Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2024 Daz Productions Inc. All Rights Reserved.
Comments
No offence intended, but "not believing" the facts doesn't change them. These guy did real hard tests and posted their results, whereas you're just "imagining" that they can't be accurate. Maybe you don't "want" them to be? And Iray being different from Cycles is irrelevant. The fact is that Blender has managed to get the GTX 1000 cards up and running using Pascal, whereas Daz Studios has not. Just because they use the technology differently, it doesn't take away from that fact. And if they've managed to tweak the 980ti and the R9 cards so that they maximize their performance, and the results are "better" than you'd like them to be, then hats off to the Blender team for their collective skills.
The fact that Titan X "always" did badly in Blender says a lot about Titan X and it's inherent bottleneck setup. The drivers for that card have been out for a very long time now and there's been plenty of opportunity to tweak them to the best of their abilities but they clearly cannot even be tweaked to be as good as a 980ti or an R9. By your logic, these results of the GTX1080 outshining the Titan X should be only a mere fraction of the performance we can expect to see from them in Iray.
Scientists calculations don't mean squat against actual results. Reality always wins...
is no more faster than 23% at max but Blender sucks with Maxwell architecture and is the slowest of all engines with , they may finally fixed it with the Pascal cards but still behind with Maxwell so for me not really comparison to begin with . My friend said 970 render faster than Titan X in Blender .. very fishy benchmarks , more like propaganda at this resolution I would render animation in 47 min with iray not one frame .
It is not the card, it is the engine that can't use the full potential of the card .. let start from that , Blender have the slowest GPU scaling rate of all engines with Cycles if they ever made Titan X run on its full potential as they did with Pascal you will see total different results in the benchmark . Comparing it to iray really not counts here . Wait for the actual iray benchmarks to make your final opinion .
Again, this doesn't really have any impact on my decisions. I don't render in cycles, i render in Iray. If the 10 series is performing incredibly in cycles, but only mediocre in Iray, then it's not worth the card swaps. If you are using blender to render your scenes, then you go right ahead and pick up some 10 series cards, It's only logical to do so. If you are wanting to use pascal generation cards with Iray/Daz then i'd wait for those specific benchmarks.
You seem to have gotten the idea that we are Titan x fan boys who want to defend that particular purchase decision with no grasp on reality. We are very aware of the different pascal integrations, but at this point, with this software, and with this renderer, It's not a good idea to start throwing away Titans for a card that might not even perform sufficiently. When the benchmarks are in, and the bugs are ironed out, i will pick up pascal cards without hesitation. If i can get 4x 1080's for the price of two titans, then that's great news ... but only when they work.
Please keep the discussion civil and remember to address the topic, not the poster.
When it comes to Iray, scientist calculation using CUDA comparaison between maxwell and pascal DO mean something. Iray is indeed a calculation itself and does not use the graphic part of the 3D engine for gamers.
I think perhaps bringing other render engines into this isn't going to achieve much at this stage, just speculation. I would expect an official nvidia framework/toolset to be far more reliable and efficient than third party equivelants, so i am quite excited to see the results. But again, i would just rather people wait for official benchmarks and driver feedback before making expensive/unnecessary changes on impulse.
I have the 1080 & Windows 10 so if it works as fast for Iray as the stats show for Blender I will be more than happy but I don't expect that so won't be disappointed if it doesn't turn out that way.
I am easy to please since I chose this over my 970 to have the 8gb Memory and will be happy when I can use the card to render & test it myself, any speed increases will be a bonus but since I had to let my 970 go to someone else I am rendering on CPU right now so whatever happens it will seem fast to me - :)
My point is that Blender already has the Cuda's working in cycles...so what's the hold up with Iray?
Because HOW they are using CUDA in Cycles is different from how CUDA works in Iray. Getting an engine to work with a new architecutre isn't that hard. But getting it to work PROPERLY with that architecture, so that it leverages all the new benefits and features is a WHOLE different matter.
CUDA (like Iray) is an API to a library of code. It provides an interface to the functionality of the cores on the device. Some elements are exactly the same as in prior versions of the library (this is to provide backwards compatibility with older apps) but they may not run as well as on the prior architecture. NEW elements in the interface may provide better performance, but require the way data is passed to them to change, require data that wasn't provided before, or any number of other factors. So using the 'new' interface elements requires a lot more effort.
Also, as the architecture has changed between Maxwell and Pascal, the very order in which you pass data can affect the performance. There are literally hundreds of interface changes, but to maintain compatibility with older devices (Maxwell, Kepler, etc.) they have to keep those interface methods available. That doesn't mean you would want to use them with Pascal, as they may actually run slower than the newer method available IF you are running a Pascal device.
