An SiSoft benchmark of Intel’s upcoming Arc “Alchemist” flagship graphics card has surfaced. Although not a very indicator of real-world gaming performance, it gives us an idea of Arc’s compute capabilities. Looking at the figures, we can see that the fastest Intel Arc GPU falls behind the NVIDIA GeForce RTX 3070 Ti in single precision compute (FP32) but managed to be nearly twice as fast in double precision (FP64) workloads.
| Arc “Alch” Flagship (512EU) | NVIDIA RTX 3070 Ti | |
|---|---|---|
| Single Float (FP32) | 20,888 | 27,029 |
| Half Float (FP16) | 35,093 | 36,510 |
| Double Float (FP64) | 1,000 | 594 |
| Quad Float (FP128) | 109 | 22 |
It’s worth noting that most games solely rely on single-precision compute for the majority of their code, with a few newer games leveraging half-precision or FP16 for certain specialized workloads such as ray-tracing and upscaling. Double precision compute is largely redundant in gaming, but is important in the compute-intensive HPC space. The same can be said of FP16 or half-precision/mixed-precision compute which is used to accelerate AI and neural networks, plus upscaling techniques such as XeSS.
This goes on to show that Intel is focusing on the HPC and the AI segments first, and it won’t be surprising to see an entire lineup follow Ponte Vecchio for data centers. XeSS, being open-source and cross-platform is Raja Koduri’s secret weapon and from the looks of it, is going to replace NVIDIA’s DLSS soon enough.
Intel’s 1st Gen Arc graphics cards are going to scale from 128 to 512 EU parts, with a total shader (ALU) count of up to 4,096 on the top-end SKUs. TSMC’s 6nm node should allow for boost clocks of up to 2GHz paired with 16GB of GDDR6 memory and a moderate amount of L2 cache (16-32MB?). With Arc, Intel is going to overhaul its graphics architecture, adding support for ray-tracing, tensors, and mixed-precision compute.
