High-Performance Computing Just Got Cheaper
Parallel Processing Price Plummet
Using GPUs targeted towards the gaming market for other applications isn't new. When cryptocurrency mining gained popularity, ASICs were ditched in favour of GPUs and the demand sky-rocketed, with prices only recently back to normal.
Application-Specific Integrated Circuits (ASICs) offer optimal performance due to their limited design scope, however are inflexible and cost-relative to their application.
How are these workloads similar? Fast parallel computation.
How do the two solutions differ? Price.
For high-performance computing projects, such as Finance AI, Computational Fluid Dynamics, and ML/DL Modelling, to name a few, GPUs provide a stellar solution in the early stages.
Hot Blue Silicon
Developed especially for Exascale Computing, the Intel Xe architecture is what enables the Intel Arc series GPUs.
First revealed at Supercomputing 2019 with Ponte Vecchio, a general-purpose GPU, Intel Xe will power the US Deparment of Energy's 2 exaFLOP supercomputer, Aurora.
Marketed for workstations, and where low power consumption is concerned, the Intel Arc Pro A Series offers two options.
The Arc Pro A40 is capable of 3.5 TFLOPs single-precision, consuming 50 W, and the Arc Pro A50 offers 4.8 TFLOPs single-precision, consuming 75 W; both with 6 GB VRAM.
TDP is where the advantages of the Pro Series end however, as the specifications are a fraction of the desktop-designed Arc A770.
Choose either 8 GB or 16 GB VRAM, and the Arc A770 computes at a whopping 17+ TFLOPs, single-precision. Against the NVIDIA RTX 3060, expect a 42% increase in performance to price.
Intel suggest that the Arc A770 may even compete with the NVIDIA RTX 3060 Ti.
The Intel Arc A770 8 GB will debut at $329 on the 12th October.
Will it be in your server?
Explore the Intel Arc A770Explore the Dell EMC PowerEdge R750XA
