MP 7,1 My x2 Duo's are slooow

[smckenzie]

For about a year I've been trying to figure out the interactive viewport lag I get when using Redshift and Cinema 4D. By lag I mean usability lag, beach balling, etc etc. After some messing about I discovered that:

1. A single 5700MPX has no lag issues, just slow at rendering.
2. A single 6900XT has no lag issues.
3. 2 6900 XT's have no lag issues.
4. My two 6800X Duo's have lag issues in any configuration, i.e. using all cards, just one from each module or just a single card from a single module.
5. A 6900XT + 6800X Duo benchmarks only 15 secs slower than the two Duo's together.
6. Two 6900XT's + 1 6800X Duo benchmarks a whopping 40 secs faster than 2x 6800X Duo's.
7. No difference in performance with or without the Infinity Link.

I know the 6900X's are slightly faster cards, but I wouldn't have expected performance issues with these modules.

Is it possible that the Duo's are just not that efficient at transferring data because they both use a single x16 slot or are they bottled necked because they share a single x16 slot?

I can't think of anything else I can do or try, except sell one!

Any insight would be appreciated here.

 

[ZombiePhysicist]

Paging Dr. @maikerukun , Dr @maikerukun ...

 

[maikerukun]

Paging Dr. @maikerukun , Dr @maikerukun ...

LOLOL Thanks for tagging me and sorry for the late reply was filming all weekend, let me take a look...

 

[maikerukun]

For about a year I've been trying to figure out the interactive viewport lag I get when using Redshift and Cinema 4D. By lag I mean usability lag, beach balling, etc etc. After some messing about I discovered that:

1. A single 5700MPX has no lag issues, just slow at rendering.
2. A single 6900XT has no lag issues.
3. 2 6900 XT's have no lag issues.
4. My two 6800X Duo's have lag issues in any configuration, i.e. using all cards, just one from each module or just a single card from a single module.
5. A 6900XT + 6800X Duo benchmarks only 15 secs slower than the two Duo's together.
6. Two 6900XT's + 1 6800X Duo benchmarks a whopping 40 secs faster than 2x 6800X Duo's.
7. No difference in performance with or without the Infinity Link.

I know the 6900X's are slightly faster cards, but I wouldn't have expected performance issues with these modules.

Is it possible that the Duo's are just not that efficient at transferring data because they both use a single x16 slot or are they bottled necked because they share a single x16 slot?

I can't think of anything else I can do or try, except sell one!

Any insight would be appreciated here.

Hmm, This is interesting and I'll have to load up Redshift. I use Octane and run R25/26 because 2023 "if you are using 2023" won't read all 4 of the w6800x GPU's...from what I remember they're not receiving updated drivers from Apple because Apple is now having them focus on Apple Silicon and so in my experience the only solution to getting ZERO LAG and realtime rendering and usage of all 4 GPUs in our config "we have the same config" is to use older versions...you may want to give this a try and see if that solves the problem for you as well.

Unfortunately, until we know what the plan is with the AS Mac Pro 8.1, there's some weird puberty stuff happening and that will keep happening with the 2019 Mac Pro in certain configs...

 

[smckenzie]

Thanks for the responses

Yes, a bit bitter because I paid for Octane last year as they told me an update was coming. They forgot to mention the part about it would no longer work on my Mac. Out of curiosity to see if it's any different I've been trying to download S26 but something is up with the Maxon server as it gets to 256 Mb and then stops, super slow connection as well.

I can't figure it out either. As far as GPU on a Mac goes, this config should be it, but it's not. If you launch the IPR with two 6900's and a Duo, the lag is super noticeable. Can be up to 30/40 secs before it updates for no reason at all. No errors in the log. Turn the Duo off and just use the two 3rd party and it's fine. Also still get the out of VRAM bug, if I enable a Duo for render I'll normally get the out of VRAM error. Ug.....

I'd be lying if I hadn't thought of just going CPU.

If you could tryout RS that would be super awesome. The lag doesn't happen on simple scenes, it needs to be something a little more complex.

Mine benchmarks in 2 mins with both Duo's. Back when BareFeats tested it was 1:47. I suspect that's a RS thing though as people have been complaining that it seem to be getting slower.

FWIW, 2 6900XT's and 1 Duo will benchmark in 1:39s.

