry doesn’t need to design a whole new solution? Because that always makes it harder for adoption.
PCMAG: That makes sense when you think about a typical off-the-shelf liquid-cooling solution, where you have a heat sink that needs to bolt onto a socket. At least in the current implementation of Compute Element, there is no socket. Or if there is, it’s all underneath the shell.
FAISAL HABIB: Right. Yes, there is no socket.
PCMAG: It’s a mobile chip, right. On that note—with the NUC 11 Extreme case, you have three large fans across the top. Can you speak a little bit about the cooling concept there? Because you have your Compute Element, which has its own internal cooling and is doing a blow-through-and-out-the-back approach, right? Then you have a video card, which presumably has its own cooling configuration, depending on the card. It may be a blower design—straight through—or out the side.
How do the three fans on top factor in? Is it just about more air volume? Is it directed mostly toward the Compute Element? What was the thinking there?
Make Way for the Graphics Card
FAISAL HABIB: It’s actually driven more toward the GPU. As you go into a full-length GPU support model, the biggest power draw is the GPU, right? If you imagine in a GeForce RTX 3080 kind of model fitting into that chassis, trying to cool that is challenging. And there is a lot of variety in how thermal solutions are bolted onto graphics cards. If you look at [Nvidia’s] Founders Edition, it works very differently than some of the cards from other vendors. Our goal was to build a thermal solution that works for most of them, or works as best it can.
That’s why you’ve seen some of the changes in Beast Canyon, like we’ve put a shroud on how you cool the Element so it’s pulling air from the outside and pushing it into the card directly so that it’s not impacted by cards like the Founders Edition, which actually blows warm air inward. So that is where we wanted to make sure the chassis would provide the cooling requirements, similar to what you would expect from an ATX kind of solution. It’s also why we did so much mesh on the side to allow for the most amount of air cooling.
PCMAG: Maybe I’m misremembering, but one of the recent Nvidia Founders Edition cards had fans on both sides, if I recall correctly?
FAISAL HABIB: The RTX 3080s have it on both sides, yeah.
PCMAG: Does that create a challenge for that sort of environment? Because you’ve got a very different airflow model, and also the Compute Element card, which in most conventional systems wouldn’t be right there?
FAISAL HABIB: Yeah, it did. That was one of the things we looked at and said, “Look, we need to create a chassis solution that accommodates customer choice.” We can’t be saying, okay, here are the [only] three cards that work. That’s not how we play in the industry.
So that is one of the reasons why some of the solutions are, you know, an excess number of fans compared to what the NUC 9 Extreme had, and so on. But having said that, some cards will perform differently depending on how optimized their cooling solutions are. But that’s true for any chassis, any kind of thermal use.
PCMAG: Is there a particular “airflow model” of card that you would say works optimally in that chassis? I know you’ve tried to accommodate them all.
FAISAL HABIB: If it were a standard airflow model, like pulling in from the side and just going straight outward from the back, that model is best. And by and large, most add-in cards follow that model, other than I think the Founders Editions.
Other Internal Issues
PCMAG: Got it. A quick aside, back to something we mentioned earlier: Will the shift to DDR5 ultimately have any bearing on upgradability, or is that again sort of at the Compute Element level?
FAISAL HABIB: It’s at the Compute Element level. As we go to Alder Lake, and assuming we implement something like that with DDR5, then those DIMM modules are on the Element. So if you had bought an Alder Lake solution, then it would have it.
PCMAG: Fair enough. And I guess I have the same question with SSD and M.2: So some of the modules are on the Compute Element and whether it was Ghost Canyon or Beast Canyon, you had one on the baseboard. Is there any sort of a limitation in terms of PCIe Gen 3 versus 4? I know that with NUC 11 Extreme, you moved to a PCIe 4-capable baseboard, but is there any sort of limitation in terms of using certain SSD types from one generation to the next?
FAISAL HABIB: Not from a PCIe perspective. So between NUC 9 and 11, obviously you wouldn’t use a Gen 4 SSD on the NUC 9 because you wouldn’t get the performance. But PCIe Gen 4 shouldn’t be a problem for NUC 11 across any of those stacks. The only place where you would see some differences, which was also true on Ghost, was if you happen to use the SSD on the baseboard, then your GPU slot becomes a PCIe x8.
