Kepler test build 2.58e Standalone (for GTX 6xx cards)
Forum rules
NOTE: The software in this forum is not %100 reliable, they are development builds and are meant for testing by experienced octane users. If you are a new octane user, we recommend to use the current stable release from the 'Commercial Product News & Releases' forum.
NOTE: The software in this forum is not %100 reliable, they are development builds and are meant for testing by experienced octane users. If you are a new octane user, we recommend to use the current stable release from the 'Commercial Product News & Releases' forum.
-
trudogolik15rus
- Posts: 17
- Joined: Fri Oct 01, 2010 3:51 pm
Thanks! And that I got confused in forum branches =)))
-
marcio_max
- Posts: 132
- Joined: Mon Dec 05, 2011 12:17 pm
- Location: Aracatuba/SP/Brazil
Is ordered. I bought a GTX 670 4GB! Now just wait to see. 
| Windows 8 Pro x64 | Octane for 3ds Max/Standalone | i7 860 | 12GB RAM | 1x EVGA GTX 670 4GB / 2x EVGA GTX 670 2GB | Cooler Master HAF 932 |
Just got my EVGA GTX670 SC card.... I get 75% of the performance that I get with my GTX480 card, but damn.. I like the 4GB vs 1.5GB 
Its sooo nice to have plenty of VRAM left.. I bought the 670 card as a test, and I can easily run three 670 cards instead of the two 480 cards I used to have and get a little better total performance.. It will be VERY interesting to see what kind of boost Kepler will get in the future.. From now on it can only get faster
Its sooo nice to have plenty of VRAM left.. I bought the 670 card as a test, and I can easily run three 670 cards instead of the two 480 cards I used to have and get a little better total performance.. It will be VERY interesting to see what kind of boost Kepler will get in the future.. From now on it can only get faster Win7, i7 980 Hexacore, 24GB RAM, 3xEVGA SC 670 GTX 4GB
-
marcio_max
- Posts: 132
- Joined: Mon Dec 05, 2011 12:17 pm
- Location: Aracatuba/SP/Brazil
Is coming! http://www.evga.com/products/moreInfo.a ... Family&sw=marcio_max wrote:Is ordered. I bought a GTX 670 4GB! Now just wait to see.
| Windows 8 Pro x64 | Octane for 3ds Max/Standalone | i7 860 | 12GB RAM | 1x EVGA GTX 670 4GB / 2x EVGA GTX 670 2GB | Cooler Master HAF 932 |
Hi all,
I have my new Zotac Geforce GTX 670 4 Giga, YES !!!
So, DEVS LISTENING, I found something that does Geforce 6xx series less fast than other card. At first Octane Kepler build not read and use the card in the right way. The first important thing is that octane run a GPU clock and read a GPU clock below the real GPU clock (see the screenshot below). You miss 240 Mhz (without OC or Boost) and 318 Mhz if you overclock the card (the boost reading in GPU-Z is the advisable OC, you can get more Mhz but at your own risk and stability
). After the worst point is you lost 320 Cuda unit... 23.8 % of the CUDA unit... That's why the card is not use at 100%. Ok it's right that Nvidia choose to down scale the computation power (with less Mhz on the GPU to avoid manies Watts and gpu temperature), but 23.8% of Cuda core not used... it's not a few. For information, in other bench, the Geforce 670 is faster than some Quadro card (middle card) and use 1344 Cuda Core.
I think it's a way to get Kepler build faster and more powerfull
Below the screenshot :
Best Regards,
Matt
I have my new Zotac Geforce GTX 670 4 Giga, YES !!!
So, DEVS LISTENING, I found something that does Geforce 6xx series less fast than other card. At first Octane Kepler build not read and use the card in the right way. The first important thing is that octane run a GPU clock and read a GPU clock below the real GPU clock (see the screenshot below). You miss 240 Mhz (without OC or Boost) and 318 Mhz if you overclock the card (the boost reading in GPU-Z is the advisable OC, you can get more Mhz but at your own risk and stability
). After the worst point is you lost 320 Cuda unit... 23.8 % of the CUDA unit... That's why the card is not use at 100%. Ok it's right that Nvidia choose to down scale the computation power (with less Mhz on the GPU to avoid manies Watts and gpu temperature), but 23.8% of Cuda core not used... it's not a few. For information, in other bench, the Geforce 670 is faster than some Quadro card (middle card) and use 1344 Cuda Core.I think it's a way to get Kepler build faster and more powerfull
Below the screenshot :
Best Regards,
Matt
-
marcio_max
- Posts: 132
- Joined: Mon Dec 05, 2011 12:17 pm
- Location: Aracatuba/SP/Brazil
My GTX 670 4GB shows the same thing in Octane. Why is it? 56 Cores?
