Intel might have revealed specs for top-of-the-line Alchemist A780 desktop GPU in its ARC Control video
Intel is doing a great job at teasing us with bits and pieces of info regarding the upcoming desktop ARC Alchemist GPUs. Up until now, only the specs for some mobile dGPUs have been officially unveiled, yet the desktop lineup is still shrouded in mystery. From previously leaked TGP estimations, we know that Intel wants its fastest desktop GPUs to match AMD’s RX 6700 XT and Nvidia’s RTX 3070 models at a slightly lower price-point, and, based on the spec info spotted by 3DCenter in Intel’s recent ARC Control software presentation video, it looks like Team Blue could really offer a competitive mid-range solution.
3DCenter is suggesting that the ARC Control video is very briefly revealing a series of specs for the ARC A780 top-of-the-line desktop GPU. The Live Performance Monitoring tab shows GPU power, GPU clock and VRAM clock specs that do not seem to match any mobile-grade SKU announced by Intel. GPU Utilization is over 99% so the 2250 MHz GPU clock, as well as the 1093 MHz VRAM clock with the GPU power at 175 W should represent the boost specs of the ARC A780.
One clue that points towards this being a desktop-grade dGPU is the GPU power consumption rated at 172-175 W. Intel’s top-of-the-line A770M mobile SKU is limited to 150 W, but could theoretically access a bit more power from unused CPU resources through the Deep Link feature. Probably just a few more W, though. Also, we should keep in mind that GPU Power only refers to what the GPU chip itself can utilize, as the VRAM and other adjacent circuitry could add at least 30 more W leading to 200+ W total board power. These wattage figures seem lower compared to the competition because Intel is supposedly using the 6 nm nodes from TSMC, so a slight of advantage over AMD and Nvidia there as well. All things considered, the power figures seem to be in-line with the 175 W - 225 W specs from previous leaks.
If the 1093 MHz boost VRAM clocks are correct for the ARC A780, we are looking at 17.5 Gbps GDDR6 memory, which is a fair bit faster than what AMD and Nvidia offer for their mid-range models. Moreover, the rumored 4096 FP32 cores on the A780 SKU should allow for the best-in-class INT32, FP16 and INT8 peak performance via the XMX architecture, easily beating AMD’s and Nvidia’s similar solutions.