Introduction

We have with us the Sparkle Arc B580 Titan OC graphics card, powered by the new B580 Battlemage GPU that's been making waves since its reveal earlier this month. Sparkle is a fairly new entrant to the Intel Arc graphics card ecosystem, but has been around for decades, notably as a former NVIDIA GeForce board partner. Now they are owned by TUL—the parent company of PowerColor. The Titan OC is the company's premium custom-design based on the B580, and pairs the GPU with a 31.5 cm-long triple-fan cooling solution that's a little over two slots thick. The Arc B580 marks Intel's second generation of the Xe gaming graphics architecture as discrete GPUs. These are modern, fulfill the DirectX 12 Ultimate API standards, and include a wide range of gaming experience improvements within the XeSS 2 feature suite.



Xe2 Battlemage succeeds the original Xe Alchemist. It debuted earlier this year as the architecture driving the iGPU of Intel's Core Ultra 200V Lunar Lake mobile processors, although that variant was a "lite" implementation of Xe2. On the Arc B580, you can expect a more complete set of the Xe2 Battlemage IP, with the GPU being geared for performance-segment gaming. The B580 logically succeeds the Arc A580, but presents a massive generational performance step-up. Intel claims (and we're confirmed in our review of the reference design), that the B580 outperforms every GPU from the Alchemist generation, including the top A770, and since the A770 was marketed as a 1440p-class GPU, this use case carries on for the B580.

In all, Intel claims a 70% gain in the SIMD performance of its 2nd Gen Xe cores over Alchemist, and a 50% gain in performance-per-watt, due in part to the new TSMC 5 nm EUV foundry node these chips are being built on. This is thanks to the significant IPC gain of the 2nd Gen Xe core, a new Ray Tracing Unit with anywhere between 50% to 100% generational gains in performance, a large 18 MB on-die last-level cache, and a faster memory sub-system than the A580, besides double the memory size, at 12 GB.

The B580 is being pitted by Intel against the NVIDIA GeForce RTX 4060. If you recall, the previous generation flagship parts, the A770 and A750, were compared by Intel to the RTX 3060, and the A580 was awkwardly positioned against the RTX 3050. In this sense, Intel has made a bigger generational leap in performance than NVIDIA, and we can only hope that Intel scales out the Xe2 architecture for even larger GPUs.

The Arc B580 is based on the 5 nm BMG-G21 silicon, and features 20 Xe cores, or 128 execution units, worth 2,560 unified shaders. There are also 20 Ray Tracing Units, and 160 XMX units, which accelerate AI. This is backed by a solid raster graphics backend, consisting of 80 ROPs, and 160 TMUs. The chip gets 12 GB of 19 Gbps GDDR6 memory across a 192-bit wide memory bus, which is both 50% larger and faster than the memory implementation of the RTX 4060 and the AMD Radeon RX 7600.

Perhaps the biggest aspect of the Arc B580 is its starting price of $250, which undercuts the RTX 4060 and the RX 7600 XT by at least $50. The Sparkle Arc B580 Titan OC looks like it's from a segment above, and yet, is being offered at just a $20 premium over this. The card is 31.5 cm in length and comes with a factory overclocked speed of 2740 MHz, compared to the 2670 MHz reference speed. It also comes with a slight increase in power limits to 200 W, from 190 W reference.