Add a LARGE GRAIN OF SALT: The GPU market is hotter than ever, fueled by gaming, AI, and knowledge heart demand. However actual competitors to Nvidia stays scarce. That is the place Bolt Graphics is available in, pitching its new Zeus structure as a disruptor. The corporate claims it may well beat Nvidia’s flagship playing cards in uncooked efficiency, however a more in-depth look suggests Zeus could be extra smoke than silicon, presumably repackaging server-class {hardware} in gamer-friendly clothes.
California-based startup Bolt Graphics lately revealed detailed specs for its upcoming graphics playing cards, claiming they outperform the Nvidia RTX 5090 in path tracing workloads. Whereas the Zeus lineup seems spectacular on paper, a number of particulars increase pink flags.
The corporate first unveiled its bold plans in March, boasting huge reminiscence swimming pools and graphics processors which might be, in accordance with them, orders of magnitude quicker than Nvidia’s.
A newly launched spec sheet takes these claims even additional – however Bolt’s vertical TFLOP numbers seem suspiciously low, and the corporate choosing older VRAM codecs solid doubt on the gaming credentials of those GPUs.
The specs for the corporate’s 4 deliberate GPUs: the Zeus 1c-032, 2c-064, 2c-128, and 4c-256, seem to outperform the RTX 5080, 4090, and even 5090 in a number of areas, all whereas consuming lower than half the facility.
As an illustration, the weakest entry-level “1c-032” is claimed to attain 77 gigarays in path tracing in comparison with the RTX 5090’s 32. It additionally contains 128 MB of cache, and Bolt is making an attempt to revive socketable VRAM, enabling configurations with as much as 160 GB of reminiscence.
All of that is achieved with an influence draw of simply 120W, decrease than even the RTX 5050. In a refined jab at Nvidia, Bolt emphasised that Zeus makes use of normal 8-pin energy connectors, which it notes are “identified to not soften.”
The upper-end Zeus fashions scale up the 1c’s specs considerably, growing cache, VRAM, gigarays, and 8K video stream help to ranges that appear virtually too good to be true.
The flagship 4c GPU, which makes use of barely much less energy than the RTX 5090, delivers 307 gigarays, options 512 MB of cache, and helps an astounding 2.3 TB of VRAM.
Carry your concepts to life with Zeus, a brand new sort of GPU:
– Expandable reminiscence: 32/64/128 GB soldered + 2x/4x SO-DIMM slots for as much as 384 GB reminiscence!
– Large enhancements to path tracing efficiency!
– 400 GbE QSFP-DD port!
– Constructed-in high-performance RISC-V CPU cores able to… pic.twitter.com/ITJyFInLGV– Bolt Graphics (@BoltGraphicsInc) August 3, 2025
Nevertheless, all of those numbers stay theoretical. Bolt’s “pre-silicon” efficiency estimates are primarily based on artificial benchmarks, not real-world gaming assessments. Moreover, manufacturing is not anticipated to start till late 2026, that means the GPUs will not be accessible earlier than 2027. By then, each AMD and Nvidia will probably have launched next-generation {hardware}.
In the event you’re questioning why any of this exists, the reply could be so simple as hype. Bolt could possibly be attempting to show heads and switch that focus into funding, by driving the present wave of GPU gold rush fever, particularly as AI accelerators proceed to be Silicon Valley’s hottest {hardware} obsession.
The spec sheet additionally reveals areas the place the Zeus playing cards considerably path Nvidia’s high-end RTX 5000 collection. These huge VRAM capacities depend on LPDDR5X and DDR5, that are probably a lot slower than AMD and Nvidia’s GDDR6 and GDDR7 reminiscence.
Moreover, FP32 and FP16 TFLOP counts – essential for gaming efficiency – fall nicely under the RTX 5090, and in lots of circumstances, even under the RTX 5080.
If Bolt’s numbers show correct, the Zeus collection could possibly be compelling for sure HPC, rendering, or server workloads. However with regards to gaming (which additionally wants optimized drivers, and that takes numerous time – simply ask Intel), their real-world efficiency will probably stay unsure for at the very least the following couple of years.
