Cuda and Poser 11 performance info ?.
Have a question about Nvidia Cuda performance with Poser 11 Pro. [O/S Windows 10]
Ive checked out a few threads here and there seems to be more than a few issues
regarding getting the best performence.
I understand what those issues are and the requirements but what i do not see from members who
say they have got setting right is what extra performance/rendering speed they are getting compared
to there previously installed GPU card.
Id like to buy a Cuda card but have no idea as to what rendering speed increases to expect.
Are we looking at a percentage of 100% eg. 10-20% or can i expect to see multiple times the rendering
speed eg. 2 to 10x.
I appreciate that this maybe a difficult question to answer but maybe the following will help.
After you bought Poser 11 and got used to the new Superfly render/setting and say you had a scene
that took eg. 3 hours to rendered then you bought an Nvidia cuda card. What was the increase in render
time that you experienced.
I have a Poser tech support answer to Poser 11/Cuda performance but i wanted to ask members here to relate
there personal experiences so i can make an educated judgment about whether to buy a Cuda card.
Thanks if you can think back to what your first impressions were.
Going from a 700 series to a 1070 is a lot faster, no idea how much thou.
If I do the same render on both machines the 1070 is worlds faster.
More than a few times as well but it varies on a scene basis as well.
Going from 2 gig on a 700, to 8 gig on the 1070 has allowed me to render much more complex scenes as well.
That in itself is a huge bonus.
There are some driver issues with the 10 series cards and CUDA 8 is still not completely finished.
I was hesitant to get the 1070 because of theses issues, but I am glad I did now.
I didn't upgrade, thou, I added another machine. I put the 1070 in a first gen I3 of all things.
Yeah it might be faster in an I7 or Xeon system, but it works fine in a old I3 so that's where it is for now.
Thanks for info much appreciated.
Glad that you were able to give a comparative test with your 700 and Pascal 1070 and that you saw a really good increase in rendering speed.
Quote "1070 is worlds faster."
The best speed report ive seen so far. [elsewere]. is that you can expect up to 3x rendering speed or as low as only 10% increase depending on system/cuda card combination/Poser scene set up etc.
But no sample scene setsups with speed results were given.
Hence the way i phrased my question.
Its difficult to judge whether buying a Pascal Titan X would see an even better speed result than you have already indicated
Ive even considered buying the Pascal 5000 so that i have the best there is as far as speeding up my whole system/software and possibly giving Poser 11 superfly rendering a real boost in speed.
But given what these cards cost i would be very dissapointed if they did not give a lot better rendering speeds than i would get with my present system.
Hopefully more members will post a few comments
The fastest card that is sort of available is the Pascal Titan X, and sadly is only about 20% faster than GTX 1080
You can buy 2 Gaming 1080s for the price of a Pascal Titan X, and be faster.
You can also buy 2 GTX 1070s for the price of a good 1080, and render faster than either of the above in single mode.
The whole point to this is something you rarely see mentioned.
Video cards have gotten so fast with the jump to the 10 series chip, that you have trouble even feeding them over the bus in the PC.
This isn't a huge problem with a render engine, simply because most or all of the scene is in the video memory.
It is a big problem with games and that why you don't see the best improvement % claimed, that the GPU is capable of.
Render hardware comes down to price, versus core count.
A Pascal workstation card is not going to look real good for the money.
If you slap 4 GTX 1070 or 1080s onto a gaming Mobo, it will blow any Titan or workstation cards doors off while rendering.
And you spent far less money if you do it with 1070s....
Thanks for explaining the benefits of buying 2x1070s/2x1080s or even 4x1080s as
apposed to a single Titan x or P5000.
You obviously know what your talking about so think it best i look to buying 1 or more
While doing a bit of research on the above cards i came across a site that gave info
about a new Nvidia card 1080 ti to be shown at the CES hi tech hardware show in Las vagas
in early January 2017. [5th to the 8th i think]
Im shore you will know which convention im referring to.
The interesting aspect of the comments was that it is expected that the new 1080 ti will be as
fast as the present Pascal Titan x which on the face of it looks as if having 2x 1080 ti would
be an interesting and possibly an affordable option.
Think i will wait till after the January convention so i get a better idea of what this new
card is all about.
