Blender’s long-awaited 3.6 model has lately launched, and whereas this text is coming later than we deliberate (it’s been a really busy summer time), we’ve been ridiculously keen to speak about a number of the efficiency modifications which have come together with it.
As with all new Blender variations, 3.6 introduces a variety of recent options and added polish, however above all, the function that stands out to us most is ray tracing acceleration for AMD and Intel graphics playing cards.
Once we speak of “ray tracing acceleration”, we’re speaking about enhancements made to the final rendering pipeline to assist a given GPU structure render samples sooner. With NVIDIA, we noticed OptiX added to Blender about 4 years in the past, and from the get-go, we may see that the advantages devoted RT cores may deliver have been immense. So, it’s lastly good to see HIP-RT making it to Blender for AMD, and Embree GPU for Intel.
We’re going to maintain this intro brief and candy in order that we are able to deal with the efficiency faster, however for all of what’s new with Blender 3.6, positively try the official launch web page.
AMD, Intel & NVIDIA GPU Lineups, Our Take a look at Methodologies
Earlier than diving into our efficiency outcomes, right here’s a fast have a look at AMD’s, Intel’s, and NVIDIA’s present product lineups, in addition to some check methodology housekeeping:
)
)
)
These lineup pictures embody each current- and last-gen GPUs from AMD and NVIDIA, however for the sake of holding charts much less busy, we’re specializing in solely the current-gen merchandise for this text. NVIDIA’s Ada Lovelace era seems full till SUPER or Ti playing cards come alongside. AMD’s present lineup entails previous-gen playing cards, as most fashions have but to be outmoded.
All the benchmarking carried out for this text was accomplished utilizing an up-to-date Home windows 11 (22H2), the most recent Intel chipset driver, in addition to the most recent (as of the time of testing) graphics driver.
Listed here are some common pointers we comply with:
- Disruptive companies are disabled; eg: Search, Cortana, Consumer Account Management, Defender, and many others.
- Overlays and / or different extras aren’t put in with the graphics driver.
- Vsync is disabled on the driver degree.
- OSes are by no means transplanted from one machine to a different.
- We validate system configurations earlier than kicking off any check run.
- Testing doesn’t start till the PC is idle (retains a gentle minimal wattage).
- All assessments are repeated till there’s a excessive diploma of confidence within the outcomes.
Word that the entire rendering tasks examined for this text could be downloaded straight from Blender’s personal web site. Default values for every challenge have been left alone, so you may set your render machine, hit F12, and examine your render time in a given challenge to ours.
Viewport Shader Compile Instances
In our Blender 3.5 efficiency deep-dive, we explored the efficiency of shader compiling when enabling the Materials Preview mode within the viewport. The uncooked efficiency of a GPU doesn’t matter an excessive amount of with this activity. As a substitute, driver and software program optimizations do.
What we present in our earlier testing was that reboots wouldn’t retain all the shader cache, so enabling Materials Preview would take longer than anticipated. Properly, it is a bug that Intel fastened, and bonus: it impacts all distributors. Let’s see the most recent efficiency:
| Controller | First | Repeat | After Reboot |
| AMD RX 7900 XTX (3.5) | 9 s | 4 s | 9 s |
| AMD RX 7900 XTX (3.6) | 9 s | 4 s | 4 s |
| Intel Arc A770 (3.5) | 23 s | 2 s | 20 s |
| Intel Arc A770 (3.6) | 23 s | 2 s | 2 s |
| NVIDIA RTX 4090 (3.5) | 21 s | 2 s | 14 s |
| NVIDIA RTX 4090 (3.6) | 21 s | 2 s | 2 s |
| Barbershop | First | Repeat | After Reboot |
| AMD RX 7900 XTX (3.5) | 54 s | 24 s | 55 s |
| AMD RX 7900 XTX (3.6) | 54 s | 24 s | 24 s |
| Intel Arc A770 (3.5) | 234 s | 12 s | 186 s |
| Intel Arc A770 (3.6) | 230 s | 12 s | 11 s |
| NVIDIA RTX 4090 (3.5) | 137 s | 3 s | 101 s |
| NVIDIA RTX 4090 (3.6) | 137 s | 3 s | 3 s |
| Classroom | First | Repeat | After Reboot |
| AMD RX 7900 XTX (3.5) | 21 s | 7 s | 21 s |
| AMD RX 7900 XTX (3.6) | 21 s | 7 s | 7 s |
| Intel Arc A770 (3.5) | 64 s | 4 s | 59 s |
| Intel Arc A770 (3.6) | 64 s | 4 s | 4 s |
| NVIDIA RTX 4090 (3.5) | 43 s | 2 s | 31 s |
| NVIDIA RTX 4090 (3.6) | 44 s | 2 s | 2 s |
It’s so nice to see that post-reboot shader compiles are successfully simply as quick to execute as it could be in case you simply reopened Blender.
