Do you hear that? That’s the sound of your GPU trying to ray trace your scene and your fans kicking in. Perhaps you’re looking for your next card to purchase, or just trying to see whether your current graphics card is up to par. You’re tired of waiting for Cycles to finish sampling. This guide will help you select the best GPUs to solve this pesky problem.

At Renderjuice, the Blender render farm, we’ve learned a lot about GPUs because of what we do. More specifically, we’ve learned a ton about how different GPUs work with Blender. A huge variety of different Blender render jobs come in on the daily, and our job is to best match the best machine specs for each job. From all of our learnings, below are the recommendations we’ve come up with for the best GPUs for Blender for 2024 and going into 2025.

NVIDIA Geforce RTX 4090 — The Best, but Luxury Option for Blender

Perhaps this is obvious, but the RTX 4090 is by far, the most powerful consumer card for Blender at the moment. Due to its high number of ray tracing cores and VRAM specs, it’s a safe choice if performance and reliability are the highest factor you’re looking for.

It is definitely pricey and sells from a whopping $1599, but it outperforms the RTX 3090 by two times its speeds in our benchmarks.

On Blender, with over 8000 datapoints we’ve tested with, the RTX 4090 will consistently render at two times the speed during the sampling process than the RTX 3090. There’s no question, the 4090 dominates in terms of ray tracing.

If cost for performance optimization is what you’re looking for, the 4090 still is technically great. The average price for an RTX 3090 at the time of writing sits at around $1000 (USD). So if you’re able to scoop up one for under $2,000 (USD) it will still be a better purchase. This, of course, makes the assumption that you’ll consistently be rendering with it. So if you’re still a hobbyist and playing around with Blender, but not on a consistent basis, this is probably overkill.

What to look out for:

The RTX 4090 is massive, physically.

At 12 inches (30.48 cm) length x 5.4 inches (13.72 cm) of width x 2.4 inches (6.1 centimeters) of height.
Make sure your machine’s enclosure can fit the 4090 in the first place. It’s not your typical card. We’ve heard countless stories where the GPU cannot fit into enclosures.

Compared that to the RTX 4060 which has dimensions of 9.4 inches (23.88 cm) of length, x 4.4 inches (11.18 cm) of width, 1.6 inches (4.06 cm) of height!

The RTX 4090 will need power

The dimensions of the RTX 4090 is in-part due to its high-power draw. Ensure you can even drive that much power to your card. It needs 450w watts for max consumption and performance does, in fact, go down if you cannot pass enough power to it. However, even if you can only pass around 380 watts, it should still perform up-to spec.

Cooling

With massive power comes the responsibility to cool properly. Yes, GPU cards typically will come with fans, but if you’re looking to use your card over the long haul, ensure your enclosure is in a cool area and that it’ll keep cool when rendering overnight.

Supply?

We’d be ignorant to recommend a GPU without acknowledging that getting your hands on one of these is not an easy feat. We expect that supply of these graphics cards won’t continue to go up in reality, demand is just too high from the surge of AI developers looking to utilize graphics cards.

Our advice here is to seek out the 4080 series instead if you can’t find a 4090 available. See below.

NVIDIA GeForce RTX 4080 - Excellent for most professionals and cost-efficient

Based on our datapoints, we currently believe that VRAM is still an important factor for most Blender rendering. The GeForce RTX 4080 has 16 GB of VRAM which is the minimum we’d recommend nowadays for professional workflows.

The RTX 4080 can be found currently at half the price of the 4090. But, with around a 24% performance loss versus the 4090, it’s a fantastic purchase.

NVIDIA Geforce RTX 3090 and the RTX 3090 Ti — It Depends!

Last year, we recommended the 3090 Ti as well as the 3090 as the best purchase available. This is no longer the case as expected. Long-story short, the architecture and series really does matter. How else would NVIDIA keep us continuously buying their graphics cards?

Architecture matters

The RTX 3090 utilizes the Ampere architecture while the 4090 uses the Ada architecture. This probably doesn’t mean much without a little explanation. The fabrication of the 3090’s process utilizes an 8nm process whereas the 4090 utilizes a 4nm process. At a high level, this nm size refers to the length of a transistor’s gate, but generally in modern marketing terms refers to the capability of packing transistors onto a chip.

In this table, you can see the power of a smaller process. The RTX 4090’s die is actually smaller than the 3090’s yet it packs more than 2.5x the number of transistors onto the die.

Pricing

The average price of a new 3090 card is still above 1 grand. Sitting at a hefty $1241.87 on Amazon, whilst giving half of the performance of a 4090. If you can shell out a couple more hundred bucks, it might be worth checking out the lower end 4070 and above graphics cards.

Ultimately, the 3090 is still a phenomenal card, but the prices haven’t dropped accordingly. So, if you’re looking to purchase a new card. Go for 4070 or above. If you’ve got a RTX 3090 or Ti already, keep it and learn how to use it with your current setup. You can read our guides on setting up your render farm for that.

What to look out for

Graphics Cards NOT to purchase for Blender

• NVIDIA GeForce RTX 4060 Ti

  • Subpar performance for a 40x series NVIDIA graphics card.
  • Limited at 8 GB VRAM — you may see render crashes.

• NVIDIA GeForce RTX 4060

  • Same as above, it has terrible performance in comparison with other 40x series NVIDIA cards.
  • Limited at 8 GB VRAM — you may see render crashes.
• Any Graphics card with 8GB VRAM

Frequently Asked Questions:

Should you buy two RTX 3090s instead of one RTX 4090?

No. The maintenance and upkeep of two graphics cards is not worth the squeeze and a bit of a headache. That applies both to you and the developers optimizing your program. We also consistently see performance loss when two graphics cards are used together on a single machine, versus when two machines are used with their individual graphics cards.

What about AMD?

We wish we could say that AMD was a contender. Perhaps next year they will be! However, most graphics developers are still optimizing for compatibility with NVIDIA cards. We don’t see the same for AMD. You’ll run into issues trying to use OPTIX or general undocumented compatibility issues with an AMD GPU. It’s a headache to debug when it turns out a feature you want to use isn’t supported in AMD. We’re rooting for AMD to continue to make great progress as it has, but right now, it’s not on our recommended list.

What about Laptops?

Our recommendations still stand, but just note that the laptop form factors of GPUs will not perform the same as desktop GPUs. We’d recommend these laptops or similar:

• Alienware m18 or m17
• ASUS ROGs
• Legion Pro 7i or 9i
• MSI Titans

Razer

Unfortunately, laptops from Razer, we recommend avoiding. Yes they are sleek and beautiful, we admit it. But they’re expensive for the offering and their manufacturing process and quality assurance needs a bit of revamping. Defects are all too common. You might get lucky and score one without any issues but it’s risky. Furthermore, the “sleek” factor of Razer typically means its cooling is less efficient. We see the same thermal throttling on other sleek form factors of MacBooks.

For Blender users, heat is going to be an issue. If you’re ray-tracing expect those rays to create some heat! Thermals are key factors here!

Our Plug:

No matter what GPU you end up going with, if you’re trying not to spend weeks waiting for a long render. Check out Renderjuice! We take our own recommendations and study this stuff day-in, day-out. While you might go purchase a single 4090 to render your scene, we frequently put 30 of those bad boys on for renders. That will be hard to beat.