 

[smckenzie]

Managed to get S26 but then realized Octane PR14 isn't supported on Ventura.🤦‍♂️

 

[smckenzie]

Well what do you know. Seen as I'd gone to the effort of installing S26, I installed Redshift and tried it with all the Duo's and bam, real time rendering! Like insane fast. Initially I installed a slightly older version of Redshift, then tried the current release and still good, so the issue is with C4D 2023 for sure.

 

[maikerukun]

Thanks for the responses

Yes, a bit bitter because I paid for Octane last year as they told me an update was coming. They forgot to mention the part about it would no longer work on my Mac. Out of curiosity to see if it's any different I've been trying to download S26 but something is up with the Maxon server as it gets to 256 Mb and then stops, super slow connection as well.

I can't figure it out either. As far as GPU on a Mac goes, this config should be it, but it's not. If you launch the IPR with two 6900's and a Duo, the lag is super noticeable. Can be up to 30/40 secs before it updates for no reason at all. No errors in the log. Turn the Duo off and just use the two 3rd party and it's fine. Also still get the out of VRAM bug, if I enable a Duo for render I'll normally get the out of VRAM error. Ug.....

I'd be lying if I hadn't thought of just going CPU.

If you could tryout RS that would be super awesome. The lag doesn't happen on simple scenes, it needs to be something a little more complex.

Mine benchmarks in 2 mins with both Duo's. Back when BareFeats tested it was 1:47. I suspect that's a RS thing though as people have been complaining that it seem to be getting slower.

FWIW, 2 6900XT's and 1 Duo will benchmark in 1:39s.

Hey bud sorry for the late response, on a film this week...

I will try and benchmark RS for you over the weekend, but that said, I'm running 2 w6800x duo's so I won't be able to give you any info vs. 6900's.

Strange situation though and yeah I didn't bother buying Octane because I knew I was going to end up just using PR14 on previous releases once they told us that the paid version would only utilize TWO GPU's at most on Intel Macs SMH.

 

[maikerukun]

Well what do you know. Seen as I'd gone to the effort of installing S26, I installed Redshift and tried it with all the Duo's and bam, real time rendering! Like insane fast. Initially I installed a slightly older version of Redshift, then tried the current release and still good, so the issue is with C4D 2023 for sure.

LMFAO! Just read this response and seeing you are having success now that you've dropped down to S26 and the updated RS lol. Actually out of curiosity...is it caught up with the speed of Octane yet?? If so, perhaps I'll start to transition over to RS finally...

 

[smckenzie]

I can't try Octane because I'm on Ventura. I don't think it's as fast os Octane though from what I remember of PR11.

Just to clarify, I have 2 6800X Duo's and two 6900XT's knocking about. The two Duo's are in the Mac right now and under S26 they do behave better. Perfect? No, but an improvement for sure.

After more messing about I noticed that enabling ODIN which was added to the current version of Redshift, 3.5.12, that certainly makes the lag worse under 2023.

RS 3.5.13 should be released soon which as I understand is supposed to address a number of this performance issues. We'll see but it will probably also need C4D to be improved to.

Although the render times suck, from a usability POV using CPU on Arnold is actually better. At least you can have material previews turned on!

 

[MBAir2010]

Paging Dr. @maikerukun , Dr @maikerukun ...

wow great post!

 

[rondocap]

When I benchmarked 2x Duo 6800x in Final Cut and Davinci, I got all sorts of weird results too, and ended up just sticking to a single Duo.

I am not sure there really is that great support out there for 4 GPUs, even when they say it is supported, accross various software types. In theory should work great, but little issues always keep it from really being implemented well imo.

 

[smckenzie]

When I benchmarked 2x Duo 6800x in Final Cut and Davinci, I got all sorts of weird results too, and ended up just sticking to a single Duo.

I am not sure there really is that great support out there for 4 GPUs, even when they say it is supported, accross various software types. In theory should work great, but little issues always keep it from really being implemented well imo.

Which OS you on? I just rolled back to Monterey, and performance on S26 with Redshift 3.5.10 is much better than with Ventura. The Redshift benchmark was also quicker by 10 seconds.

I compared Redshift realtime rendering with Octane which has always been fast and it's comparable.

Perhaps in my case the mix of Ventura and bugs in the current version of Cinema and Redshift are the culprits here.