INTEL ARC PREPS FOR A FLOOD
The big two GPU names are about to get some muchneeded competition from Intel’s Arc, or at least in theory. Arc, formerly DG2, aka Xe-HPG, has seen quite a few delays. We originally thought it would launch by the end of 2021, but now it’s looking like a mid-2022 launch for the desktop parts. It could be too little, too late for gamers, though the media aspects of the chips remain promising.
Intel recently revealed the initial lineup for laptop Arc A-series models, which will inevitably be paired with Intel CPUs. Given Intel dominates the laptop landscape in sales, it’s probably a smart business move, but it doesn’t give us a lot of hope for high performance.
Intel compared its lowertier A370M part against its own integrated Iris Xe Graphics, from a Core i71280P. That’s a 28W mobile chip with 96 Execution Units (now called XVE, for “Xe Vector Engine”), going up against a 35–50W A370M with 96 of the new XVEs. Performance was over 50 percent faster, but our previous testing suggests that will be quite a bit slower than even a GTX 1050.
The highest tier Arc chip, A770M, will come with 512 XVEs and faster memory. That’s 4096 shader cores and a potential compute performance of 13.5
TFLOPS FP32. Theoretical performance isn’t the same as real-world performance, but even in a best-case scenario, that puts the A770M on par with Nvidia’s mobile RTX 3070.
Intel isn’t just talking about graphics performance, however. Arc also includes XMX units (Xe Matrix eXtensions), which are basically Intel’s equivalent of Nvidia’s Tensor cores. Each XMX can perform 128 FP16 operations per clock, so the A770M has a peak throughput of 108 TFLOPS for deep learning and AI acceleration. Intel will be using the XMX engines for its XeSS (Xe
Super Sampling) upscaling algorithm, which sounds a lot like DLSS. XeSS will also be able to run on DP4a (INT4) hardware, for non-Arc GPUs including AMD RDNA and newer and Nvidia Pascal and newer architectures.
Finally, Intel will include up to 32 ray tracing units (RTU) in its top Arc GPU. Unless Intel’s RTUs are far more potent than those of Nvidia or even AMD, we don’t expect much from Team Blue when it comes to its first generation of ray tracing hardware. The lowest tier Arc A350M for example has just six RTUs clocked at 1150MHz, with 4GB VRAM on a 64-bit interface. Based on those specs, it might make even the RX 6500 XT look fast.
We’ll find out just how the Arc chips stack up in the real world in the coming months, but we have a sinking feeling.