FAISAL HABIB: Now, traditionally that hasn’t had too much of an impact. But we document that very clearly for our customers, who, in the extreme space, absolutely care about every percentage of performance [and] have to make their choices accordingly. There’s obviously some impact in what kind of PCIe Gen 4 SSD you choose, in terms of its heatsink and stuff like that. There are a lot of M.2 SSDs out there with really fat heatsinks which may or may not fit. So would they work? Yes. Would you be able to close the system?
PCMAG: Or experience throttling, or things of that nature.
FAISAL HABIB: Exactly.
Right-Sizing the NUC
PCMAG: Here’s just another sort of aside question, in terms of the marketing for the term “NUC.” So the name has been around for quite a few years, has stood for “Next Unit of Computing” for as long as we remember. But it has largely been associated, I think, in customer minds with small things, right? And NUC in the last year, or couple of years, has expanded into larger, expandable desktops and also into the notebook space, though I don’t know how much consumers are really aware of that initiative.
What are the thoughts around the NUC concept as a brand and the potential for—maybe not customer confusion, but you probably want it to mean a lot of things, and it has meant one thing for most of its life.
FAISAL HABIB: Yeah, I was kind of prepared for that [question]! We spend a lot of time reading all the reviews our friends in the press publish, right? And the comments on them. So we hear what you’re saying. Ultimately for us, the “Next Unit of Computing” is a term view for how we drive innovation. Now, traditionally, we’ve been very careful on how we do form-factor innovation. As we feel a little more confident as we hear back from our customers on where they would like to take the market, or where they see pain points, our vectors of innovation change.
Our 4-by-4-inch line obviously continues to be the bedrock of our product line. We still drive a lot of innovation there, to a point where, if you were to look at the PCB now, you wouldn’t find any green spots. It’s completely filled, right? Because it serves so many different segments and so many places, you just fill [the space] up. But then we added the performance line, like our Skull and Phantom products that you’ve been seeing lately. That’s a different vector, which truly is the smallest possible discrete graphics solutions. That’s also a swim lane in which we plan to continue.
But as we move further up—in terms of really driving true desktop-class gaming performance, where you want add-in cards and stuff like that—that’s a relatively new place for us. But the key conversation we need to continue to drive is the “Next Unit” part of that, which is not the chassis. The Next Unit part of it is the Element inside it, which is where we want to be driving that innovation.
I’ve seen commentary out there like, “This is now eight-liter, which feels like a Mini-ITX product…why?” But that is a way of us helping the industry come to market and showcase how we drive this kind of innovation. But what we really want to do is work with our ecosystem partners to go build out an ecosystem around the Element. And that is where we’re spending a lot of time with our partners.
PCMAG: In terms of the Element itself, I know you can’t talk about unannounced products, but could you see other form factors or situations in which this Element could be deployed that we haven’t seen today? Just by the nature of what it contains and sort of the architecture around it?
FAISAL HABIB: The short answer is yes, we are already seeing some of it. There are customers who are taking the Ghost 9 Element and doing creative things. There are people who are looking at doing microservers. There are people who are looking at doing very different kinds of things. They’re all pretty kludgy right now, but it is important for us as we look at those usage models to decide, okay, if that is where customers want to take this, then our job is to enable those features so they can.
I would say it’s pretty nascent at this stage, but we absolutely see our customers taking it into new and innovative places which we never thought about, which is awesome. And we want to figure out how we continue to enable that kind of innovation.
PCMAG: Yeah. Are there…size limitations, because, obviously, you’re putting a relatively high-powered CPU onto the existing Compute Elements here? Is there any reason that the [Compute Elements] can’t go significantly smaller with lower-powered CPUs, or are there just fundamentals, you need to have a certain amount of acreage of chips on the PCB that you can’t go much smaller?
FAISAL HABIB: You can go smaller. We were trying to think forward in terms of keeping a certain GPU size, so we were trying to follow the PCIe spec for a graphics card. Therefore, the size was defined by that, which actually is not bad for us, because it allows us to do a self-contained unit for RAM and SSDs and stuff as we go forward.