Friday, January 9th 2026

New GPU Startup Bolt Graphics Promises 2.5x Path Tracing Performance of RTX 5090

Bolt Graphics is a new computer graphics hardware startup based out of Sunnyvale, California. This company wants to shake things up in the consumer graphics hardware market with a brand-new, grounds-up graphics architecture for graphics rendering, gaming, and HPC. What makes this startup unique and refreshing is that its architecture, or at least the marketing material of it, doesn't use the most uttered two-letter acronym in human history.

Called ZEUS by Bolt Graphics, the architecture offers a contemporary raster 3D graphics, ray tracing, and path tracing hardware pipeline, using an in-house designed SIMD architecture that's controlled by a RISC-V command processor that doubles up as a CPU. The prototype graphics card features a PCI-Express 5.0 x16 host interface, and LPDDR5X as its graphics memory, with additional DDR5 SODIMM slots for the RISC-V command processor. Together, the board can have up to 384 GB of memory to deal with large data sets. The card also features a 400 Gbps or 800 Gbps enterprise network interface, and a fully-fledged BMC/IPMI circuit on board, so you can manage large clusters of the card in a rackmount render server or workstation. The most surprising aspect of the card is power, and it only features an 8-pin PCIe power input, for a total of 225 W of power draw.
Besides hardware, Bolt Graphics is offering a fully-fledged software that that includes Vulkan and DirectX 12 interfaces, including support for Unreal Engine and Unity. Applications for the architecture support all modern programming environments. Bolt innovated a path tracing pipeline.

Bolt Graphics has made some astoundingly tall performance claims that indicate an ASIC-like approach towards popular use-cases, using fixed-function hardware. In FP64 EM simulation benchmarks, the Zeus 4C is claimed to offer 300 times the performance of an NVIDIA B200 "Blackwell." In another FP64 math benchmark, the Zeus 1C (single silicon) is shown offering 3 times the performance of a GeForce RTX 5090. Perhaps the most fascinating claim is path tracing performance, and here, a Zeus 1C is claimed to offer 2.5 times the path tracing performance of an RTX 5090, 5 times with 2C, and 10 times with 4C.

The part about Bolt Graphics marketing that caught our eye is their claim that the Zeus 1C is "the fastest graphics processor ever," made possible due to a fundamental reimagining of conventional graphics processing workflows.
Add your own comment

114 Comments on New GPU Startup Bolt Graphics Promises 2.5x Path Tracing Performance of RTX 5090

#1
Chomiq
Here's hoping they'll actually deliver on their promises and won't get eaten by Nvidia.
Posted on Reply
#2
Quicks
Don't help much if its good at path tracing, but sucks at everything else.

If anything it will be bought by one of the major players very quickly if claims hold true. But probably just playing the game like Elon of overhyped products and then under deliver.

Intel, needs a win.
AMD, there head are screwed on the wrong way lately.
Nvidia, there head is so far up their own arses doubt they see anything.
Posted on Reply
#4
Theswweet
Why do I get the feeling that their test methodology is testing the VRAM versus the actual performance of the chip? It's odd that a card targeting data center usage is comparing itself to the 5090 rather than the RTX Pro 6000 with 3x the VRAM.
Posted on Reply
#5
Dr. Dro
They've been making this claim since like 2024? I'm more than just a little skeptical they have anything of substance to show by now.
TheswweetWhy do I get the feeling that their test methodology is testing the VRAM versus the actual performance of the chip? It's odd that a card targeting data center usage is comparing itself to the 5090 rather than the RTX Pro 6000 with 3x the VRAM.
I'll just paraphrase myself on the last time this thing came up
Dr. DroCompletely out of the question, the rendering pipeline doesn't work that way, and with when ray traced graphics are so computationally expensive, you're not getting that price or that wattage even if this was somehow feasible. There's a matrix to follow:

