RTX speed increase?

[charlyxxx]

Hm, today tested on a simple scene with hair and had these results =) 2x2080Ti + NVLink + core i9 9900K

CUDA - 2.30
CUDA + CPU - 2.26
RTX - 7.45 =)

[Joelaff]

Hm, today tested on a simple scene with hair and had these results =) 2x2080Ti + NVLink + core i9 9900K

CUDA - 2.30
CUDA + CPU - 2.26
RTX - 7.45 =)

Wow! So RTX was over twice as slow?? Even with a processor like the i9 9900K (not slow, but exactly fast) the hybrid is still faster, and that is with two 2880Tis! Holy cow. I would not have thought that.

(Or am I mis-reading?)

It is very scene dependent here.

[a7az0th]

It’s exactly post like these that mislead people and start false trends. RTX in my tests (and they are hundreds) has never been slower tha cuda on RTX hardware. It might be a tiny bit faster, or a lot faster, depending on the scene, but never slower, especially with hair where it particularly shines.

What build version did you use to get that result? Stable or nightly? Nightlies tend to break every now and then and it could have been a broken one.

Could you send the scene you used to get these numbers? I am very curious to see it.

[a7az0th]

How are you currently handling it for the mixed setups on the same machine? (Like the example you provide of 750ti, a 1080 and an RTX 2070). Surely large blocks for all of these would result in waiting for Last Blocks on the slower GPUs, no?

You are absolutely right about that and the bucket size is chosen to be uniform across the whole rendering while being large enough to suffice the needs of the biggest GPU (in the example the 2070)

[francomanko]

It’s exactly post like these that mislead people and start false trends.

Well clearly your statement that its always faster isnt correct, otherwise i wouldnt have started the thread. The thread being based on the release version of next. Maybe you could give me one of your test scenes?

[charlyxxx]

Wow! So RTX was over twice as slow?? Even with a processor like the i9 9900K (not slow, but exactly fast) the hybrid is still faster, and that is with two 2880Tis! Holy cow. I would not have thought that.

(Or am I mis-reading?)

It is very scene dependent here.

RTX is slower. than CUDA with Hair in my scene. In other tests - RTX is faster x2-x5 times without hair, but with many objects =)

[charlyxxx]

It’s exactly post like these that mislead people and start false trends. RTX in my tests (and they are hundreds) has never been slower tha cuda on RTX hardware. It might be a tiny bit faster, or a lot faster, depending on the scene, but never slower, especially with hair where it particularly shines.

What build version did you use to get that result? Stable or nightly? Nightlies tend to break every now and then and it could have been a broken one.

Could you send the scene you used to get these numbers? I am very curious to see it.

Ok. No problem. I will prepare the scene and video. Give me some time =)

[charlyxxx]

It exactly posts like these that mislead people and start false trends. RTX in my tests (and they are hundreds) has never been slower tha cuda on RTX hardware. It might be a tiny bit faster, or a lot faster, depending on the scene, but never slower, especially with hair where it particularly shines.

What build version did you use to get that result? Stable or nightly? Nightlies tend to break every now and then and it could have been a broken one.

Could you send the scene you used to get these numbers? I am very curious to see it.

I don't want to make antiPR, but I saw what I saw =) Hope you will fix this in the next build and my scene helps to do this...

So. This is a simple scene

Video with my test -> https://www.youtube.com/watch?v=p77qDDzxKwg

Scene -> (3ds max 2020, V-Ray 5 GPU Hotfix 2, Ornatrix 6.3.5.24029 [Latest Stable]) --- > https://bit.ly/3iTreoN

RTX - 9:29
CUDa - 2:43
CUDa + CPU - 2:37

Standard GPU settings =)

Why RTX slower in this case? What I'm doing wrong? =)

Thank you and have a great day
Best regards,
Andrew.

[Muhammed_Hamed]

In this scene, it is 60% faster with RTX compared to Cuda,
This renders in 12 minutes on a single GPU, 3600 x 3200 resolution ..which is quite impressive

https://www.youtube.com/watch?v=q__Erbr6VGU

I have never had a case where RTX was slower than Cuda.. (it has to be a side case/bug honestly otherwise) people should report these to be honest if they they really care.
For pretty much any interior scene, RTX has been at least 50% faster for me.

I will send you this seen shortly francomanko so you take a look

On another note, I don't know if you realized this or not.. Andrey kozlov used your post to publicly bash Vray GPU on the Fstorm group, claiming that RTX is useless and it is why he never implemented it for Fstorm
It is shame honestly, and he got a lot of backlash on that post.
And not the first time he does that, last time he did a comparison on an evermotion scene that doesn't use Adaptive domelight. Then when he was called out on that by Tomasz and others he started banning people and deleting comments.

I will share more examples here tonight.

[francomanko]

Firstly, scene looks great and much appreciated. Secondly, no I wasn't aware that my post was used by the fstorm guy as I dont go on their forums Bit odd that he did to be honest but I guess thats his call.

Having only recently bought a rtx card and had quite a frustrating time with bugs in both next and vray, ( which ive reported, sent scenes etc)..my initial rtx tests were showing little or no improvement so it was compounding my frustration a bit. Anyway, I wanted to see if someone had an example of a scene that took advantage of it, mainly to check it wasnt something else etc.

So will be good to test..thanks for that.

[charlyxxx]

a7az0th Any news? =) Did you check my scene with RTX?

[a7az0th]

charlyxxx Didn't have time to look into it yesterday. I just downloaded it so I would let you know my findings. In any case, we're talking about a bug here. This is not the standard RTX behaviour. Once we figure out whats going on we will fix it.

[charlyxxx]

charlyxxx Didn't have time to look into it yesterday. I just downloaded it so I would let you know my findings. In any case, we're talking about a bug here. This is not the standard RTX behaviour. Once we figure out whats going on we will fix it.

Ok. Thank you. I will wait the fix and then test again =)

Have a great day.

[a7az0th]

Just to let you know I have not forgotten Both the issues with On-demand mip-mapping and RTX have been reproduced locally and we're working on a fix. This is definitely not the expected behaviour. It's vacation season and half the office is out, so it might take a bit longer than usual to resolve, but when we have a fix at hand I will let you know. Cheers

[francomanko]

awesome, thanks for that.