I've done several tests in Poser 11 with SuperFly,I've got two different cards GTX1080 and Titan X (Maxwell or if you want previous generation of Titan X)
Regarding which card is best to get as above @shvrdavid said GTX1070 or 1080 series is best right now,those cards have probably best performance for money(I will post my results bellow),I've been personally very surprised by performance of Titan X (Maxwell) in rendering
Here is render which I've done in Poser 11,its rendered with SF,V4 is simple my old morph,textures are MDD with EZSkin 3 applied,Hair is SAV,clothing is normal nothing extra,EZDome I've used and two area light(similar setup to clampshell),render has been postworked in Gimp(only curves has been edited),rendered up to 45 samples with Progressive rendering unticked
Render times per GPU at 45 samples and times are from Poser log:
Titan X has rendered that image at 29.1 minutes(1746 seconds)
GTX1080 has rendered that image in 8.70 minutes (525 seconds)
GTX1080 with Titan X has rendered that image in 4.60 seconds(281 seconds)
As you can see from those render times I would say,GTX1080 is best card to have,I don't have anymore GTX780 which I would love to test too and Titan X (Pascal) I've promised,but when I will have hard to say,I've been surprised by Titan X performance there as I expected,this card will be closer to GTX1080
As above @shvrdavid said you can buy two GTX1080 for price of one Titan X pascal and if its worth I would say,single card is OK,but if you are looking for best performance then I would go with two 1080 or 1070,you will gain mainly on faster render speeds,where Titan X is better is 12GB VRAM farmebuffer which would help if you do large scenes
I wouldn't go with workstation cards,have used in our work,we are changed our workstation Quadro to GTX series and if we are using workstation cards then we are went with AMD FirePro which we are using mainly in OpenCL and tese cards are cheaper than Quadro cards although they don't don't have CUDA cores,but we are using in house developed GPU renderer and now we are starting to use AMD own renderer which have pretty awesome results
Hope this helps
Thanks Jura for troubling to provide additional info that expands on what shvrdavid has already posted and also for the nicely rendered image.
Great that you have given some test results which is exactly what i was hoping someone would post.
Suprised how much slower the Titan x is against the 1080 but then again
the 1080/Titan x combination was twice as fast as the 1080 on its own.
Initially these test results messed with my head a bit :[
But now that ive got used to the idea thats 2 cards are better than one
Titanx or one Pascal 5000 i think ill be going the 2 card rout.
In my previous reply i mentioned about the new 1080ti which is expected to be shown at the CES convention early January.
There is lots of speculation on youtube about specifications but the most common comment is that the new 1080ti is virtually identical to the Pascal Titan x with slightly fewer cuda cores and a couple of other differences
My only concern here given that the new 1080ti is said to be very similar to the Titan x is wether it will perform poorly relatively speaking or will this new 1080ti actually out perform the standard 1080.
Would welcome any comment you care to make about this.
Here is a link to a site that shows the expected specification of the 1080ti compared to the Pascal Titan x and 1080.
You will need to scroll down a bit to see the chart.
Regarding your mentioning AMD FirePro and Open cl.
My present GPU is an AMD FirePro W5000 with Open cl v 2.00.
I heard a while ago that Poser are trying to get there software
to be compatable with Open cl but have not heard anything since.
One thing to keep in mind, is that all the Nividia cards coming out right now use standard memory.
GDDR5 and GDDR5x.
Because of this, the GPUs are sort of hitting a bandwidth wall, when HMB2 memory is sorted out that wont be as much of an issue. Cards using that style memory will have a speed advantage over the current cards, but for a price. Nvidia is going to pump out as many models as they can prior to using the new memory, they always do.
The difference between a standard GPU and a TI, is basically; Core width and price.
Yes there are some other minor differences.
I went with the 1070, because it was the cheapest at the time.
It is also faster than just about any Maxwell card, overclock it, and the story is over.
The card is factory overclocked, its small, it doesn't get hot (surprisingly) and it isn't that loud.
The small part is one of the nicer parts, it fits into just about any case. And has one power connector versus two.
It all comes down to how much you want to spend, on new stuff that is already outdated by design.
It's a game, and it can be a very expensive one to boot.
Dollar per Cuda core, is the best way to go.
Sure if you spend a lot more it will be slightly faster, and that's pretty expensive slightly....
Here is the 1070, happily running at over 2 gig core at 52c. Try that with a Maxwell....
I am still waiting for the new bios to raise the memory speed any higher.
Micron memory gets odd much over this, Samsung will usually hit 10,000 without much issue.
A lot of what your saying here chimes with what ive seen on youtube/comments.
Your reference to standard memory GDDR5 and GDDR5x being an issue
was something i did not understand to well when i saw it mentioned elsewere you have now clarified that.
Don,t think ill waiting for the newer memory to come out with next
generation Nvidia cards.
