RTX 5090 beats RTX 4090 by 36% in the Blender benchmark — RTX 5090D is 40% faster than RTX 4090D
Not exactly a huge boost.
Nvidia’s upcoming RTX 5090 and RTX 5090D graphics cards have appeared in the Blender benchmark database, demonstrating decently higher performance than the outgoing RTX 4090 (via VideoCardz).
The RTX 5090 was tested in Blender 3.6.0, while the RTX 5090D used version 4.3.0, the latest. The Blender benchmark database doesn’t offer any details about specific runs, but we know the exact scores since there’s only one result for each card. However, it would have been nice to know details about other components like the CPU and RAM; the database’s search refinement options indicate that the PCs were both running Windows.
As for why these cards are even in the database before launch, it is a mystery, but since Blender will automatically upload benchmark results if it has internet access, there’s a good chance this was an accident. We do at least know that these results were probably from two different computers since the Blender versions differ between the two cards; plus, the RTX 5090D is a China-exclusive GPU, while the RTX 5090 isn’t legal to export to the country, so it would be weird for someone to have both at this time.
| Header Cell - Column 0 | RTX 5090 | RTX 4090 | RTX 5090D | RTX 4090D |
|---|---|---|---|---|
| Median Score | 17,822.17 | 13,064.17 | 14,706.65 | 10,516.64 |
| Blender Version | 3.6.0 | 3.6.0 | 4.3.0 | 4.3.0 |
Compared to their direct predecessors, the two graphics cards perform decently better. The RTX 5090 is roughly 36% faster than the RTX 4090, while the RTX 5090D beats the RTX 4090D by 40%. However, take these numbers with a grain of salt, not only because they’re leaked but also because the sample size for all these GPUs, except the 4090, is relatively small. Even if all the data is legit, having so few results means the numbers aren’t all that definitive.
Still, both results demonstrated the same uplift, even though they were almost certainly derived from different PCs. That lends a significant amount of credibility that the data is both legitimate and reflective of the cards’ actual performance since these scores happening by chance doesn’t seem very likely.
A ~40% performance improvement over the previous generation seems alright on paper, but it’s a far cry from what we saw with the 40 series compared to the 30 series. Across several versions of Blender, the RTX 4090 is roughly twice as fast as the 3090, according to the Blender benchmark database. The RTX 3090 was similarly twice as fast as the RTX 2080 Ti.
It’s hard to come to a firm conclusion based on just two benchmarks, but if the results are accurate, then the RTX 50 series' performance might not be awe-inspiring. That aligns with what we see on paper, as the RTX 5090 only has about 33% more CUDA cores than the RTX 4090, a significant factor in overall performance. It’s already clear that Nvidia is relying on AI to achieve its lofty performance claims for the RTX 5070, and the story may be the same for the RTX 5090.
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
Matthew Connatser is a freelancing writer for Tom's Hardware US. He writes articles about CPUs, GPUs, SSDs, and computers in general.
-
jp7189 Every article about the 5090 is the same.. we get it. Multiply everything about the 4090 by 1.33333 and you get a 5090.. cores, vram, power, price, performance.... and so on.Reply
Oh.. and we get fp4 support. -
Pierce2623 How can the 5090D be any faster at native without falling afoul of restrictions? Wasn’t the 4090D right on the limit? Or is the fp32 compute about the same and the 40% faster is using frame gen BS?Reply -
OldAnalogWorld We've already discussed this - now normalize the performance of the 5090 relative to the 4090, taking into account the difference in consumption (575W vs 450W). And it turns out that the difference in performance is ridiculous - only +5-10%. And this is with a significantly higher price and huge heating of the room even with the 4090...Reply -
dipique Reply
That's like normalizing horsepower on cars to fuel economy. No, a 400hp car @ 17mpg is not about the same as a 200hp car at 34mpg. Those cars are obviously for different people, and one of them obviously costs (and should cost) more.OldAnalogWorld said:We've already discussed this - now normalize the performance of the 5090 relative to the 4090, taking into account the difference in consumption (575W vs 450W). And it turns out that the difference in performance is ridiculous - only +5-10%. And this is with a significantly higher price and huge heating of the room even with the 4090... -
ihateregister Synthetic tests in blender have absolutely nothing to with texture heavy video games. Also, it's pointless to check specs on 5090. The rumor mill is that it's another paper launch. There will be no 5090s for sale at launch.Reply -
Alvar "Miles" Udell But the 5090 can predict the future and generate multiple fake frames so it doesn't need to actually be faster!Reply -
Joseph_138 What the heck is a 4090D? Is it a disabled card for the Chinese market, like the RX 7900 GRE?Reply -
AndreX86 It's about 23% faster than my stable and overclocked OC 4090 AERO. That being said overclocking the 5090 might lead to some interesting results. Think i'll wait for the 6090 and AMD ****X3D @ 6GHZ before I upgrade again, barring the 6090 isn't $2.5k.Reply -
Peksha Reply
For example, to melting of the power connector;)AndreX86 said:That being said overclocking the 5090 might lead to some interesting results. -
Phaaze88 Reply
Yep. 4090D(ragon).Joseph_138 said:What the heck is a 4090D? Is it a disabled card for the Chinese market, like the RX 7900 GRE?