And (unless I've missed an announcement from nVidia) CUDA 8 is available as a RC (Release Candidate) which is NOT the same as a release. A Release Candidate is a BETA, that IF nothing major is found to be having issues, will be promoted to a Release. That means they BELIEVE they've fixed the bugs, but they aren't 100% certain they didn't break something else in the process. Any commercial software that depends on the library is NOT going to release a build using a RC library. They will wait for the full Release version. I expect DAZ will do the same. In fact, I expect DAZ to wait for the full release, then port it in, and release it as BETA for DS until they work out all the new API details for Pascal in Iray.
+1 , there is more to be updated than just the iray engine , beside Cudas, there is also new Optix and additional functions where DS need to be adjusted , and on top of that it has to works with all GPUs optimal and not just focus on Pascal , I would not wanted a software that run on 1 leg , everything need time and if it is not ready then they need more time .Pushing it will not help and other render engines has nothing to do with iray , as it give people false hope .
Don't even think about , Blender is not as fast as iray and never was and is really drag with Maxwell architecture from the beginning especially Titan X as I read a lot of complains since the release plus couple of my friends did not liked the performance , again it is not the card but the engine . I am sure Pascal cards will perform great with iray but never 4 times faster as technically it would be less possible. Better to stay optimistic than disappointed especially with things that haven nothing to do with iray at all .
Evidence? (other than the Titan aspect, I'm aware of that issue).
Saying the numbers are suspect I mean that they don't jive with how the cards compare against another with other applications.
For example, if I was strictly a gamer and relyed on that alone, I'd "upgrade" from a Maxwell Titan X or R9 Fury, expecting those cards to blow the Titan X out of the water for games. As you say, reality will be much different.
Yeah yeah, I know all that. Nobody is claiming that Blender has everything working "PROPERLY". The link is to their release candidate - V 2.78 RC. The threads are clear that this is all Beta. I said that they have the Cudas "working" in Blender, not perfected...The point is that they have "something" useable up and available to users.
You keep talking about Blender being slow compared to Iray, but that's like comparing Daz Studio to cycles. You should be comparing Iray to cycles and Blender to Daz Studio, if you want to make comparisons. How Cycles will implement the Cuda 8's and the Pascal architecture is irrelevant to how the Maxwell architecture and Cuda 7's were implemented in the past. There are many examples in technology where something was crap for many versions and suddenly started shining and eclipsing their own predecessors. Daz Studio is a perfect example. I came into it at version 4.9 and was stunned at how wonderful it is compared to Blender, which I've been using for years. (didn't realize the physics are no where near as good as Blender though). When I look at the previous versions of Daz Studio though, and the earlier models, I see a tremendous leap from V4 to present day models. I hardly ever heard "Daz Studio" mentioned when I was using Blender and MakeHuman, and now it is the buzz of the 3d world in those circles. Until you "see" this new technology performing, you cannot make a judgement based upon past performance. The benchmarks I posted earlier are the first glimpses of what the new technology can do in Blender, using cycles. Perhaps when Daz3d implements it we will find that it isn't all that good. On the other hand, perhaps we will be blown away by the performance increases. For now though, the only evidence I can find for the new technology suggests that it is really good.
The benchmark numbers are for a render, not games. A gamer wouldn't be upgrading based upon results about render times for a single image :)
Daz3D is a company not a program ;P everyone know we talk about Cycles but you have also Cycles in Poser so for that reason I said Blender .. Blender is a great program and better than Daz Studio , Cycles just are not the number 1 engine for knowing reasons .
The benchnarks are for Cycles in Blender what does not affect in anyway the perfomance of the cards in other programs and should be not commpared to other engines ,what happen in Blender stay in blender .. but glad you guys getting finally the correct perfomances.
But even thinking that 1 x 1080 will replace the power of 4 xTitan X in iray is absurd . The power is max 23% faster compared to stock Titanx X Max
You're missing the point - those numbers are not typical of the performance difference between a Maxwell Titan X and GTX 1080.
While they may mean something for Blender, they don't tell us about Iray. In fact for the performance difference of a Titan X vs. the other cards listed for Iray, the way you worded your post is misleading.
+1
There are a couple of features of Pascal (though not necessarily the 1080) that are interesting for Iray specifically.
One is 'Compute Preemption', the ability of the card to know when an instruction is for the monitor and give it priority over any instructions for Compute.
The other feature is on the Tesla cards, that's NVLink. A high speed connection, fast enough to pool GPU memory.
The spec. for NVLink v1 is a maximum of 4 GPUs with max.16GB each, which will allow Iray scenes of 64GB!
It's not clear whether NVLink will work with PCIe slot Tesla cards, or if it requires a Tesla with Mezzanine connectors, making it a server only option.
