At this point it does not look good. Let’s wait for the Cycles results.
Honestly, not that surprised. Gen to gen, improvement has usually been in the 15-20% range.
All the “10 times faster!” Nvidia talking points are always related to games… Or lately, AI.
I’ve also read that the power consumption will be quite a bit higher.
And I wonder how loud the fans will imitate a vacuum cleaner.
It’s all speculation at this point, card isn’t out yet. This thread is wayyy early
At what point is this speculation? RT performance in Far Cry is taken from Nvidia’s official slide, and the rest is also official data.
Where is this graphic from?
And even then the “ten times faster” claim is only applicable to the NVIDIA GPU in the abstract and could be possibly hindered by other bottle necks that may exist on a persons particular system config or even the software task being performed.
I see Daz studio users getting excited about the new 5 series when all they ever do is load daz content and render stills in an obsolete version of Iray that does not even support RTX path tracing .
It’s official marketing data, sure. It’s not official field data
Are you suggesting that RTX5090 performance is even worse?
I wrote that you have to wait for the results in Cycles. Maybe it won’t be so bad.
I’m getting soo tired of the AI hype.
Where did anyone imply it would be worse?
It’s a “we don’t really know the actual performance in Blender” … no more nor less.
AI is what is now pumping the balloon called Nvidia. It’s what’s selling right now.
At this point it looks like Nvidia just put the RT cores from the RTX4090 into the RTX5090 and cranked them up a bit.
OK. From this point of view, it is “speculation”.
In theory it shouldn’t be too bad for Cycles. I mean 33% actual more RT cores and they are next gen cores, so 1 for 1 should be faster.
Add to that, with Cycles and OptiX, rendering does scale fairly linear, so at this stage, I would expect around a 30-40% jump in Cycles rendering speed.
Of course, compared to the 4090, you are also getting a near 30% jump in price and power usage, so from a relative performance point of view, it’s not going to be all that big an improvement. The 32GB VRAM is very nice of course.
Even so, we aren’t getting a 3090 to 4090 type jump this time, were the 4090 largely was twice as fast.
Not that it matters, for me I’m looking at $4K min, and given that my Youtube channel isn’t exactly pulling in MrBeast levels of cash, I don’t have that sort of money to splash around.
RTX 5090 | RTX 4090 | Difference | |||
---|---|---|---|---|---|
CUDA Cores | 21760 | 16384 | 33% | ||
Texture Mapping Units | ¹ | 680 | 512 | 32% | |
Render Output Units | ¹ | 340 | 176 | 93% | |
Ray Tracing Cores | ¹ | 170 | 128 | 32% | |
Tensor Cores | ¹ | 680 | 512 | 32% | |
Die Size (mm²) | 744 | 608.5 | 22% | ||
Die Node | ¹ ² | TSMC 4NP | TSMC 4N | ||
Ray Tracing (TFLOPS) | 318 | 191 | 66% | ||
FP16 (TFLOPS) | ¹ | 104.8 | 82.58 | 27% | |
Boost Clock (GHz) | 2.41 | 2.52 | -4% | ||
Base Clock (GHz) | 2.01 | 2.23 | -10% | ||
VRAM | 32 GB GDDR7 | 24 GB GDDR6X | 33% | ||
Total Graphics Power (W) | 575 | 450 | 28% | ||
Req. System Power (W) | 1000 | 850 | 18% | ||
MSRP | 1999$ | 1599$ | 25% |
¹ - from wiki: 5090;4090, rest from official nvidia product specs page
² - allegedly: Its not N4 nor N4P, but its custom node made specially for nvidia, 4NP is an extension of the N5 platform with an 11% performance boost over N5 and a 6% boost over N4
When I’m fiddling with Geometry Nodes and it doesn’t work…
“Die, node!”
With a quick&fast letter change, you could say (and create) an Italian swear that could be actually said when things goes wrong with GeomNodes.
I bought a nvidia gpu for over $3k once, the Titan V. I found that I can do everything I want to do with a much cheaper card. I’m not looking to compete with a studio on one card, it just doesn’t make sense. I find myself spending the vast amount of time debugging nodes and creating tools and rigging and animating parameters. When I want to render I can render just fine on what I already have which is 2 x 3060