Official Statement from Otoy towards GTX 980/970

[grimm]

2-Running these tests using Octane demos and the same licensed versions should not make a difference, especially if the comparator does the same, using the same version. The only reason that I suggested using the demos was to make it easier for others, who may not yet have Octane paid licenses, to replicate.

Not necessarily, you have to be careful, as over time the devs have improved the kernels so they might have a slower Ms/sec but will converge faster to a nice image. Also as Octane gains more and more features the kernels will slow down as they have to do more branching, etc. GPGPU is still an evolving science and software speed will always lag hardware capabilities.

I have noticed something strange with my setup so I will test with the scatter ocs file tonight. It does look like my card is getting throttled for some reason but I need to do some more tests.

[smicha]

Sample/px and time is useful to compare different versions of Octane to test for convergence, but you have to compare the images to make it useful. There is a group out there that developed a method to compare noise levels in images but I forget who and where I saw that.

Yes, for noise levels it is important. ( I used it for comparing of how 'path term power' 0 and 1 values affects noise. Thank you for commenting this.

[glimpse]

Sample/px and time is useful to compare different versions of Octane to test for convergence, but you have to compare the images to make it useful. There is a group out there that developed a method to compare noise levels in images but I forget who and where I saw that.

Yes, for noise levels it is important. ( I used it for comparing of how 'path term power' 0 and 1 values affects noise. Thank you for commenting this.

I'm actually working on this topic, as it's quite interesting for me =) will try to test different versions + diffferent parameters tweaked =) not so long ago Marcus wrote a small script, so I can automise things a bit =) here's a brief idea of how I'm going to do that - Testing Noise Levels In Octane Render =)

[Tutor]

Tutor,

I assume you ask someone to do the test on both 6 and 8 powered 980 ,right?

Absolutely, pBarrelas who has a dual 6-pin GTX 980 and Grimm who has dual 8-pin GTX 980.

If so it does not matter which version of octane you'll use - just be consistent for both of these cards.

At this stage, "No." The reason why I picked the two versions that I did was to try to determine whether there is any significant variance based on Octane versioning, but any two other versions might provide the same check.

I assume (probably I am mistaken) that the only difference in 8 powered 980 is that you can overclock it higher, so to compare these two I'd set same clocks for them and do the test.

Absolutely, correct again. The reviewers of the initial GTX 980s (who Glimpse mentioned earlier) opined that the first GTX 980s' (dual 6-pin) energy saving features when out the window when the 980s had to perform compute tasks and that they hit a TDP brick wall and began down clocking. That's why I asked Grimm if he could down clock his GPU because I consider that easier than having pBarrelas up clock his GPU to Grimm's GPU level (+ that would give Grimm's GPU even more headroom). So, in essence, I'm trying to see whether and the extent a headroom issue plagues the GTX 980s (and consequently the other current Maxwells when it comes down to what we do, i.e., render).

Your suggestions are and will be appreciated.

[glimpse]

Absolutely, correct again. The reviewers of the initial GTX 980s (who Glimpse mentioned earlier) opined that the first GTX 980s' (dual 6-pin) energy saving features when out the window when the 980s had to perform compute tasks and that they hit a TDP brick wall and began down clocking.

do You know the fact that nVidia asked to pull of OC 980 results out form the site as it wasn't the best way to reflect cards usage =DDD..I don't remember what were these numbers exactly, but If I'm not mistaken..nearly 400W =DDD

[Seekerfinder]

What I'm trying to determine is whether there is any significant difference in the performance of a Maxwell GPU powered by two 6-pin connectors as opposed to the same GTX versioned Maxwell (here 980) powered by two 8-pin connectors, with both GPUs otherwise configured the same or as closely as possible.

Tutor,
It's a worthy pursuit and I'd be keen to see a comparison done in a reasonably scienific (i.e. consistent) fasion, as you're suggesting. However, I'm throwing my hat in with Smicha and I don't have much hope for the current Maxwell version with Octane. I picked up a long time ago that Octane, while GPU hungry, does not utilise power with the same linear intensity that games seem to do. My own theory is that it's about memory usage - once the scene is loaded Octane seems to do relatively little with it (don't know about animation in this regard).

So, I think we should see pretty much exactly the same results between the two same cards only with different power connectors because Octane will not draw any more power. I'd love to be proven wrong here because it might give the Maxwell's a bit more Octane-cred.

I find it very interesting, and somewhat confusing, that the different Cuda architectures seem to change so much. Glimpse's theory that Nvidia are, as it were, reducing the high-end game cards' compute capability could well be on the mark. But who knows?

Best,
Seeker

[pBarrelas]

pBarrelas did you turn off alpha shadows? Those times seem slow to me.

Like I said, I rendered at default settings, so I didn't turn off alpha shadows.

[Tutor]

Sample/px and time is useful to compare different versions of Octane to test for convergence, but you have to compare the images to make it useful. There is a group out there that developed a method to compare noise levels in images but I forget who and where I saw that.

Yes, for noise levels it is important. ( I used it for comparing of how 'path term power' 0 and 1 values affects noise. Thank you for commenting this.

Smicha and Glimpse,

Bottom line, since at this stage all that I'm trying to determine is whether throttling/headroom issues affect even the dual 8-pin GTX 980 in the same way and to the same extent that they may be affecting the dual 6-pin version, is it your opinion that we also need to know "Sample/px?" I'll be following your opinions on this.

... .do You know the fact that nVidia asked to pull of OC 980 results out form the site as it wasn't the best way to reflect cards usage =DDD..I don't remember what were these numbers exactly, but If I'm not mistaken..nearly 400W =DDD

No, I didn't know that. When you reference "OC 980" are you including the dual 8-pin powered Classified-like versions? Were results, other than power consumed - 400W, asked to be pulled? Were they pulled out? What does that mean?

[Tutor]

... .
So, I think we should see pretty much exactly the same results between the two same cards only with different power connectors because Octane will not draw any more power. I'd love to be proven wrong here because it might give the Maxwell's a bit more Octane-cred.

Before anyone else, like me, purchases an even more expensive GTX 980 for compute tasks, I though it would be good to know exactly what they're up against. If you're correct, there'll be a continued run by content creators to snap up Kepler and secondarily Fermi GPUs.

... . Glimpse's theory that Nvidia are, as it were, reducing the high-end game cards' compute capability could well be on the mark. ... .

I don't doubt that the trend is to move content creators away from low to moderate cost GTX GPUs. I'm just curious about whether it reaches the GTX 980 Classified, as I am/was looking at getting an 8G version when it drops. In the mean time, I guess I have to make due with my GTX 480 Cs (1.7g), 580 Cs (3g), 590 Cs (1.5G), 680 OCs (4g), 690 (2g), 780s (6g), 780Tis (3g) and Titans (6g) and if I need more firepower get it now (from the past offerings) or wait out an uncertain future.

[pBarrelas]

Ok, so here are the results with Alpha Shadows Off:

Scatter.ocs - Octane 2.12.1

DL - 8.11 Ms/sec, 1000 samp/px - 00:01:05
PT - 4.11 Ms/sec, 1000 samp/px - 00:02:08
PMC - 3.06 Ms/sec, 1000 samp/px - 00:02:52

Scatter.ocs - Octane 2.06

DL - 7.75 Ms/sec, 1000 samp/px - 00:01:08
PT - 3.42 Ms/sec, 1000 samp/px - 00:02:33
PMC - Failed

Scatter.ocs - Octane 1.55

DL - Failed
PT - Failed
PMC - Failed