The Star project rendered in 2m16s in HF3 vs. 2m20s in HF2, so something of a wash.
Now, I mentioned it in my other post, but there was a striking visual difference in your other project; are photos vs. Dropbox an issue, since I tried both a .PNG and a .JPG and neither display below properly. Anyway:
For some reason the link is dropping the 'dl=0'. You'll need to copy between the < > and paste it into your browser. Sorry.
I'm not able to see them.
When i'm rendering this project (atomic particles) in HF3, I get strange yellow/red artifacts. That's it.
It's from your lens dirt effect, when it's in Screen blend mode.
Dropbox Link to screenshot comparing HF2 and HF3 render
Sorry, Dropbox was giving the wrong link
Your link is not working (at least for me).
Just an update - this is being worked on for the next update.As mentioned, Atomic is doing more now than before and therefore moving more data around as a result. Furthermore we have fixed some bugs since HF2 which required us to make some of these changes which have impacted performance. Having said that our devs are working on optimisations to bring performance closer to HF2 even with the new features and fixes. They are, as always, doing an amazing job and we'll keep you posted with their progress. It won't be long.
Thanks Joshua !
Do you think that it will be possible that Nvida family cards will handle particles with similar performance as Radeons?
Here my benchmark on the Star project
Laptop CPU i5 2520M 2.5 Ghz, GPU Intel HD3000, 8GB RAM, SSD 240GB
CPU load around 0% with some peak at 20%, GPU load 100%.
Tomorrow test with my eGPU set.
Here my benchmark on eGPU
Same laptop + eGPU Radeon HD 7950 Boost (= R9 280), Catalyst version 14.9
CPU load 25%, GPU load 87% on HF2U and 85% on HF3P
Meaning with eGPU there is a x15 speed gain, not bad for the very same laptop with an 200 Eu hw update.
Just another update...The dev team are doing well and there is a possible reason the pretty old Nvidia drivers are performing better for some people. We're using a feature which was well supported on both Nvidia and AMD when Atomic was being updated to speed up rendering when multiple instances of the same item were present. This functionality appears to have suffered a major performance drop in more recent drivers so we're refactoring what we are doing to work differently on Nvidia...AMD on the otherhand still performs well.Anyway, we have a plan! More news soon.
Good news, great to constantly get reminded that you're actively working to make HitFilm better on all ends
We're trying to be a bit more vocal about what's going on behind-the-scenes with HitFilm 3. It benefits everybody to know what's going on, and that there's a plan.
Just ran a test on my work machine, which has an i5 and an NVIDIA 970. Bear in mind that NVIDIA cards are the most affected by this issue, due to not-entirely-brilliant changes to drivers made by NVIDIA over time.
I rendered the first second of Arek's sample project when conducting these tests. We'll be doing further, more extensive tests but the results of this were too good to not share.
Here's how it shakes down, for rendering the same 1 second of Arek's project:
HitFilm 2 Ultimate: 9 minutes 40 seconds
HitFilm 3 Pro public release: 14 minutes 50 seconds
HitFilm 3 Pro development build: 31 seconds
Soooo....the devs have not only brought HitFilm 3 back to HitFilm 2's performance, they've actually smashed past it.
We're aiming to release this update a week on Monday (8th December), once it's gone through some more testing. There'll be lots of other improvements aside from this atomic performance boost, too.
Performing the same tests as Simon above but on my iMac with the NVIDIA GeForce GTX 780M 4096 MB...
The problems were not the same driver ones as under Windows so the issue was never so huge, but the changes we've made have still helped a great deal... Check it out
HitFilm 2 Ultimate: 1 minutes 6 seconds
HitFilm 3 Pro public release: 1 minutes 56 seconds
HitFilm 3 Pro development build: 32 seconds
Slightly annoyed that Simon beat me by 1 second!
YES YES YES :D:D:D!!!
@ SimonKJones Holy crap! The fact that the dev team was able to get a 15 minute render down to 30 seconds blows my mind! Great work guys!
I would have to see that dev speed in a project on my system to actually believe it. I am guessing it is a hardware specfic result with a $500+ card with lots of cores.
I wonder what that would be on my Nvidia GTX550 TI 2047MB card, doubt 32 seconds.
Gotta luv the devs though.. keep going with the optimizations!
A 970 vs a 570? It won't be 30 seconds. Maybe 3 minutes...
Yeah, your mileage will vary depending on your card, but the improvements will be significant on Windows machines with Nvidia cards.
I finally found time to test the latest build on my home machine, which has a considerably older and less powerful NVIDIA 560ti. I haven't done Arek's benchmark, but in general usage it is immediately apparent that atomic is vastly faster. Can't wait for you guys to get your hands on it.
Will there be an update tomorrow (8th December)?
Intel Core i7
Gigabyte GeForce GTX 780 OC - 3GB
Just found the time to read through this post.
Kudos to the devs!!!!! My NVIDIA is a 555M- but I'll bet I'll see an improvement too...........I hope.
@Lars_H - we had planned for a Monday release but it's been delayed slightly due to illness in the team, which put a couple of people out of action last week. The intention now is to release mid-week - keep an eye on the blog.
Sign in to comment