Turned v-sync off but Soft doesnt seem to be obeying that.
I will say you will not be disappointed in Titan for CUDA rendering. In our testing a single Titan beats a Quadro K5000+Tesla K20X in Arion pretty easily. $1K for Titan. $7K for Quadro + Tesla combo.
actually I was looking more into 780. from some reviews is doesn’t fall too much behind performance wise but price wise pretty much.
but still would love to see v-sync off results as any card now can push that to steady 60fps.
for example here are some results I did when I replaced gt580 with 7970:
GTX580:
cinebench OpenGL: 60
Softiamge, cube subd 831x800x800 shaded view rotation: ~17fps, after totating it for like 30sec or something it speeds up to 80fps. doesn’t work all the time.
Maya 1 riged cahracter simple scene: rotating cam 35fps-40fps, playing 4-5fps
HD7970:
cinebench OpenGL: 90
Softiamge, cube subd 831x800x800 shaded view rotation: ~90fps, no need for half of min rotation, goes smooth from start.
Maya 1 rigged character simple scene: rotating cam 70fps-115fps, playing 24-26fps
Same system just card replaced. Witht that much performance gain in viewport and CUDA rendering not so high in proprity list back then that was simple solution.
Now with redshift testing and looking so sweet, time to look at some CUDA cards again. Maybe 
They also mentioned working on OpenCL version to but not any time soon 
hi
amd opteron 8350 at stock + 7970 at stock settings 172,5 fps. Catalyst 13.4
Cinebench 72,2 fps
best regards
Sven, render4you
that sounds abut right. forgot those are ooold results with some old drivers.
Getting ~170 in Si 2013, a bit lower in SI 2014 which I’m not happy to see but…
Also after some overclocking play it pushed up to 200fps.
In any case compared to gtx580 it was world apart. Now wondering about titan and 780
I get this when I try to run that script:
ERROR : Cannot use parentheses when calling a Sub - [line 1]
Someone up there metnioned some problem ad did this:
I just ran this in a fresh install of Soft 2014. Had to change the “True” to “1” to get it to run.
Also check if script language is python selected
set the script editor to Python, it’s probably defaulting to VBS.
Make sure you turn vsync and cap playback off before you run it though, or everybody will get 60 fps 
As tests go it’s not indicative of a lot though.
You can get videocards doing ok in that test that will absolutely crap themselves in a lot of scenarios.
All it gets you is a semi-accurate draw/fill rate clocking.
I’ve finally had a chance to bench both the 780 and the Titan (remotely, I compiled a few things for a friend to run and send me the output of).
Sadly, while not as badly as the 680, the 780 is still driver crippled when it comes to quite a few things. In some (non artificial) tests it still gets beaten by a 580, where a titan easily doubles out the performance of the 580, and is only an inch behind a k5k.
OCL shoot-outs that have been tossed around in the last few weeks also all agree.
It seems for now the only recent GTX out there that truly excels at workstation work remains the Titan.
This was a set of DP tests written in CUDA, so it’s as telling as it gets with no OCL bias questions. Multisampling, ByN naive reduction and large set FFTs all show the 780 crippled.
Shame.
Blah… Heres hoping amd pull their finger out and release something decent this decade…
I share that hope, but I have too much of an investment into CUDA to consider giving it up.
I have limited (very, very limited) experience with OCL, and despite my closet socialist desperately wanting an open standard to prevail, the level of maturity, availability and accessibility of CUDA and nVIDIA’s reliability in many regards, even if they weren’t completely dominating the performance segment to begin with, would make considering AMD not a short term proposition.
At least the push for competition would be nice.
Fingers crossed for mid-gen 9ks.
The Titan -is- a nice card though, if only it was a couple hundred bucks less (it’s 1.2-1.3k here in Oz) I would have probably bought one already. The crippling of the 780 is a severe disappointment though.
Thanks for sharing the results of that test, very useful information! Can find titans around £730 so they’re not that much of a leap from the 780s at £550 in the UK.
I ended up getting one. Will have a better chance to test when I’ll get back to some GPU centric projects.
Guys, Im almost buying a new budget GPU.
I was going with the hd 7950 since it had more raw power, and also because autodesk recommends the hd 7970, so I thought the 7950 would be fine too.
However, after reading what you guys say, it seems it’s still not the time for buying amd stuff.
Is that really true? Will I get issues running new version of maya with an hd 7950, lets say?
Compared to a few years ago things got a lot better, and in dx11 the differences are small (we’re talking reliability, performance for money AMD/ATI has always done very well), I do still routinely hear of glitches with AMD cards which are almost unheard of with nVIDIA cards (especially classic viewport, transparency drawing and so on).
I think a lot of the stereotype is an old reputation they aren’t quite able to shake off when it comes to 3D DCC with Autodesk products, things are better, but they don’t seem 100% reliable all the time yet.
Personally I’m hoping AMD will do well, but am not buying yet. Even if I didn’t have an investment in CUDA development (and I do), I would probably still have bought a Titan. A GTX 580 can be had really cheap and still does well with Maya if you’re on a 250-350 budget.
This is mostly hearsay though, I haven’t owned an AMD in quite a while, and the last time I benched one before purchasing was at least a couple years ago (when it still had the horrible selection lag issue which, from what I hear, is gone now).
I heard also that with recent drivers, all possible issues with amd and 3d apps would be solved.
Anyway, I was thinking about the gtx 660 ti. It scores lower than the hd 7950, probably due to that nvidia crippling thing and lower ram.
So you would still trade the so called nvidia reliability for higher performance?
Personally I’d never, ever, trade reliability for performance. Nothing kills productivity like glitches and latency.
Sheer draw and fillrate performance mean absolutely nothing when it can take up to 10 seconds just to select a heavy mesh (this particular issue allegedly now solved, but just for the sake of example).
The 6xx cards are also crippled only when it comes to DP (double precision computation soft tuned down to 1/24th), it’s not something cropping up much when you’re just modelling and animating in Maya to be honest. The 7xx are crippled too, but not as badly, the Titan is fully DP capable like any Kepler quadro would be.
I would at least stretch it to a 670 though, that can be found for the same price of a 7950, at least here.
I would also buy 2nd hand, with the 780 out a lot of 670 and 680s that are less than a year old are cropping up for cheap on eBay and local websites.
Alright think Im going with nvidia than.
But something curious is that autodesk does not recommend the gtx 670, but only the 660 and 680.
Why is that? Are there any real causes behind that or they just forgot to add the 670?