If a product is fast and cheap, it's not good
If a product is good and fast, it's not cheap
If a product is cheap and good, it's not fast

You can't have all three
I just don't trust these guys. It took Nvidia 30 years of constant evolution to achieve Blackwell. I will be delighted if they actually deliver, but I treat this as nothing more than a proof of concept and an attempt to upsell this startup and keep capital coming.
Posted on Reply
#6
Rus4kova
Actual pictures of the hardware itself or it didn't happen.
Posted on Reply
#7
Dr. Dro
Rus4kovaActual pictures of the hardware itself or it didn't happen.
It was never shown, they just claim it "exists" and provide internal, controlled benchmarks. JPR has what's pretty much the most details ever released on this Zeus "GPU", it seems to be RISC-V based and there are several buts, ifs, and caveats to their marketing. The claim last year was that the lowest configuration, 1c, had 3.25x GB202/RTX5090 RT, which is simply a ludicrous claim - not even AMD and its vast expertise could hope to match that level of performance as it stands today.

www.jonpeddie.com/news/bolt-graphics-zeus-gpu-technical-overview/

The Zeus 1c configuration is purported to have only 10 TFLOPS FP32, this makes it somewhat slower than a laptop RTX 3060. The usage of RISC-V clusters is also reminiscent of Intel's old Larrabee GPU, and it likely suffers from the same problem: absolutely abysmal, unscalable general purpose and raster performance, leaving it - provided that their claims are both true and provable - as a hyperspecialized ray tracing accelerator or a (slow) compute device that has a very high memory capacity. The thing they're doing with RAM was tried before by AMD with the Radeon Pro SSG, which was a Vega 64 that could take M.2 NVMe SSDs as a memory expansion.

Everything I know about GPUs and how they work at a fundamental level simply don't click when I look at this thing.
Posted on Reply
#8
halcyon
You know we are at the top of bubble dynamics when scams like these get so many headlines.
Posted on Reply
#9
TSiAhmat
Dr. DroIt was never shown, they just claim it "exists" and provide internal, controlled benchmarks. JPR has what's pretty much the most details ever released on this Zeus "GPU", it seems to be RISC-V based and there are several buts, ifs, and caveats to their marketing. The claim last year was that the lowest configuration, 1c, had 3.25x GB202/RTX5090 RT, which is simply a ludicrous claim - not even AMD and its vast expertise could hope to match that level of performance as it stands today.

www.jonpeddie.com/news/bolt-graphics-zeus-gpu-technical-overview/

The Zeus 1c configuration is purported to have only 10 TFLOPS FP32, this makes it somewhat slower than a laptop RTX 3060. The usage of RISC-V clusters is also reminiscent of Intel's old Larrabee GPU, and it likely suffers from the same problem: absolutely abysmal, unscalable general purpose and raster performance, leaving it - provided that their claims are both true and provable - as a hyperspecialized ray tracing accelerator or a (slow) compute device that has a very high memory capacity. The thing they're doing with RAM was tried before by AMD with the Radeon Pro SSG, which was a Vega 64 that could take M.2 NVMe SSDs as a memory expansion.

Everything I know about GPUs and how they work at a fundamental level simply don't click when I look at this thing.
also apparently 3.25x times faster in ray-tracing on a 120W budget (compared to a 5090 with 575W ~ 600W )

Wouldn't that also imply it was at least 2x faster and 8x more efficient than a 5090 in raster? (or is that not how it works??)

Also no mentions on what system specs they tested it... (OS, Processor etc...)

---

Also that Graph comparison looks funny (yellow bar goes over the line)

Posted on Reply
#10
Dr. Dro
TSiAhmatalso apparently 3.25x times faster in ray-tracing on a 120W budget (compared to a 5090 with 575W ~ 600W )

Wouldn't that also imply it was at least 2x faster and 8x more efficient than a 5090 in raster? (or is that not how it works??)

Also no mentions on what system specs they tested it... (OS, Processor etc...)