There always newer stuff on the horizon isn,t there and it is a never end waiting game.
Im prepared to wait for the new 1080ti to come out and hopefully it will
be significantly better than the present 1080 if so i may go for 2 of these cards.
Those render times surprised me too,I've expected closer render times,as I've done comparison in other renderers and render times are much closer or more in favor Titan X(DS 4.9 IRAY) ,but this has been issue with SuperFly/Cycles from beginning as looks like devs never optimized enough for newer GPU(Maxwell) and looks like Pascal is well optimized
I would go with two cards if budget allows,but still GTX1080 or Ti will be good choice for rendering,Titan X pascal no sure if I would choose that card,have owning previous generation of Titan X and is overpriced and those cards should be have lower price from start as £1000-£1500 for one GPU is way too much although is designed for enthusiast,but still those prices are way too high
Not sure if this card(Titan X Pascal or Ti) well perform in SuperFly or in other SW or renderer,I'm still not sure,as right now most of renderers are not optimized for new CUDA 8,but usually from what I've tried and done few tests like with Ti and Titan X in past,Ti is more orientated for OC'ers I would say,these cards are designed for OC if right cooling is applied,i would stay away from stock or FE(Founders Edition) cooling,best are ZOTAC,ASUS,PALIT and few others like MSI,usually aftermarket or AIB coolers will be released after few weeks of release of the NVIDIA stock cooling and mainly stock cooling is way too loud and runs way too hot there,due this stay away,only I would go with stock cooling if I want to go route of custom water loop
Difference between those two cards in actual rendering,Ti with OC applied can match speed of Titan X,this has been same in gaming where Ti always matched OC Titan X,Titan X has been always poorer OC'er in my view,my Titan X can run max 1450-1475MHz on core and have owned ZOTAC GTX980Ti which I could push to 1525MHz,but usually everything above 1500MHz is possible on that card
Regarding the OpenCL and Poser,I asked while ago if its possible to ask AMD if they can port their renderer (Radeon ProRender),this renderer is still based on OCL 1.2 and due this any card will or can run this renderer there like AMD or NVIDIA or even Intel IGPU,OCL is still only supports I think AMD and Intel,NVIDIA is still stuck at 1.2 and many devs don't want to build OCL 2.0 renderer for only one platform or one vendor,its not very cost effective way how to do it
GDDR5X is not bad memory,as right now only Micron is supplier of these RAM,many people can push their 5X beyond 5500Mhz(11000MHz effectively),but still we are still limited by NVIDIA BOOST 3.0 which downclocks the card when is reaching temps,we are simply need something like has been Maxwell BIOS tweaker and Pascal BIOS tweaker not sure when will be released
Regarding the newer memory,HBM2,this will be using Pascal P100 and AMD VEGA which will have next gen memory and that card should be same or similar fast as Titan X Pascal from reports
GDDR5X is still fast enough for NVIDIA although they are using HBM2 on the P100,but for hefty price and that card is more orientated for datacenters etc not for actual gamers and normal people like we are,I think maybe with Volta they will start to use HBM2,but I wouldn't be surprised if they will stick with GDDR5X for another or two generation
Yes I would agree,GTX1070 is and will be faster than Maxwell card,but this depends on lots of factors,friend have Palit GTX1070 and we are compared against the GTX 980Ti from ZOTAC and in many games both cards are tied,his GTX1070 has reaching the Ti in games when is OC and Ti is still fast enough for most of people
I'm running Titan X with Raijintek Morpheus II core,this GPU cooler will take 3-4 PCI slots,but cools my Titan X with OC at 1450-1475Mhz at 45-50C in rendering and never seen higher than 40-43C in rendering if card is running alone and with stock cooling I would need to run 65-75% fan profile and still I would hit easily 70-75C in rendering,OC no way I would OC this card to such levels like I'm right now running,tried EVGA Hybrid AIO and that allowed me to run 1400MHz as most at 60C in rendering and after my H100i V2 failed and then EVGA AIO failed went back to air cooling and I'm very happy
Did you updated BIOS on yours,these issues are well known and many manufacturers has released BIOS update and mainly update to fan curves etc,I would check if there is update,Gigabyte has been always slower with updates and many people do have own issues with this
OC VRAM not sure,but this doesn't help in rendering,have look tried to run with OC only core(currently I'm running on mine +100MHz on core/effectively at 2060Mhz and on VRAM +200MHz/5400/10800 and I know few guys can run their VRAM above that figure,usually at +500-650MHz) and there has been little difference and this in my view limited my OC of the core,with VRAM at stock speeds I can push my card close to 2100MHz,but still we would need Pascal BIOS tweaker which would help us to combat this bloody GPU Boost 3.0 on Pascal which downclocks core
Hope this helps and sorry for such long post
Don,t mind at all that you have replied at length as all info is usfull and helpfull in deciding what to buy.
