Intel shows off DG1 discrete GPU: Small and efficient. A sign of things to come?

Intel finally showcased an Intel Xe DG1 GPU in the flesh at CES 2020. We have to admit, the diminutive single-fan card sports one of the snazziest reference designs we have ever seen.

It’s important to keep expectations in check, though. The DG1 showcased was part of Intel’s Xe LP lineup, with parts that scale up to 50W in TDP. This means that this particular unit, at least, isn’t going to deliver revelatory performance. Moreover, the DG1 unit won’t be shipped to end-users. Intel is exclusively sending this to software vendors including game developers. Why is this? To make sure that, in the months to come, the industry is familiar enough with Xe’s architecture to ensure that games, machine learning tools, and other GPU-intensive software will be optimized to effectively use Xe graphics.

What do we know about the DG1’s performance? Well, the ever-useful Eurasian Economic Union portal leaked basic hardware specs a while back: DG1 will apparently feature 96 Execution Units (EUs), meaning we’re looking at about 50 percent more shader hardware than the top-end configuration for Gen 11 graphics. Intel claims a 100 percent improvement in performance, though, which would indicate IPC gains, clock speed hikes, or both.

The folks over at Linus Tech Tips got hands-on with the Destiny 2 demo from CES and confirmed that it was indeed running at 1080p 60 FPS. Destiny 2 benchmarks give us a good idea of how the integrated variant of Xe (which is related to, but not identical to DG1) performs. We’re talking about a part that’s faster than Vega 10 and in the rough vicinity of the GTX 1050. We expect the discrete variant to be marginally faster, putting it in contention with the GTX 1050 Ti.

This actually gives us a good idea of how efficient Xe is. Intel confirmed that the DG1 is a sub-75W card, as well as saying that Xe LP will scale up to 50W. This strongly indicates that the DG1 is a 50W part. If DG1 delivers GTX 1050 Ti levels of performance at 50W, that means that Xe is substantially more power-efficient than Pascal, at the very least, and possibly comparable with Turing. We wouldn’t be surprised if a hypothetical baseline Xe HP variant in the 75W range keeps with the GTX 1650. This is good news for Intel because it implies that Xe is likely competitive with Turing and Navi in the performance stakes. Just how Ampere will change the equation remains to be seen, however.


Penguin-published author, and journalist. Loves PC hardware but has terrible hand-eye coordination. Most likely to be found playing Total War or watching weird Russian sitcoms.

Related Articles

Back to top button