Intel’s Arc Pro-A series initially includes three graphics accelerators and serves entry-level workstations with a maximum of 75 watts, 4.8 TFLOPs and 6 GB of memory.
Intel announces three graphics accelerators based on the Arc Alchemist architecture: The two PCIe 4.0 plug-in cards Intel Arc Pro A40 and A50 are flanked by the notebook chip A30M. All three are scheduled to launch in preconfigured workstations later this year. When asked, Intel is currently not considering a retail sale, nor are concrete individual prices given.
But they shouldn’t be too high, because all three models are aimed at compact entry-level workstations. The plug-in cards have half the height and are apparently quite short. Its four mini DisplayPorts serve a corresponding number of screens with up to 3840 × 2160 pixels and 60 Hertz (Hz). Alternatively, two screens with 5120 × 2160 pixels at 120 Hz or one display with 5120 × 1440 pixels and 240 Hz are also possible. The maximum is two 8K displays (7680 × 4320 pixels) at 60 Hertz. Intel’s Arc Pro A models also support Dolby Vision HDR playback.
A current unique selling point is the integrated AV1 encoder, which (almost) without the CPU having to do anything can convert videos using a bit-rate-saving and, above all, license-free AV1 codec – good for twitch streamers and streaming services.
ARC GPU with 50 to 75 watts
With a power consumption of 50 watts for the Intel Arc Pro A40 and 75 watts for the A50, the PCIe 4.0 x8 cards do not require a PCIe power connector. The A30M is said to have a configurable TDP of 30 to 50 watts.
A40 and A50 have 6 GB GDDR6 graphics memory, which achieves a transmission speed of 192 GB/s with 96 data lines; the A30M only has 4 GB (128 GB/s).
TSMC manufactures the graphics chips of the cards with 6-nanometer technology like AMD’s Navi23 on the Radeon RX 6500 XT. In addition to the Xe cores, they also have eight ray tracing units and Intel’s AI accelerator called XMX.
The A40 and A50 chips have 8 Xe cores with 128 Xe vector engines and, according to the product page, a base clock of 2 GHz each, and the A30M should also be able to do that with 50 watts.
However, the TFLOPS information on Intel’s overview page is not enough, since each of the 128 Xe vector units creates 16 FP32-FMA calculations per clock, i.e. a total of 2048 FLOPS/clock, since the multiplication and addition in FMA are each performed as individual floating-point count surgery. At 2 GHz that would already be 4.1 TFLOPS, the A50M with a nominal 4.8 TFLOPS would then have to clock at 2.35 GHz in Turbo, which the Arc A380 gaming variant easily achieves. The 3.5 TFLOPS of the A40 and A30M then indicate a clock speed of around 1.7 GHz.
In addition to DirectX 12 Ultimate, Vulkan and OpenCL, the card drivers also support Intel’s OneAPI interface, for example for use in Blender. However, they are to be made fit for other programs in the fields of architecture, construction and design.
