Intel Arc Alchemist A380 OpenCL Performance Benchmarked by SiSoftware, GPU Deemed “a Decent Effort”

With the release of Intel’s new Arc gaming processor, many are asking if it is worth buying. Though it has all-new features for PC gamers and a higher boost clock speed than its predecessor, it still ranks below AMD Ryzen 7 1800X in performance tests by SiSoftware.

The “intel arc graphics benchmark” is a test that has been created by SiSoftware. The test was performed on the Intel Arc Alchemist A380, where the GPU was deemed to be a “decent effort”.

Picture: Intel

For Intel’s forthcoming Arc “Alchemist” A380 GPU, SiSoftware has released some of the first OpenCL benchmarks, showcasing its computational prowess in a variety of tests that contrast it with some of NVIDIA and AMD’s rivals. There are still some interesting tidbits in there for potential adopters, such as what appears to be the complete specifications for the entire Arc Alchemist family, even though the verdict seems to be that the GPU isn’t all that exciting when compared to already-available options like the Radeon 6500 XT and GeForce RTX 3050. According to SiSoftware, the top-tier Alchemist series is expected to go by the codename A700 and have 512 Execution Units, 4,096 Shaders, and 16 GB of memory:

512CU / 4,096 SP maximum
- 128CU/1,024SP A300 series
- 384CU / 3,072SP A500 series
- 512CU/4,096SP for the A700 series
16GB maximum GDDR6
- 6GB GDDR6 96-bit 14Gbit/s A300 series, 192GB/s
- 12GB GDDR6 192-bit 16Gbit/s, A500 series, 384GB/s
- 16GB GDDR6 256-bit A700 series, 16Gbit/s, 512GB/s
TSMC’s 7nm process
Support for OpenCL 3.0 and DirectX 12 Ultimate
Half-floating point support for FP16
NO FP64 support for double floating-point (!)
Tensors for Deep Learning (also known as XMX or Matrix Engines)
- Support for FP16 and Int8 (but not FP32/FP64)
- It must be properly coded to utilize the OpenCL extension.

Performance of the Intel ARC (“Alchemist”) A380 GP-GPU Graphics with OpenCL (SiSoftware)

Intel-Arc-Alchemist-A380-OpenCL-Performance-Benchmarked-by-SiSoftware-GPU Photograph: SiSoftware

Since we have only examined low-end models here, nVidia need not be concerned. Despite being more costly, it is absolutely worthwhile. The most reliable performance is provided by CUDA, and even the “ancient” Turing can easily defeat the Intel ARC and AMD “Navi II” rivals. Again, both Intel and AMD’s support for OpenCL still leaves a lot to be desired, and DirectX is primarily designed for games rather than for computing. However, AMD should be your second option since Navi has been around for a lot longer and the drivers are mature. The A380 is still competitive and could become better with drivers and optimizations — assuming Intel chooses to help ISVs like us.
We don’t have FP64 support, just as with integrated graphics versions, but at this (entry) level, it doesn’t matter. However, at the top-end (Xe-HP), it would be terrible. Fortunately, nVidia has limited the FP64 on its consumer cards to a 1/64x rate, making it essentially useless. If you want FP64 native support but lack the funds (who does these days? ), AMD is still your best option.
Tensor (matrix multiplier) support is restricted to FP16/Int8 and won’t aid typical FP32 performance, but it should enhance performance (GEMM, CNN/RNN neural-networks, image processing). As nVidia did with “Ampere,” we will probably have to wait for upgraded tensors in next releases.
Perhaps expecting an nVidia/AMD killer was too much, but overall, it is a respectable attempt. YMMV (Your Mileage May Vary).