(Mezzanine is 5-10 times the speed of PCIe3)
You guys are arguing about benchmarks, software, drivers, and API's that are irrelevent to DAZ's Iray, which in itself is a only a specific version of Iray. To compare this product to anything else (i.e. game benchmarks, cycles finally working properly, Octane) or anything other than Daz's Iray is a pointless discussion. Each engine is different, along with each specific application of CUDA and Pascal. They are unique and not mutually exclusive in terms of relative performance. You're comparing apples to oranges; and apples to cantelopes in some cases.
We have to wait for Cuda 8 SDK and the Iray SDK update in their final release, then all the development and testing of those SDK's as applied at DAZ. This takes time, so cool your cherries and stop speculating on performance. We will not know what improvements are in store for DAZ's Iray on Pascal until the final DAZ Iray update is released.
You are mistaken Blender with Luxrender and Cycles. Iray is a rendering engine. Daz3D Studio si a software like Blender but doesn't render by itself. You can use Luxrender with Daz3D studio.
Luxrender is a true physics engine, like Maxwell (Ndidia's render engine like Iray), unlike Iray. Iray is partially only a true physicis engine, so it's partially biased (aka also optimised to render false reality). Despite Luxrender being a true physics engine, Nvidia has tought the guy who makes the benchmark based on Luxrender (Luxmark) how to bias the engine a little bit to make Nvidia cards look better. Nvidia cards don't use ISO maths, unlike AMD, to render in Luxmark 3.1. If you like ISO, by book results, you must go with Luxmark 3.0.
Mind that many officiel benches are based on Luxmark 3.0 or 3.1 depending on the website. Nvidia card look bad in Luxmark 3.0, very far from AMD, because when using ISO full 32bit (not even 64bit) precision AMD cards show huge value. AMD GCN graphics cards are all about parallel computation made by the book, made to fully replace the CPU. There are no biased or any kind of optimization on thses cards. All the silcon is dedicated to those maths. This is how Nvidia beats AMD in recent gaming. Nvidia has all silicon drawn for optimizations and biased rendering. So that's it. The more Nvidia (or AMD) optimize their cards for gaming, the more they'll look bad in comparision in 3D rendering engines. So the same applies to AMD RX480 8GB. This card looks about the same as the Radeon R9 390/390X in performance in gaming and application, but it's more optimized for this. Si in fact it's weaker in brute calculations needed for rendering.
About Titan X compared to GTX 980Ti. Mind that Titan X has bigger use of the same silicon and much bigger VRAM. How is it the GTX 980Ti beats the Titan X in, benchmarks ? There is very simple explanation. You can overclock the GTX 980Ti and its RAM more. But don't be fooled. You'll gain maybe 10% in small average scenes but you will be stuck compared to the Titan X in bigger and huge scenes. Some scenes will evene refuse to render on GTX 980Ti. If you have to choose, and you get some good pricing on a some considered end of life GTX Titan X Maxwell, grab it.
About benchmarks related to Luxrender and Cycles. There have been benchamrks on other GPU render engines of the same kind, and Maxwell (aka Iray unbiased) was among them, and Luxrender showed most of the time to be the best performer, and Cycles was close. So Luxrender IS FAST for what it does and Iray may not be that fast for what it's meant to do.
Thanks for taking the time to post this info, prixat.
- Greg
It is true that Iray is unique in how it uses their GPU.
There may be some people who purchase their new graphic card just based on the Iray benchmarks.
But please keep in mind that some other users actually use other render engines as well.
-> The intention is to find the right GPU that yields great performance in ALL the render engines, games and cuda applications that are being used.
The underlying issue is that it is not a practical solution
- to plug in pascal GPU to play the latest games in highest resolution, and to work with all those software that allready support Pascal
and then
- to plug in the old GPU again to just render in Iray
- - -
-> The missing Pascal support in Iray is preventing users to upgrade their GPU for all other render engines and applications with cuda support as well.
- - -
To get the full picture I am grateful for all those who shared information how the GTX 1080 is performing in other applications.
- - -
The private build change log has this new entry:
Thats the one that support Pascal cards
I suspected this from the developpers, trying to recompile with the new pascal compatible library. It is really encouraging good news.
Well, we are comparing fruit then, at least. It's not like we are comparing apples to vegetables, or apples to rocks, or something. The benchmarks that we are all comparing are giving all of us a "glimpse" into what we might be able to expect from these cards. There is nothing wrong with the comparisons being made in this thread to similar render engines that are using this new technology now.
Why should we all "stop speculating on performance"? If you find this thread to be "a pointless discussion", as you say, then why are you joining in it? Some of us like speculating about this stuff and what we might expect from these cards when they are implemented in Daz. It helps to pass the time while we wait for this to happen, and helps us educate each other about the technology. Why should we all wait until all of this development is completely finished, fully hashed out, and in its final release before we discuss it? What is it that bothers you about us discussing this that you feel you need to tell us to stop?