Following plans first unveiled last year during the launch of their DG1 GPU, Intel sends word this morning that the first Iris Xe video cards have finally begun shipping to OEMs. Based on the DG1 discrete GPU that’s already being used in Intel’s Iris Xe MAX laptop accelerators, the Iris Xe family of video cards are their desktop counterpart, implementing the GPU on a traditional video card. Overall, with specifications almost identical to Xe MAX, Intel is similarly positioning these cards for the entry-level market, where they are being released as an OEM-only part.

As a quick refresher, the DG1 GPU is based on the same Xe-LP graphics architecture as Tiger Lake’s integrated GPU. In fact, in broad terms the DG1 can be thought of as a nearly 1-to-1 discrete version of that iGPU, containing the same 96 EUs and 128-bit LPDDR4X memory interface as Tiger Lake itself. Consequently, while DG1 is a big first step for Intel – marking the launch of their first discrete GPU of the modern era – the company is planning very modestly for this generation of parts.

Intel Desktop GPU Specification Comparison
  Iris Xe
Tiger Lake
Ice Lake
Kaby Lake
ALUs 640
(80 EUs)
(96 EUs)
(64 EUs)
(24 EUs)
Texture Units 40 48 32 12
ROPs 24 24 16 8
Peak Clock 1500MHz 1350MHz 1100MHz 1150MHz
Throughput (FP32) 2.11 TFLOPs 2.1 TFLOPs 1.13 TFLOPs 0.44 TFLOPs
Geometry Rate
2 2 1 1
Memory Clock LPDDR4X-4266 LPDDR4X-4266 LPDDR4X-3733 DDR4-2133
Memory Bus Width 128-bit 128-bit
VRAM 4GB Shared Shared Shared
TDP 30W Shared Shared Shared
Manufacturing Process Intel 10nm SuperFin Intel 10nm SuperFin Intel 10nm Intel 14nm+
Architecture Xe-LP Xe-LP Gen11 Gen9.5
GPU DG1 Tiger Lake
Ice Lake Integrated Kaby Lake Integrated
Launch Date 01/2021 09/2020 09/2019 01//2017

The first DG1 GPUs were shipped in the fall as part of Intel’s Iris Xe MAX graphics solution for laptops. At the time, Intel also indicated that a desktop card for OEMs would also be coming in 2021, and now, right on schedule, those desktop cards have begun shipping out.

Overall, Intel is taking a very OEM-centric approach to their DG1 products, and that goes for both laptops and the desktops. Even the desktop Iris Xe cards won’t be sold as retail – as entry-level cards, they are unlikely to fly off of shelves – and instead are only being sold to OEMs for use in pre-built systems. And even then, the cards were co-designed with ecosystem partners – of particular note, ASUS – rather than Intel building and shipping out their own video cards. So by desktop video card standards, Intel is being somewhat hands-off at the moment.

In a curious twist, the desktop cards will have slightly lower specifications than the laptop parts. While I’m still waiting to hear what the TDPs and final clockspeeds will be, Intel’s announcement confirms that the Iris Xe cards will only ship with 80 of 96 EUs enabled, rather than being fully-enabled in the case of the laptop parts. Given that this is an entry-level part, any further drop in performance isn’t doing the part any favors, but at the same time it was never going to be a speed-demon to begin with. At any rate, given that no chip has perfect yields, we now know where salvaged DG1 chips are going.

Meanwhile, like their laptop counterparts, the Iris Xe desktop cards will ship with 4GB of LPDDR4X memory. Intel has also confirmed that the cards will ship with up to three display outputs, with ASUS's card using a mix of HDMI, DisplayPort, and even a DL-DVI-D port.

Another DG1 Card

As for Intel’s target market, the company is targeting what they’re calling the “high-volume, value-desktop market.” Notably, unlike the Iris Xe MAX launch, Intel’s (admittedly brief) news release doesn’t spend much time focusing on the cards as a secondary accelerator, and instead are promoting these as a superior solution over existing graphics options. Given the focus on things like AV1 decoding, HDR support, and deep learning inference performance, I’m assuming that these will primarily be showing up in Atom (Gemini Lake Refresh) systems. Though it may also show up in low-end Comet Lake Celeron and Pentium systems, where vendors are looking to add a few more display ports and take advantage of the additional hardware accelerator blocks for things like video encoding, similar to how Intel positioned Iris Xe MAX for laptops.

