New Release Candidate coming this week?...

[pixelrush]

Well I would be a little concerned that pre Fermi cards can't be supported. What's the issue here? I want to see Octane develop but I'd prefer not to have rapid redundancy of hardware.
Is it likely this will continue? ie Fermi could be obsolescent next year when Maxwell arrives or is this some jump in cuda capability that we can settle into for a while?

[abstrax]

Well I would be a little concerned that pre Fermi cards can't be supported. What's the issue here? I want to see Octane develop but I'd prefer not to have rapid redundancy of hardware.
Is it likely this will continue? ie Fermi could be obsolescent next year when Maxwell arrives or is this some jump in cuda capability that we can settle into for a while?

No, GPUs of compute model < 2.0 have severe limitations when it comes to functions. As a result the CUDA compiler creates one giant code spaghetti without function calls which it then tries to optimize. Unfortunately, the optimizer is a 32-bit application and eventually runs out of memory (with some hackery we could make it use the full 4GB available). There has been no development on the old CUDA toolchain since CUDA 3.x. The changes needed for architectural glass, seem to break the optimizer's neck.

But I can't say for sure yet. We got a 64-bit Windows build working, we need to see how it behaves with other bitnesses and platforms. Even if we get it working this time, it can fall apart any release now. So sooner or later pre-Fermi GPUs will not be supported anymore.

Cheers,
Marcus

[pixelrush]

Just to add to this
GTX 2xx only came out 4 years ago.
While I realise there are quite a few users here who will have a full time job rendering and afford the latest and greatest many others will be infrequent users.
Could Otoy make general policy to support hardware up to 5 years old? (Edit: if possible)
I full well realise technology moves on and that in general people are keen to benefit from it but...
The other thing that concerns me a little is driver march. To keep up with Octane advances we are moving from one cuda release to the next. Sometimes this will impact other apps on your PC that have certified/recommended drivers for them. Just throwing this topic in for general discussion

Most likely people have 64 bit builds these days so perhaps you could drop pre Fermi 32 bit and that wouldn't be so bad

[abstrax]

Just to add to this
GTX 2xx only came out 4 years ago.
While I realise there are quite a few users here who will have a full time job rendering and afford the latest and greatest many others will be infrequent users.
Could Otoy make general policy to support hardware up to 5 years old?

It's not about not wanting to support the old GPUs, but not being able to because of the limitations of the hardware. You can still use Octane 1.10 with them and maybe some more releases, but we really don't see a way to support them forever without stalling the development of Octane.

I full well realise technology moves on and that in general people are keen to benefit from it but...
The other thing that concerns me a little is driver march. To keep up with Octane releases we are moving from one cuda release to the next. Sometimes this will impact other apps on your PC that have certified/recommended drivers for them. Just throwing this topic in for general discussion

As far as I know Octane works fine with all drivers down to 301.42.

Cheers,
Marcus

[pixelrush]

OK I understand.
So we can expect that after v1.1 pre Fermi will no longer be supported for 32 bit and that for v2 pre Fermi is definitely out.
That's cool. As long as thats out there for people.
Perhaps it should appear in the Octane system requirement notes for new purchasers to be aware of.

[abstrax]

OK I understand.
So we can expect that after v1.1 pre Fermi will no longer be supported for 32 bit and that for v2 pre Fermi is definitely out.
That's cool. As long as thats out there for people.
Perhaps it should appear in the sys requirement notes for new purchasers to be aware of.

Again, we try to support the old GPUs as long as possible and on as many platforms as possible, so I can't say for sure. I'm just saying that an end can come any time now, maybe after 1.10, after 1.20 or some time later. I really don't know. Of course, when that's the case we will write that into the specs.

Cheers,
Marcus

[mib2berlin]

Hi, just for info, the Blender Foundation kicks out the support for cards with Cuda Capability 1.3 (GTX 200) for the next release (2.67) because of hardware limitations.
The GTX 200 series start in 2008, I had a GTX 260 but sell it and start with a GTX 550Ti and get nearly double performance and could use the PMC kernel first time on linux.

Cheers, mib.

[pixelrush]

Yup gotcha

My point was that I didn't want people to buy Octane soon thinking of using their old 2xx sli gaming rig and find out it wasn't supported a couple of months later.
I am a little surprised there will be a 1.2 but OK I'm grateful for all donations

[gabrielefx]

don't forget to buy one Titan

I'm going to buy 4 pcs then I would like Octane kernel optimized yet.

[pixelrush]

Yeah OK but what say optimisation means 10-15% code speed improvement, which I suppose is realistic, and then everyone with a slightly older card suffers say 20% loss or something, and still older cards won't run at all earlier than they otherwise might if a compromise of new and old capability had been gone for.
Titan is Tesla based and that is a minority card availability wise and also not in consideration for many people because of cost even though it is good value in comparison to Quadro or Tesla.
Which is more work for coders? Reworking and optimising for each release or even individual cards such that previous generations are retired early, or trying to code with adequate speed for a span of generations and encountering compatibility issues?
I would rather have many features done very well than only some done extremely well.
No sooner the Titan is here and we get Kepler II and then Maxwell. Your shiny Titan will be not so much use in short order because you ask for it to be especially efficient for a short time. In the extreme you will burn customers off with technology churn.