- Bolt Graphics unveils Zeus GPU, targeting the rendering, the HPC and the game
- Complaints include a faster 10x rendering, 6x FP64, 300X simulation gains
- Zeus supports up to 2.25 TB of memory, 6x 800gbe, 8x PCIe Gen5
Bolt Graphics has announced what he says is the fastest graphic processor of all time, which is completely claiming.
Based in Sunnyvale, California, the semiconductor startup claims that Zeus has a “completely new GPU design for high performance workloads, including rendering, HPC and games” and that it “deals with performance, efficiency and functionality limitations with inherited GPUs”.
Bolt Graphics adds that Zeus offers up to 10 times a faster rendering, 6 times more FP64 HPC performance and up to 300 times the faster electromagnetic wave simulation speeds compared to inherited GPUs.
Breathtaking track tracing performance
It supports the expanded memory via the architecture of Bolt, allowing up to 384 GB per PCIe card and up to 2.25 TB per unit Zeus in a 2U server, with rack configurations reaching 180 TB. Zeus also incorporates 400 GBE and 800 GBE Ethernet interfaces directly into the GPU, reduces energy consumption despite higher performance and will be available in the PCIE formal factors, and Cloud.
“Zeus increases performance while simultaneously reducing energy consumption,” said Darwesh Singh, founder and CEO of Bolt Graphics. “I am proud of the dedicated effort of the Bolt team to create a solution that addresses the pain points of key customers, allowing them to be more productive and to give life to their ideas.”
Slides published by Serve Show Zeus is based on a scalar kernel outside RISC-V RVA23 command and there are a certain number of different architectures.
The ZEUS 1C26-032 is a unique chiplet design and supports DisplayPort 2.1A and HDMI 2.1b, as well as a BMC RJ45 port – generally visible on DPUs, servers and other infrastructure components. There is also a QSFP-DD port for 400 GBE and two PCIe Gen5 X16 locations. The 1C26-032 includes 32 GB of LPDDR5X and up to 128 GB of DDR5.
The ZEUS 2C26-064 / 128 is a two chips module, while 4C26-256 is a configuration with four chiplets. In the case of the latter, which has 500 W of chip power, each chiplet is associated with 64 GB of LPDDR5X and up to 512 GB of DDR5, bringing the total memory capacity to 256 GB of LPDDR5X and up to 2 TB of DDR5. Combined, the GPU supports up to 2.25 TB of total memory.
It also includes six 800gbe (OSFP) ports, which Serve Notes is “a lot of connectivity comes out of a GPU. By putting this in perspective, it is ~ 12x PCIe Gen5 x16 bandwidth ways. »»
The slides show that the Zeus 2C26-064 / 128 surpasses the NVIDIA GeForce RTX 4090 and RTX 5090 in the tracing of the path of approximately 9x and 4.8x, providing 154 gigarays compared to 17 gigarays (RTX 4090) and 32 Gigarays (RTX 5090). In FP64 calculation performances, Zeus reaches 10 TFOPS, significantly in advance on 1.4 TFOL (RTX 4090) and 1.6 TFOP (RTX 5090) – approximately 7.1x and 6.25x higher, respectively.
By comparing the Bolt Zeus 4C26-256 to the NVIDIA consumption GPU, the track tracing performance is approximately 18 times higher than the RTX 4090 and almost 10 times higher than the RTX 5090, while the FP64 calculation flow is more than 12 times more.
All of this should be taken with a pinch of salt of course. As Serve Observe: “It looks like an ad where we really want to see the product. The fact that we are still a few months old from the first developer kits made it feel a little early. On the other hand, if Bolt Graphics A Zeus competition with a combination of GPU Nic and the general public, while offering more memory, then it could be a really neat combination. I am always a little skeptical about the products until we saw them live. »»