---

Also that Graph comparison looks funny (yellow bar goes over the line)

Yeah, which is why I mentioned that older post of mine with the matrix of good/fast/cheap. You know what would really impress me? If they sent one of their engineering guys to actually clarify and debate us on their approach once this product launches (if it ever does). Not with a press release, controlled AMA or interview: just have the greater nerds posting here on the forum alongside us, lesser nerds dropping knowledge bombs. Massive way to build credibility as a startup, but eh.
Posted on Reply
#11
TSiAhmat
Dr. DroYeah, which is why I mentioned that older post of mine with the matrix of good/fast/cheap. You know what would really impress me? If they sent one of their engineering guys to actually clarify and debate us on their approach once this product launches (if it ever does). Not with a press release, controlled AMA or interview: just have the greater nerds posting here on the forum alongside us, lesser nerds dropping knowledge bombs. Massive way to build credibility as a startup, but eh.
tbh I would be very impressed if they (just out of nowhere) could produce/manufacture a GPU that would be around a 5060ti/9060xt performance level with decent, even if faulty, documentation & drivers.
(even if the end product wasn't viable cost wise/too expensive)

We already saw with intel that it's no easy task to "just" create dgpu division (and Intel is humongous company, with a lot of smart heads & resources)
Posted on Reply
#12
lexluthermiester
ChomiqHere's hoping they'll actually deliver on their promises and won't get eaten by Nvidia.
I'll second this, on both points!
FouquinDeja vu.
That's what I thought of. Perhaps we're looking at an evolution of that card?
TSiAhmatWouldn't that also imply it was at least 2x faster and 8x more efficient than a 5090 in raster? (or is that not how it works??)
If true, that is possible. Raster rendering is a different beast from RTRT, so there will be differences and trade-offs.
Posted on Reply
#13
Darmok N Jalad
An actual working and functional GPU takes time to execute. Look at the supposed B770 we've been hearing about--it's been at least a year since the first design was scrapped and we're still waiting for the followup to show (if it does). The problem this company (and any others that may dare to try) will have is that running games is a fickle business, especially when the Green Giant is helping devs optimize them. We'd all love to see another player in the space, but if it sounds too good to be true...
Posted on Reply
#14
WonkoTheSaneUK
This smells very suspect to me.
Why would a GPU in 2026 have a PCIe slot on both sides?
Pics are likely AI slop at the very least
Posted on Reply
#15
john_
Well, if those numbers where true, Nvidia would have bought them long ago.
Posted on Reply
#16
lexluthermiester
WonkoTheSaneUKThis smells very suspect to me.
Why would a GPU in 2026 have a PCIe slot on both sides?
Pics are likely AI slop at the very least
It's a prototyping development board. There are number of reasons for a PCIe slot on both sides, including real-time processor monitoring and debugging. And no, this is very unlikely to be "AI slop".
john_Well, if those numbers where true, Nvidia would have bought them long ago.
Not necessarily. NVidia is big enough now that them buying ANY other entity/company has to go through regulation checks and approvals. There is a solid chance that such a purchase would be disallowed, for very good reasons.
Posted on Reply
#17
Bobaganoosh
Part of me thinks the RT numbers are possible, but only because Nvidia is soooo focused on other things. Also, there are no raster claims here, so it's one thing to calculate rays, another thing to render the polygons being lit up (or not). If it's purpose-built for ray-tracing calculations, it could be considerably better than Nvidia at it...but not game for beans because it can't handle rastering at an acceptable speed.

