Intel has given its upcoming gaming graphics cards a name — Intel Arc. The first generation of Arc Alchemist graphics cards is set to arrive in early 2022, and Intel confirmed that three future generations are already in the works. We’ve been waiting a long time to see what Intel has in store for gamers, and now, we have a clearer, if still slightly smudged, picture of what’s coming up.
Intel Arc is the brand that Intel is using for all gaming hardware, software, and services moving forward, and the company confirmed that it will span multiple hardware generations. Here’s everything we know so far about Intel Arc Alchemist, the release date, and how the cards will perform.
Intel Arc is the branding that Intel is using for “upcoming consumer high-performance graphics products.” The name encompasses multiple generations of graphics cards targeted specifically at gaming, and you shouldn’t confuse it with Intel Xe. Arc products — at least the ones we know about — use the Xe architecture, but not all Xe products live under Arc.
Intel Xe is the architecture for all of Intel’s recent graphics products, including the integrated graphics in Tiger Lake laptops and data center GPUs like Ponte Vecchio. Xe is a scalable architecture that Intel can shrink or expand in a single package to accommodate multiple different applications.
Intel Arc products are based on the Xe HPG. Intel is using this same architecture for mobile and desktop applications, starting with the first generation of Intel Arc cards.
With the naming out of the way, we can talk about the Intel Arc products we know about. The first-generation cards were formerly known as DG2, but they now have the code name Alchemist. Alchemist cards are the only ones we even remotely have information about at this point, so we’ll be focusing on those here.
Intel revealed the code names for upcoming generations, too: Battlemage, Celestial, and Druid. For these generations, all we know right now are the names. However, we know that Battlemage will follow Alchemist, Celestial will follow Battlemage, and so on.
Intel hasn’t revealed pricing information for Arc cards yet and only offered a vague hint at the release window. Keep in mind that Arc is the branding for all Xe HPG graphics products, so beyond this point, we’re talking about the upcoming generation of Arc cards — Alchemist.
For release, Intel says the first cards will arrive in the first few months of 2022. Information is sparse at this point, so there’s no saying when in that time frame the cards will release or how Intel will handle the rollout. It’s possible that Intel will release cards to system makers and laptop designers first, or everything could hit at once. We don’t know yet.
Previously, rumors suggested that Intel would release the cards at CES 2022. That event happens in January, so it’s possible that Intel will launch the cards then, with general availability following shortly after.
Unfortunately, we know even less about pricing. We’re purely speculating, but if rumors are to be believed, the flagship card should cost around $600. That’s in between the RTX 3080 and RTX 3070, which it looks like the top card is targeting. Intel isn’t a stranger to abnormally high prices, however, so it could be higher.
Some rumors suggest Intel is targeting an aggressive price to steal some gamers from Nvidia and AMD. We still don’t know how aggressive Intel is willing to be, though, and the GPU shortage has driven up the cost of components. Intel’s cards could be more expensive purely because the cost to make them is higher.
Intel hasn’t announced what partners it’s working with for Arc Alchemist graphics cards. However, speculation by other outlets points to Gigabyte, Asus, and MSI as possible partners. Given Intel’s history of working with motherboard makers for its CPU launches, these partners are possible, if not likely.
Intel gave us a deeper look at the Xe HPG architecture powering Alchemist cards at Architecture Day 2021. However, the company still didn’t announce or confirm any specs. So far, we know that the flagship card will feature 512 execution units across 32 Xe Cores. Otherwise, we have to rely on rumors and speculation for the the time being.
|GPU||Execution units||Shading units||Memory||Memory bus|
|Intel Arc Alchemist a170 or a700||DG2-512EU||512||4,096||16GB GDDR6||256-bit|
|Intel Arc Alchemist a150 or a500||DG2-384EU||384||3,072||12GB GDDR6||192-bit|
|Intel Arc Alchemist a130 or a300||DG2-128EU||128||1,024||8GB or 4GB GDDR6||96-bit|
The specs above haven’t been confirmed by Intel. They’re the product of rumors, leaks, and speculation, so they’re subject to change. The names haven’t been confirmed, either. We’ve seen a leak about branding, but the card names are still subject to change.
Originally, rumors suggested Intel would launch five Arc Alchemist models, though the most recent rumors point to only three models at launch. The others may be for mobile configurations, but we don’t have any word about mobile Arc Alchemist implementations at this point.
Based on the specs, the flagship unit looks fit to compete with Nvidia’s Ampere and AMD’s RDNA 2 ranges. It comes with 512 EUs for a total of 4,096 cores, as well as up to 16GB of GDDR6 memory. Both 16GB and 8GB models have been rumored, though it’s possible the slimmer version is being reserved for laptops.
Otherwise, the cards step down similar to AMD and Nvidia. The 384 EU model looks like it will compete with the AMD RX 6700 XT with 12GB of memory, though rumors suggest that the card will perform closer to the RTX 3060. This points out one of the big problems with comparing GPU specs — oftentimes, it doesn’t say much about performance.
The bottom card could come with either 8GB or 4GB of memory, though it’s not clear right now. It’s the lowest-end card of the range, targeting the performance of a GTX 1650 Super, according to rumors. The killer for this card looks to be the 96-bit memory bus, which will choke the card regardless of the memory configuration it ships with.
Intel hasn’t revealed any performance numbers for Arc cards yet. However, the first-generation Alchemist cards look capable of running a good chunk of recent AAA video games, hopefully at playable frame rates. Following the announcement, Intel released a montage of gameplay footage captured on preproduction silicon.
