OTOY Forums

at 1150 mhz - as yours - and 7000 mem - 40ms/s (alpha off, coherent off)

smicha wrote:at 1150 mhz - as yours - and 7000 mem - 40ms/s (alpha off, coherent off)

So basically, You can save money, buy less expensive model (regular instead of Ti) then, spend money You've saved for good cooling & thus as a result: would run way cooler, less noisy =) but how about stability, longevity? any thoughts.. does higher OC or Higher temps shorten life of GPU more?

PolderAnimation wrote:nice thnx!

Edit: does memory speed effect the performance?

Here
http://render.otoy.com/forum/viewtopic. ... 0&start=70
I posted some results from an econometric model



It states (almost surely, p-value close to 0) that about 24% performance comes from memory, 76% from core. (coefficients at l_core_mhz, l_memory_mhz - these are logarithms, same interpretation as elasticity in economics).

a thnx, I will look in to that.

glimpse wrote:but how about stability, longevity? any thoughts.. does higher OC or Higher temps shorten life of GPU more?

For usual purposes I use stock clocks. Some time ago I asked the question on overclock.net if it is safe to set 1.2V (stock max volts) for Titan. The answer was - very safe. I saw some guys with modded bios and 1.4V and even 1.7V - this is crazy. So I assume that 1.2V (which is a stock max voltage) should not affect longevity that much, but how exactly, I cannot say.

Yeah, i would not go much higher then that. The manufacturer set this limit, so they dont get any problems with warranty. If you got the money anyway, you can of course try to experiment and see where it leads you.

Anyway, here is the benchmark:

v2.11.1

EVGA GTX 780 SC

~19 DL
~6.1 PT
~1.8 PMC

GPUS: 2xEVGA GTX780s 6GB VRAM @ 1110 MHz (stock - no overclocking)
CPU: i7 4770L @ 3.5GHz (stock - no overclocking)
Octane Standalone v 2.12

Test file: Octane_22_benchmark.ocs
Alpha shadows=off

GPUs in SLI mode

DL: 74 Ms/Sec (first run), 43 Ms/sec (subsequent runs)
PT: 13 Ms/Sec (first and subsequent runs)
PMC: 9 Ms/Sec (first and subsequent runs)

GPUs in Non SLI mode

DL: 74 Ms/Sec (first run), 43 Ms/sec (subsequent runs)
PT: 14 Ms/Sec (first and subsequent runs)
PMC: 9 Ms/Sec (first and subsequent runs)

SLI mode enabled appears to cause only a small speed penalty - possibly within margin of error.

In SLI mode you just have less memory available, it doesn't affect rendering speed.

Interesting. I have been religiously leaving SLI off based on reading somewhere that it 'reduced performance'. How much memory is sacrificed for SLI mode? Yet to strike a memory problem with the one large project so far attempted, though am complete noob with Octane. Found other serious model size bottlenecks however (at least using the ArchiCAD Plugin) that would likely block progress as model gained in complexity long before GPU memory is full. This has been improving with later plugin releases, though at least some of them appear to be caused by limitations in the ArchiCAD API rather than octane itself.

Reading through all the impressive/insane specs posted (5 GPUs, water cooling etc!) am guessing those guys are probably not charging by the hour for their rendering work? Would be better off with a single GTX260.

TitanZ (Octane 2.10 PT ): 15.94 Ms/sec