OctaneRender™ Standalone 3.00 alpha 4

[Notiusweb]

postby ff7darkcloud » Sat Feb 13, 2016 3:13 am
Also using 1x riser configuration for my 3 of my 4 cards, I am having problems too, should I use a 4x splitter from amfeltec?, Is it reliable?, still I hope this will get fixed.

Hello, I have tried both single PCI 1x USB 3.0 risers (16x to 1x), as well as an Amfeltec 4-way expansion cluster, which splits 16x to 1x, and they both cause the problem (for example, if I isolate any one card as the rendering card, it causes a freeze.)
However this is on the current Version 3 Alphas (1-4).
In a Version 2x Octane Render, I find The Amfeltec 4-way expansion cluster to be an awesome rendering weapon, very reliable.

Now, when you say a "4x splitter", I don't know if you just mean a 4-way splitter of some sort, or a splitter that splits from PCI 16x to PCI 4x (as opposed to 1x).
We know 1x causes a problem, so just by mere fact that it would be a 4-way splitter at PCI 1x from Amfeltec, as opposed to some other splitter, will not be something that could help dodge the issue.
With regards to a PCI 4x speed, no one has come forward with PCI 4x being a problem in Version 3, although it could be and we are not aware of it. So PCI 4x it is an unknown currently.
In my case, I can reproduce issues at 1x, whereas 16x never has the issue.

Like you, a couple of us have our fingers crossed on this one.
-Notiusweb

[ff7darkcloud]

Hi Notiusweb, these are the ones I refer to, someone brought them up at the facebook group: http://amfeltec.com/products/flexible-x ... -oriented/

They said those are better than using 1x. Though I think this is the same piece you are mentioning?

BTW the 1x risers I use are USB 3.0 too.

[Notiusweb]

Postby ff7darkcloud » Sun Feb 14, 2016 12:59 am
Hi Notiusweb, these are the ones I refer to, someone brought them up at the facebook group: http://amfeltec.com/products/flexible-x ... -oriented/

They said those are better than using 1x. Though I think this is the same piece you are mentioning?

BTW the 1x risers I use are USB 3.0 too.

Hi ff7darkcloud,
Your link would be a product having a 4x bandwidth connection (as opposed to 1x). It connects the card at the 16x side and plugs to mobo at 4x. Yeah, that's the maximum PCI-speed you could get off a 4-way split from a 16x PCI-lane (16/4 = 4). Mine is in fact different, it is this one, connecting at 1x speed http://amfeltec.com/products/gpu-oriented-cluster/, giving an extension of up to 10ft
The tradeoff with that product your Facebook group is mentioning is the connection length. Now, in my scenario where I need the length, I would imagine possibly trying to daisy chain a powered 16x,or 4x(?) extender to it, but from what I read in bitcoin forums and product reviews, people who tried this type of thing got a non-working stability, if that.

As for the Version 3 Alpha, maybe someone who has that Amfeltec product in your link, or a 4x-bandwidth speed of some other make, could comment on stability in Version 3 with baking cam or high res scenes.
Or if you had a contact in the Facebook group who uses Octane and has the product in your link, maybe they could test a scene of yours that regularly crashes with 1x and give you some feedback how it goes with 4x.
You know, I found I can in fact use baking cam at low res with 1x, the issue is 'provoked' by higher res in a linear way, so a theory of mine is that it's possible the freezing/crashing issue is always present in all speed scenarios, with perhaps 16x yielding max stability, while 1x yields lowest stability. Which means possibly a 4x connection speed may only yield stability up until a point. Yet It is after all, an Aplha, so if 1x stability gets introduced, then it becomes a mute point.

At minimum, the product in your link would potentially be a better performer on Version 2, it would probably load the scene into the display quicker than a 1x connection does and maybe give some minor rendering speed gains, so if connection length was not an issue, it would maybe be a good pickup in any event.

Best,
Notiusweb

[itou31]

Hi f7darkcloud,

I hace the one you linked. it's a 4X PCIe to 4 X 1X PCIe (confirmed by amfeltec), but this has a PLX chip that handle better the "extension" compared to USB 1X that only extend a minimum PCIe signals.
The one Notiusweb has is better in term of bandwith transfer as it can run at PCie 2.0, the handshake/sync signal allow to double the transfer rate as opposed to the on I have. But I think that is not enough for the new V3 Kernel.

[Notiusweb]

Hi f7darkcloud,

I hace the one you linked. it's a 4X PCIe to 4 X 1X PCIe (confirmed by amfeltec), but this has a PLX chip that handle better the "extension" compared to USB 1X that only extend a minimum PCIe signals.
The one Notiusweb has is better in term of bandwith transfer as it can run at PCie 2.0, the handshake/sync signal allow to double the transfer rate as opposed to the on I have. But I think that is not enough for the new V3 Kernel.

Itou31, thanks for updating on that 4x unit. So let us ask you, do you get V3-related crashes/freezes using GPUs connected to the Amfeltec you have, or is it only with the USB riser?

Notiusweb

[itou31]

Yes, I have the issue on the two with the V3, but less (not instanly freeze, or it render a bit before freeze) on amfeltec.

[BorisGoreta]

This is tricky business with PCI extension cords, I found out that specific manufacturers work with my setup but some don't and make GPUs freeze or crash. Seemingly they look the same but they are not. I am not talking about amfeltec but single PCI 16x to 1x extensions. So what I do is buy twice as many as I need and from different manufacturers and then pick the ones that work.

[BorisGoreta]

About that change with 3.0 where now all GPU results are first added and then tone mapped :

I have a lot of hot pixels now when I load scenes done in Octane 2.0.
Now the way it was before where GPU results were first tone mapped and then added together was a very efficient way of getting rid of hit pixels when using multuple GPUs. How pixels are a big problem with GPU renderers and I am not happy about this change. I know there are other benefits for this but still.

Is there a way to have some clamp feature for each GPU before they add up like highlight compensation works on the final image so in effect we would have the same benefit we had before regarding hot pixels ?

[Notiusweb]

This is tricky business with PCI extension cords, I found out that specific manufacturers work with my setup but some don't and make GPUs freeze or crash. Seemingly they look the same but they are not. I am not talking about amfeltec but single PCI 16x to 1x extensions. So what I do is buy twice as many as I need and from different manufacturers and then pick the ones that work.

Boris, can you identify the spec or make that works, which differs from the ones that do not? And, do the PCI 1x ones you have that work, work properly with version 3 alpha on high res baking cam scenes?

In my case, as is with itou31, the ones we have work and do not cause freezes or crashes with any other programs or versions of octane render (ie v2). Crashing only occurs with high res/baking can scenes on v3, indicating there is something working differently with v3 compared to v2 as it relates to stability with a PCI 1x connection.

[Notiusweb]

Yes, I have the issue on the two with the V3, but less (not instanly freeze, or it render a bit before freeze) on amfeltec.

Thanks. This seems like in your case there is added stability with a 4x connection vs a 1x connection, but only up to a point, as the freezing is still there with that Amfeltec.