That is a type of assessments the place the efficiency from a given vendor goes to rely a bit on the challenge. Total, although, NVIDIA proves sooner with repeated shader compiles, however preliminary compiles are quickest with AMD GPUs. Intel’s preliminary compiles take longer than the opposite distributors, however the repeated compiles are tremendous fast.
Cycles GPU: AMD HIP, Intel oneAPI & NVIDIA OptiX
As talked about above, Blender 3.6 introduces ray tracing acceleration for each AMD Radeon and Intel Arc playing cards. AMD features it by way of HIP-RT, and Intel by way of Embree. Embree isn’t new to Blender, however the GPU assist is.
In all of our testing, Intel’s RT implementation is extra secure than AMD’s. It feels secure sufficient to depend on Arc proper now, however since some tasks did not render with HIP-RT, we are able to’t say the identical about AMD for the time being. That stated, we invite customers to offer us suggestions so we are able to higher gauge this.
The beneath graphs present the efficiency variations between having RT on and off. We’re leaving the Secret Deer challenge for final, because it’s one which did not render with HIP-RT. That stated, the efficiency upticks are downright notable:
)
)
)
Not each challenge will magically present a large rendering velocity achieve from RT acceleration, however when the celebrities align, the enhancements are unimaginable. Even the lowbie AMD Radeon RX 6500 XT and Intel Arc A380 noticed dramatic efficiency enhancements within the Secret Deer and Scanlands tasks.
Contemplating the truth that Blender gained RT acceleration for NVIDIA 4 years in the past, it’s actually nice to lastly see the function added for different distributors in Blender 3.6. We imagine Intel’s take to be secure sufficient for actual use, however for AMD, it may be finest to attend for Blender 4.0 (due November).
Eevee GPU: AMD, Intel & NVIDIA
Earlier than tackling Eevee efficiency, we must always observe that the RT acceleration famous above doesn’t influence Eevee. Which means every vendor has a singular alternative to impress, since uncooked efficiency and driver optimizations will matter so much. Let’s test it out:
)
)
)
With Eevee, NVIDIA nonetheless proves the quickest general, however relying on the challenge, AMD can carry out very well, too. Sadly, the identical can’t be stated for Intel proper now. We’re in unhealthy want of both Blender or driver optimizations to lift the set of Arc playing cards from the underside of those charts.
Viewport: Materials Preview, Strong & Wireframe
Viewport efficiency is unchanged between Blender 3.5 and three.6, so most of this knowledge is getting back from the earlier deep-dive. GPUs which have launched since that final deep-dive have been added in, nonetheless:
)
)
)
)
)
)
)
)
As with most scalable assessments, it’s exhausting to foretell which vendor will come out on prime, and that’s just about the case for viewport efficiency. Whereas NVIDIA’s GeForce RTX 4090 is tough to topple, AMD beats it out on some events with its top-end 7900 XTX.
Viewport efficiency scales fairly effectively up and down any given GPU lineup, however for the smoothest doable efficiency, you’d wish to intention for a mid-range card at the very least. Sadly, viewport efficiency is one other space Intel wants work on – the Materials Preview efficiency is okay, however the Strong and Wireframe modes are in want of optimization:
)
)
For distributors apart from Intel, it’s not tough in any respect to get nice Wireframe and Strong viewport efficiency.
Wrapping Up
After poring over the entire knowledge right here, Blender 3.6 strikes us one of many greatest updates to drop shortly, however that’s to not downplay how main every launch feels. The addition of ray tracing acceleration for AMD and Intel can dramatically speed-up render instances, so we’re glad to see the function carried out.
Additionally good is the repair that was carried out for Materials Preview shader compiles after reboots. With every new Blender launch, increasingly polish leads to continuous efficiency enhancements.
In November, Blender 4.0 goes to launch, and it’s set to deliver a powerful quantity of enchancment, together with for efficiency. We’re nonetheless months out, so we’re undecided all of what’s going to make it in, however like at all times, we stay up for digging in!
Help our efforts! With advert income at an all-time low for written web sites, we’re relying greater than ever on reader assist to assist us proceed placing a lot effort into any such content material. You’ll be able to assist us by changing into a Patron, or by utilizing our Amazon purchasing affiliate hyperlinks listed by our articles. Thanks to your assist!