Nvidia's focus has not been gaming for years now. Their processors are still capable of raster and RT, but the focus is entirely on data-center workloads. So it wouldn't surprise me if someone else came along, completely focused on raster and ray tracing and beat them, while not doing well at data-centers. Do I think this company has done that? no lol. At least not yet. I can't even tell if they have real hardware.
Posted on Reply
#18
Arcturio
Too outlandish to be believed.
Even if the 'stated' performance and power use is correct, other performance aspects will likely suck, limiting its use.
I'm inclined to take another look only when a trusted review or appraisal is published.
It has that whiff of Star Citizen.
Posted on Reply
#19
lexluthermiester
ArtcurioToo outlandish to be believed.
That's what was said about Core2, Ryzen, GTX10X0 and RTX, yet here we are.. I choose to remain objective and optimistic.
ArtcurioEven if the stated performance is correct, other performance aspects will likely suck, limiting its use.
That is an assumption, not a good one.
ArtcurioI'm inclined to take another look only when a trusted review is published.
While that's a fair point of view...
ArtcurioIt has that whiff of Star Citizen.
...this isn't. Not even the same area of tech.
Posted on Reply
#20
Patriot
lexluthermiesterThat's what was said about Core2, Ryzen, GTX10X0 and RTX, yet here we are.. I choose to remain objective and optimistic.
Objectively those claims were a magnitude of difference.... and when it came to raster on gtx to rtx a lie.
Posted on Reply
#21
LastDudeALive
I think there's certainly potential here. Gaming will never happen, there's too much legacy knowledge and infrastructure required to succeed in gaming. How are they going to ensure this works on games from 2006 and 2026 alike? They just don't have the manpower to do that, let alone keep up with new game releases. Just look at Intel's struggles, and they're a massive company that's had years of iGPU development.

However, the high-end rendering and HPC markets are certainly ripe for disruption. Nvidia is in the difficult situation of trying to balance a lot of different markets. AI markets want pure tensor cores, gamers want low-precision calculation for high frame rates, HPC and rendering want high-precision and care less about speed. No architecture can meet all those needs. So if they focus on supporting common rendering engines (including UE5) and make a versatile framework for writing HPC code, they can certainly make a better (on paper) product than Nvidia in those specific markets.
Posted on Reply
#22
lexluthermiester
PatriotObjectively those claims were a magnitude of difference.... and when it came to raster on gtx to rtx a lie.
What? Could you explain that a bit better?
Posted on Reply
#23
john_
lexluthermiesterNot necessarily. NVidia is big enough now that them buying ANY other entity/company has to go through regulation checks and approvals. There is a solid chance that such a purchase would be disallowed, for very good reasons.
Nvidia knows it, that's why they bought Groq.... without buying Groq.
Posted on Reply
#24
melkor_unlimited
TSiAhmattbh I would be very impressed if they (just out of nowhere) could produce/manufacture a GPU that would be around a 5060ti/9060xt performance level with decent, even if faulty, documentation & drivers.
It may or may not be compatible with specifications.
Flawless support of all games with their mess, optimization, mitigations, workarounds, ... not likely.
TSiAhmatalso apparently 3.25x times faster in ray-tracing on a 120W budget (compared to a 5090 with 575W ~ 600W )
Wouldn't that also imply it was at least 2x faster and 8x more efficient than a 5090 in raster? (or is that not how it works??)
Does it only RT things or does it something more (RR denoising, upscaling, raster, ...)?
That is the question.

Its not a big thing to get higher RT performance than rtx 5090 as all rtx series cards have fixed ratio 1 RT core : 4 tensors : 128 shaders.
rtx 5090 has 170 RT cores .. with 640 its wont be a challenge.
TSiAhmatAlso that Graph comparison looks funny (yellow bar goes over the line)
Exaggeration.
Very common in advertisement.
Posted on Reply
#25
Athena
There are so many patents they must avoid or pay licensing agreements, that it doesn't seem possible

However, if they want to make $$$ fast, they just spit out these claims, and just use it as a tactic to get someone to buy them out / "invest" in them, pretty much like Scam Altman, and people buy into the BS
Posted on Reply
Add your own comment
Nov 18th, 2026 04:49 CST change timezone

New Forum Posts

Popular Reviews

TPU on YouTube

Controversial News Posts