- Attachments
-
Last edited by marcio_max on Sat Jul 07, 2012 4:26 pm, edited 1 time in total.
| Windows 8 Pro x64 | Octane for 3ds Max/Standalone | i7 860 | 12GB RAM | 1x EVGA GTX 670 4GB / 2x EVGA GTX 670 2GB | Cooler Master HAF 932 |
@MattRM
You should have seen 56 cores, so 1288 are missing!
What a beast! 1344 cores and 4go...
Wish I'll can change my workstation before the apocalyptic end of the year
F
You should have seen 56 cores, so 1288 are missing!
What a beast! 1344 cores and 4go...
Wish I'll can change my workstation before the apocalyptic end of the year
F
Win XP x64 | Q6600 2.40GHz | 8Go | GTX470 1.28Go // Win 7 x64 | I7 Q740 1.73Ghz | 6Go | GTX460m 1.5Go
Yes, of course, I read the wrong line, yes, 1288 cores missing. But now core are in cluster on the new geforce 6xx maybe, there are something really wrong in CUDA 4.1 for 6xx series.
@Tchoa : Yes 4 giga is really good and 1344 cores will be better
@Tchoa : Yes 4 giga is really good and 1344 cores will be better
TO DEVS :
I found an other problem a long time ago and don't have time to post, so here the problem and one request :
Problems :
- There're a problem with two deep channel kernel. Too much sampling for Z-depth and position pass (not see in others passes but check could be good). There're not a number artefacts
The Z-depth is after downscale gain in compositing software.
Request :
- It'll be really good to have a matte material, that capture directly in the computed frame (with premultiplied alpha channel on) the shadows from light source, reflection and gi bounce other alpha channel. It's a really good feature that open octane to vfx production for compositing result picture in live footages and open way to complex scene. The material can have multiple checkbox to activate shadows hit, bleeding hit, reflection hit, and maybe refraction. It's quite easy to do it in RSL (renderman shading language) so I think not so hard in a raytracer (standard or gpu).
I Know it's from Fryrender doc...
- I know that the goal of Octane is to be only GPU, but an hybrid mode like GPU/CPU will be good, all others engines started to done it. It's an essential way to Octane to survive to the next two years, like open it to vfx world.
- Subdivided surface and displace is also something missing to Octane.
- Instances also
- And of course get all CUDA devices and real GPU speed
It's my point of view but as TD rendering/lighting/compositing, I think it's right. Your soft is really good, but have limited by obj format (for anim and single frame, I don't know about your plugins, but I wait a blender plugins to buy one of them ). Plugins is a good way to avoid a real and boring pipeline or put a timeline in Octane and a real supported scene description language as RIB (a very easier language, with a lot of doc).
Best Regards,
Matt
likedin: http://www.linkedin.com/profile/view?id ... rk=tab_pro
PS : Tchoa, Tchoa from Montpellier, France ?
PS 2 : if someone can give me the code to have minimised picture in the message corpse, it'll be great, not found it... so boring big pictures...
I found an other problem a long time ago and don't have time to post, so here the problem and one request :
Problems :
- There're a problem with two deep channel kernel. Too much sampling for Z-depth and position pass (not see in others passes but check could be good). There're not a number artefacts
The Z-depth is after downscale gain in compositing software.
Request :
- It'll be really good to have a matte material, that capture directly in the computed frame (with premultiplied alpha channel on) the shadows from light source, reflection and gi bounce other alpha channel. It's a really good feature that open octane to vfx production for compositing result picture in live footages and open way to complex scene. The material can have multiple checkbox to activate shadows hit, bleeding hit, reflection hit, and maybe refraction. It's quite easy to do it in RSL (renderman shading language) so I think not so hard in a raytracer (standard or gpu).
I Know it's from Fryrender doc...
- I know that the goal of Octane is to be only GPU, but an hybrid mode like GPU/CPU will be good, all others engines started to done it. It's an essential way to Octane to survive to the next two years, like open it to vfx world.
- Subdivided surface and displace is also something missing to Octane.
- Instances also
- And of course get all CUDA devices and real GPU speed
It's my point of view but as TD rendering/lighting/compositing, I think it's right. Your soft is really good, but have limited by obj format (for anim and single frame, I don't know about your plugins, but I wait a blender plugins to buy one of them
). Plugins is a good way to avoid a real and boring pipeline or put a timeline in Octane and a real supported scene description language as RIB (a very easier language, with a lot of doc). Best Regards,
Matt
likedin: http://www.linkedin.com/profile/view?id ... rk=tab_pro
PS : Tchoa, Tchoa from Montpellier, France ?
PS 2 : if someone can give me the code to have minimised picture in the message corpse, it'll be great, not found it... so boring big pictures...
The number Octane is getting from the CUDA runtime is probably wrong. If Octane starts a render kernel, it always runs on all CUDA cores. The reported speed may be the speed when the GPU is idle. How fast does Octane render the benchmark scene?marcio_max wrote:My GTX 670 4GB shows the same thing in Octane. Why is it? 56 Cores?
For the deep_channel_kernel: switch on the alpha channel.
--
Roeland