— Intel Gaming (@IntelGaming) August 16, 2021
The video shows the Crysis Remastered trilogy, Forza Horizon 4, Days Gone, and Metro Exodus, among other titles. Crysis Remastered and Metro Exodus, in particular, are fairly demanding. That’s a good sign for Intel’s first foray into gaming graphics cards, but it’s important to reiterate that we don’t have any performance numbers at this time.
A benchmark leak from August showed a DG2 card (now known as Alchemist) reaching speeds of up to 2,200MHz, which is faster than most cards from AMD and Nvidia. That doesn’t mean the card is faster than its competition, just that Intel is able to drive its silicon harder.
Another leak from April showed some performance information. The leak showed the flagship model performing about as well as the Nvidia RTX 3080 and AMD RX 6800 XT. In 3DMark TimeSpy, the leaker claimed results could even rival Nvidia’s $1,500 RTX 3090.
At this point, that kind of performance seems like wishful thinking. A Geekbench leak of a mobile configuration showed Arc Alchemist underperforming compared to Intel’s own UHD 750 graphics. That’s just as unlikely as the top card performing near an RTX 3090, but it’s important to consider both ahead of launch.
The most recent rumors suggest that the flagship card will fall between an RTX 3060 Ti and RTX 3070 Ti, which seems more reasonable. The next step down should perform between an RTX 3060 and RTX 3060 Ti, while the lowest card could match the GTX 1650 Super. It’s important to wait for real testing, though. We’re dealing purely in rumors and speculation in terms of performance right now.
We have more concrete information about overclocking, though. Intel has confirmed that Arc drivers will feature a built-in overclocking utility, allowing you to push the clock speeds past their rated spec. We don’t know what the overclocking tool will look like, but we hope it matches up to the offerings from AMD and Nvidia.
As we’ll get to in the next section, each core in the Xe HPG design gets its own dedicated ray-tracing unit. At Architecture Day 2021, Intel confirmed that its upcoming Alchemist cards will support DirectX 12 and Vulkan ray tracing at launch, which compromise the vast majority of titles that support ray tracing available today.
Ray-tracing cores aren’t all built equally, however. Currently, Nvidia uses dedicated ray-tracing cores, while AMD packs a “ray accelerator” into each compute unit. The end result is vastly better ray-tracing performance on Nvidia graphics cards compared to AMD, as you can see in our review of the AMD RX 6600 XT.
Intel looks to be taking a similar approach to Nvidia, suggesting decent ray tracing performance on the upcoming range. It’s too soon to say, however. Given how new ray tracing is for gaming and the hardware it requires, Intel’s implementation may not be perfect at launch.
Alongside ray tracing, the cards will come with Intel XeSS. This is an artificial intelligence (A.I.)-assisted supersampling feature that works similarly to Nvidia’s Deep Learning Super Sampling (DLSS). Intel uses a neural network to train a model, then utilizes dedicated cores on the graphics card to perform the upscaling.
In practice, XeSS might be identical to DLSS. Intel recently announced that it hired the person who helped create DLSS at Nvidia, so the two features will likely look similar. Intel says XeSS can scale from 1080p up to 4K without a noticeable quality loss and that it offers up to a 2x frame rate improvement over native rendering. That’s exciting, but we’ll have to wait to see how XeSS stacks up when it launches.
Although similar, XeSS isn’t the same as DLSS. It seems Intel took some notes from rival AMD and its FidelityFX Super Resolution (FSR) upscaling tech. Like FSR, XeSS works across a wide range of hardware, not just Intel graphics cards. To achieve broad support and A.I. upscaling, Intel developed two software development kits (SDKs).
The first utilizes dedicated cores on Intel graphics cards, similar to how DLSS uses Tensor cores on Nvidia RTX cards. This is the full, fat XeSS experience, and Intel says developers can start implementing it in late August.
Another SDK uses DP4a instruction, which is used in A.I. applications on recent Nvidia graphics cards and recent Intel integrated graphics. Intel says this version has some quality and performance differences compared to the normal version of XeSS. However, it opens up much wider support for other hardware, which is great to see. This SDK is coming in 2021, but Intel hasn’t said when.
As for how XeSS, DLSS, and FSR stack up, we’ll just have to wait and see. There are a few key things Intel needs to do to make XeSS successful, and there’s still a lot we don’t know about the feature.
All of the currently announced Intel Arc graphics cards will use the Xe HPG architecture or a version of it. Intel announced that it has three future versions in the works: Xe2 HPG, Xe3 HPG, and Xe Next. These will show up on the second, third, and fourth generation of Intel Arc cards, respectively.
Unfortunately, we don’t know anything about these architectures right now. However, we have the code names for the upcoming generations: Battlemage, Celestial, and Druid.
The first-generation Alchemist cards are being built on chipmaker TSMC’s N6 node, which is a revision of the N7 node used on AMD RX 6000 graphics cards. Intel has already provided a full deep dive of the Xe HPG architecture, so we have a good idea about how it works ahead of launch.
The basis of Xe HPG is an Xe Core, which features 16 vector units and 16 Xe Matrix Execution (XMX) units, along with an L1 cache. Intel combines four of these Xe Cores into a render slice and adds a shared L2 cache between them, as well as dedicated ray-tracing cores for each Xe Core. These slices are what will separate the various Arc Alchemist cards.
Intel says it can add up to eight slices to a graphics card, totaling 32 Xe Cores and 512 XMX and vector units. We don’t know how future architectures will work, but Intel says that Xe HPG is a scalable architecture. In the future, we’ll likely see smaller, more efficient processes alongside more render slices on a card.