Normally if you want to speed up your computer system with respect to GPUs
then all you have to do is buy the next model up from the one you have already and thats the end of it.
But if you want an Nvidia Cuda card then its a whole different ball game altogether.
You really have to know and understand the ins and outs as it would be so easy to spend a lot of money [As i nearly did] on a card that gives you a disapointing increase in rendering speed.
"I would go with two cards if budget allows,but still GTX1080 or Ti will be good choice for rendering"
Money is not really an issue but value for money is.
Im focused on the 1080ti [x2] as a prefered choice at the moment but need to wait to see what final spec is after January release and ill take it from there.
Regarding GPU cooler.
My system came with water cooling and im informed by system tech support that adding 1 or 2 extra gpu cards would not be an issue also my power unit 750w is more than enough to cope with extra cards
My bios is up to date as of about 3 months ago.
Regarding Radeon ProRender
By coincidence only yesterday on the Lightwave forum there was a link provided to view this new renderer. Checked it out and looks very impressive.
Will wait to see what the next version of Lightwave will be like as it will have a brand new renderer so may not need to have a third party renderer.
Regarding Open cl.
I was very suprised to learn that the new Pascal range have not had the open cl changed from v1.2 to v2.00 and remains the same as ever.
Nvidia would have there reasons for not updating Open cl which could be technical and or commercial and obviously they want to promote there own Cuda.
To my mind if Nvidia updated to Open cl 2.00 which may not be as popular as cuda they could through there advertising make great play on the fact that they have 2 options for increasing the rendering speed.
Even if Open cl. is prefered by relatively few professionals these few could easaly be persauded to buy into Cuda as they would have the benifits of both Cuda and Open cl.
Don,t know if what ive just said is valid or not but just seems Nvidia is missing on an opportunity here to gain more business by leaving Open cl at v1.2
You never know someone from Nvidia may see my suggestion and think its a good idea and send me a voucher to have any Cuda card i fancy for free.
No chance really just wishfull thinking on my part :].
If you buy good cards designed for hardcore gaming, etc, your not going to need to mess with the cooling at all, for the most part. I can get my 1070 to never hit 50c just by changing the fan profile, and it only has one fan on it. Thermal kicks in at 60c on my card, and it wont hit that rendering all day long.
Many of the 1080s will stay even lower. The drop in nm of these GPUs allowed for a huge drop in current used, thus less heat.
As far as a bios update for my 1070, there isn't one yet. You can adjust the fan curve yourself on the gigabyte oc cards.
What you can't adjust on them without hacking the bios, is the voltage to the memory. That is the issue with the micron memory, the bios does not increase the voltage as the speed goes up. You don't have to do that with the samsung memory, and they changed memory during production, IE after development....
Appreciate your adding a few more comments.
I have to say at this point that i never developed an interest in gaming.
Having said that when i started with an Amiga 500 computer many years ago
i did enjoy playing a game called "DXBall" which i became very addicted to. But my main interest was graphics and animation with Delux Paint software that came with the computer. Thats how i started and have never looked back.
Ps. Ive installed DXball on every computer ive had and its worked fine accept for the computer i have now.
DXball installed ok but playing it is a bit stuttery and noticable a bit slower then usual.
But ill keep it on my system as a sort of keep sake.
Have accasionally thought about buying a computer game.
Grand theft auto [i think its called] comes to mind as i
used to be a car mechanic and had an interest in motorbikes.
May spend some time checking this game and others out and
see if it leads any were.
matb last edited by matb
Actually I made an extremely detailed post on this subject less than a month ago showing precise comparisins between a Titan X, and various processors. The increase for me from an nVidia 560ti to a Titan X was dramatic BUT Superfly rendering is MUCH slower than Firefly, so you're kind of back to the old days of a day per frame for high quality, complex scenes with lots of reflective/transparent elements.
For me, the card memory was an important issue as SLI connecting two cheaper cards did not share memory.