Finally, given the OEM-centric nature of today’s launch, Intel isn’t publishing any specific availability dates for their Iris Xe video cards. But we expect that they’ll begin showing up in short order.

Source: Intel

Comments Locked


View All Comments

  • cyrusfox - Wednesday, January 27, 2021 - link

    I still plan on buying on ebay and testing it unless I hear this more concretely. I expect there to be drivers out there to enable it jsut like they did for early DG1 samples they sent out for driver development.
  • lmcd - Thursday, January 28, 2021 - link

    Not silly at all. This is barely a working discrete part at all. Notice the LPDDR4X on a desktop card? That 4GB probably costs about as much as the same amount of GDDR6, if not more, but the memory controller is copy-pasted from TGL. The PCIe 4 lane limitation also sticks out. There's a simple answer here: it's not a PCIe dGPU. It's an onboard GPU decoupled from onboard fabric, still being initialized like it's an onboard GPU. It probably doesn't have a real vbios and needs to be initialized by the chipset.

    It might not even work in every Intel machine, honestly.
  • PingSpike - Thursday, January 28, 2021 - link

    The videocardz article suggests it only works on coffee lake S and comet lake S, only a selection of the mid range chipsets and only ones with a special bios. So it certainly looks like it doesn't work on many Intel machines.
  • michael2k - Tuesday, January 26, 2021 - link

    But the performance delta of the DG1 over the 1030 is essentially zero.

    And talking about SFF, you can use MSI Afterburner and GPU Boost on an GeForce to target a lower power threshold than the 75W TDP:

    The tools are normally used for overclocking, but apply just as much to underclocking and undervolting. The article undervolts a 1070 -> 1060, shaving 54W TDP from a 1070.

    I don't know if you can shave 35W from a 1650, but essentially if you limit the clock the GPU ever tries to hit, it won't ever reach the 75W TDP.
  • Spunjji - Wednesday, February 3, 2021 - link

    Given what Nvidia managed in transforming the 1650 into the MX450, I suspect you could happily chop 35W off a 1650 and get much better performance (and perf/watt) than the 1030. It would be a LOT slower than a stock 1650, though.
  • jabber - Wednesday, January 27, 2021 - link

    Buy a Quadro/FirePro on Ebay for very little.
  • Rudde - Saturday, January 30, 2021 - link

    I was looking at upgrades for a nVidia 710 that was used for streaming. I noticed that the 1030/1010 doesn't support hardware accelerated video encoding. The 710 does. For that niche, the Intel DG1 or an AMD card might be the best. Unfortunately, the DG1 is OEM only.
  • ceisserer - Tuesday, January 26, 2021 - link

    Intriguing - Intel is willing to spend precious 10nm capacity on such low-margin parts.
    Or could it be 10nm isn't that tight anymore with AMD eating Intels market share slide-by-slice.
  • Deicidium369 - Tuesday, January 26, 2021 - link

    There are 3 DIFFERENT 10nm lines

    Vanilla 10nm - Ice Lake & Ice Lake SP
    10nm SF - Tiger Lake U & H and this tiny GPU
    10nm ESF - Golden Cove (Alder Lake, Sapphire Rapids and un named U SOC)

    so other than Tiger Lake and maybe Xe HP that line has tons of capacity it seems

    AMD is not eating anything and is hardly shipping anything for the PC market - TSMCs long fragile supply chain broken... Q1 is going to be a bloodbath for AMD...
  • Spunjji - Wednesday, February 3, 2021 - link

    "Q1 is going to be a bloodbath for AMD"
    Deicidium predictions are like Qanon drops: They never come true, and it never stops the true believers from buying in anyway.

Log in

Don't have an account? Sign up now