Intel’s upcoming high-performance DG2 desktop graphics card will pack up to 512 execution and competition units AMD’s latest RDNA 2 GPU For powerful raw graphics processing. This is the interesting inclusion, if not fully confirmed, of the device ID entries included in the most recent Intel graphics driver dump.
Previous rumors about Intel’s new GPU pointed to a more modest number from 128 to 384 in the European Union. But I also spotted it Videocardz.comIntel’s latest GPU 100.9126 driver has Hardware ID entries referring to two variants of the new DG2 card, high performance takes Latest Intel Xe graphics architecture.
The two entries read “DG2128 SKU” and “DG2 512 SKU” respectively, which unambiguously refers to the 128 EU and 512 EU variants of the new GPU. If the 512 EU variant is accurate, it should deliver raw graphics power of roughly the same pitch as AMD’s latest Radeon RX 6000 GPUs, if far behind Nvidia’s Ampere chips.
Intel’s Xe architecture is already a known amount in the form of DG1, a discrete GPU designed for thin and light laptops. It’s also the integrated graphics core of Intel’s Tiger Lake CPUs.
The DG1 offers 96 execution units running at 1,650MHz and matching 25W TDP, all of which are enabled by an Intel 10nm production node. DG1 is claimed to be good for about 2.5 TFLOPs of raw FP32 computation strength. This provides a clear reference point for the 512 EU variant of the DG2.
Intel has already said that the DG2 will be produced on a third-party production node, which is believed to be the TSMC 7nm or 6nm node. Because the DG1 is aimed at laptops that are constrained by power and heat, it is reasonable to assume significantly higher hours for the DG2 desktop.
Factor in 512 EU and higher clocks and it looks like the DG2 should offer raw graphics processing power to rival AMD’s latest Navi chips. Assume clocks are 30 percent higher than DG1 and you’ll have a DG2 board with 17TFLOPs. Upgrade 50 percent higher and DG2 would be good for about 20TFLOPs. This just happens to be AMD’s claims for the Radeon RX 6900 XT.
Of course, the strength of raw shading is not a clear guide to actual gaming performance. After all, Nvidia says the GeForce RTX 3090 is capable of more than 35TFLOPs of FP32 throughput, which means a totally unrealistic performance advantage over the best AMD GPU. There is much more to the actual in-game frame rates than the pure processing power as measured by TFLOPs.
However, the fact that Intel’s DG2 almost feels the right pitch for raw shading power is definitely promising. This latest information also matches previous rumors that DG2 could offer it Performance is roughly in line with Nvidia’s RTX 3070 panels. With the provision of both new AMD and Nvidia GPUs very narrow and Pricing is increasingly out of control, It can only be a third market participant a good thing.
In short, Intel Xe graphics doesn’t need to be the fastest thing ever to have a very positive impact on the overall market. More options and lower prices would be very welcome indeed.