There is lots of pros and cons when you buying the multiple cards and running them too,I've went via few GPU and have been fortunate too on one hand as many of my friends are able to borrow me multiple cards and I can test them and decide if its worth to get
We are still little bit in dark with Ti and I'm still not sure what performance will have,from earlier reports will be somewhere in Titan X "Pascal" performance and should have 10GB GDDR5X
Watercooling,you mean probably AIO or CLC? I've owned few of the kits from Corsair and all of them failed after while,mainly pump failed and if you are really looking for best CPU coolers then have look on EK Predator and Swiftech series which are best and can be upgraded if you want to cools yours GPU water and its very easy on those kits,which you can't do that on Corsair and if budget allows and case allows get EK Predator 360 or Swiftech H320 x2 Prestige with GPU water blocks for GPU of yours choice... Those cheap AIO or CLC are based on same pump(Asetek) which are just poor pumps and they're as first very noisy and those cheap kits use aluminium radiators with high FPI and you need to run there high static fans to have good performance there
If you looking on BeQuiet,Alphacool Eisbaer ,EK Predator or Swiftech H series all of them use copper radiators,good pumps with good flow and they tend to be lot quieter than Corsair or NZXT kits
Personally I'm running Noctua NH-D15 on CPU and comparing this one to H100i V2 which have run,I've gained 5-10C on temp front and can OC my i7-5820k to 4.5GHz,with H100i I've couldn't as my temps would reach 82C under stress and with NH-D15 I'm hitting only 72C and in rendering I'm hitting still just 61-63C,with H100i I would hit 72-75C
Yes ProRender they're planning the too for Lightwave,3DS MAX,Blender and SolidWorks and few others,just hope so Smith Micro will speak with AMD and we will have that renderer too
Nvidia always has been slow with adopting new tech,have look LuxRender devs don't want to create OCL 2.0 version of LuxRender until NVIDIA will be supporting the OCL 2.0
Nvidia I don't think they will in near future support OCL 2.0 as you are said only due one reason and this reason is CUDA,CUDA still don't support shared virtual memory which would be awesome if you two cards like 8GB and 8GB,as you know this means you will have 16GB worth of VRAM for render not like right now you have only 8GB
Sadly Titan X has been always sold with stock cooler which is loud and card always will run quite warm or rather hot and GTX1080 this depends,my GTX1080 usually during the rendering if its renders alone will sits in 62-65C range and if I render with both cards then temps on GTX1080 is in 70-72C and Titan X still hits just 45C during the rendering which is pretty awesome
If I would run one card all day long,I think I wouldn't hit those temps,with multiple cards you need to optimize airflow of the case and play with the most of the settings to achieve better temps,that's cons of the running the multiple SLI or CrossX cards I was planning to watercool both cards via custom loop,which is still on the cards but this depends on the next card too as I will be selling current Titan X and will be getting GTX1080 again
And temps that's depends on card and cooling,stock cooling known as FE,they run quite hot 83C and most of the gaming AIB version of the GTX1080 runs in region 68-82C,but most of the reviewers tested these cards in open air bench,not in actual cases,happens to my friend bought GTX1080 STRIX and his card from review should hit 68C,but actually his card hitting 76C,we are tried to improve his case airflow but we are failed miserably few times,we are tried in my case to run his card and temps has been slightly lower but not as in review and if he wants to have 68C then he needs to run 90% fan profile
I'm running custom fan curve on the both cards,when I render with both cards,my fans on GTX1080 will go beyond 80%(they sits at 85-88%) and in gaming this is the same for me
This BIOS tweaker would be helpful for people who wants push their cards bit further like me,adjust GPU BOOST 3.0 dynamic clocks and raise TDP plus few bits would do to my card
I've read yours post and I wanted to respond,but I forgot to add my view..
If you are reading my post above where I rendered same scene and same image in Poser 11 with SF like with Titan X and GTX1080,you will see GTX Titan X is lot slower than GTX1080,why is that,I would say Blender/Cycles devs never managed to optimize Maxwell generation of the cards for Cycles/SuperFly,in many cases I tried Titan X vs old GTX780,GTX780 has been similar fast as Titan X which shouldn't be possible at all,similar this apply for IRAY where Titan X is faster than GTX1080 which shouldn't be possible,everything depends on devs how they managed optimize and depends on skills of devs and Nvidia tools
Right now I'm thinking to sell GTX Titan X and get GTX1080,8GB is enough for most of my works although I will be limited again which is bad idea,but still 8GB is lot more than 6GB or 4GB VRAM of previous Maxwell cards
I only use GPU for rendering like in Poser and DS,but in 3DS MAX I use rather my CPU as rendering with GPU there I will need few more cards at least to get reasonable times
Hope this helps guys and again sorry for wall of the text
Here is my PC which I use mostly for rendering or PC in question
Hope this helps