This monstrous video card probably has more memory that you have space on your SSD; Bolt Graphics Zeus has 2.25TB VRAM

Bolt Graphics unveils Zeus GPU, targeting rendering, HPC, and gaming
Claims include 10x faster rendering, 6x FP64, 300x simulation gains
Zeus supports up to 2.25TB memory, 6x 800GbE, 8x PCIe Gen5

Bolt Graphics has announced what it says is the fastest graphics processor ever, which is quite the claim.

Based in Sunnyvale, California, the semiconductor startup says Zeus has a “completely new GPU design for high-performance workloads including rendering, HPC, and gaming” and that it “addresses performance, efficiency, and functionality limitations with legacy GPUs.”

Bolt Graphics adds Zeus offers up to 10 times faster rendering, 6 times higher FP64 HPC performance, and up to 300 times faster electromagnetic wave simulation speeds compared to legacy GPUs.

Stunning path tracing performance

It supports expanded memory through Bolt’s architecture, allowing up to 384GB per PCIe card and up to 2.25TB per Zeus unit in a 2U server, with rack configurations reaching 180TB. Zeus also integrates 400 GbE and 800 GbE Ethernet interfaces directly into the GPU, reduces energy use despite the higher performance, and will be available in PCIe, server, and cloud form factors.

“Zeus increases performance while simultaneously reducing power consumption,” said Darwesh Singh, Founder and CEO of Bolt Graphics. “I’m proud of the Bolt team’s dedicated effort to create a solution that addresses key customer pain points, enabling them to be more productive and bring their ideas to life.”

Slides published by ServeTheHome show Zeus is based on a RISC-V RVA23 out-of-order scalar core and there are a number of different architectures.

The Zeus 1c26-032 is a single-chiplet design and supports DisplayPort 2.1a and HDMI 2.1b, along with an RJ45 BMC port – typically seen on DPUs, servers, and other infrastructure components. There’s also a QSFP-DD port for 400GbE and two PCIe Gen5 x16 slots. The 1c26-032 includes 32GB of LPDDR5X and up to 128GB of DDR5.

The Zeus 2c26-064/128 is a two-chiplet module, while the 4c26-256 is a four-chiplet configuration. In the case of the latter, which has 500W of chip power, each chiplet is paired with 64GB of LPDDR5X and up to 512GB of DDR5, bringing the total memory capacity to 256GB of LPDDR5X and up to 2TB of DDR5. Combined, the GPU supports up to 2.25TB of total memory.

It also includes six 800GbE (OSFP) ports, which ServeTheHome notes is “a lot of connectivity coming out of a GPU. Putting that into perspective, that is ~12x PCIe Gen5 x16 lanes worth of bandwidth.”

The slides show the Zeus 2c26-064/128 outperforms the Nvidia GeForce RTX 4090 and RTX 5090 in path tracing workloads by roughly 9x and 4.8x, delivering 154 Gigarays compared to 17 Gigarays (RTX 4090) and 32 Gigarays (RTX 5090). In FP64 compute performance, Zeus achieves 10 TFLOPS, significantly ahead of 1.4 TFLOPS (RTX 4090) and 1.6 TFLOPS (RTX 5090) – roughly 7.1x and 6.25x higher, respectively.

Comparing the Bolt Zeus 4c26-256 against Nvidia’s consumer GPUs, path tracing performance is approximately 18x higher than the RTX 4090 and nearly 10x higher than the RTX 5090, while FP64 compute throughput is over 12x greater.

All of this should be taken with a pinch of salt of course. As ServeTheHome observes, “This feels like an announcement where we really want to see product. The fact that we are still a few months away from early developer kits makes this feel a bit early. On the other hand, if Bolt Graphics has Zeus competing with a combination of NIC and consumer GPU, while offering more memory, then it could be a really neat combination. I am always a bit skeptical of products until we see them live.”