Graphics Card Addendum 1: DAZ Studio Display Optimization and Preview Improvements



Since putting up the graphics card upgrade article a few things surfaced. The first came to light on the DAZ Studio forum at DAZ 3D. The latest beta includes some chunky OpenGL, and therefore, graphics card related updates.

DAZ Studio 4.6x Introducing Texture Shaded 2.0

In their latest Beta, DAZ 3D have added a new preview mode to their viewport currently referred to as “Texture Shaded 2.0”. This new preview DrawStyle gives a much more accurate preview of bump, normal, and specular effects with DAZ Studio’s default distant, spot, and point lights. At this point the new preview mode is very slow when compared to the older “texture shaded”, but produces a much better preview. To put the new system to the test I loaded one of my heavier scenes (600k polygons), which I also used to test OpenGL response in the previous article. Graphics card used in both articles is a tired Radeon HD 5770.

Texture Shaded 2.0

DS’s new DrawStyle is slow but pretty

DAZ Studio Texture Shaded preview

Texture Shaded – lightening fast with Display Optimization

Using Texture Shaded 2 in this scene caused the viewport to all but grind to a snail’s pace. In this instance the difference in quality between the two preview modes is very noticeable.

The current state of Texture Shaded 2.0 is unlikely to be a reflection of performance in future releases. At the moment it appears to put extra load on the CPU when idle, but seems to have similar usage to the previous texture shade version while viewport manipulations are occurring. The new preview mode seems to move less of the load on to the GPU than the previous version.

 Display Optimization – It Makes A Difference!

The second addition I would make to the previous article is that since then I have discovered the “Display Optimization” in the Interface preferences. With this turned to “Best” the Texture Shaded preview was much, much quicker. In fact, this scene ran almost as smoothly as a simple scene with a single Genesis figure. What display optimisation does, according the documentation, is load more of the scene’s geometry into the graphics card’s memory.

Embarrassingly, this somewhat changes my findings from the last article, but perhaps only slightly. With just about any graphics card of the last 4 years or so, with the most current drivers installed, it should be able to get good viewport response with quite large scenes. With larger scenes, particularly those with lots of textures with the Texture Resources (also in interface preferences) turned all the way up, having a graphics card with a gig or more of RAM will make a huge difference.

DAZ Studio 4 texture shaded 2.0

Genesis 2 default with current Texture Shaded preview.

DAZ Studio Texture Shaded 2.0

Same scene with Beta Texture Shaded 2.0 preview

To push my card as far as it would go I loaded a jumble of scenes which resulted in 15.5 million polygons (+ subd), and 108 megabytes of textures (much less than I expected). The mish-mash produced more texture information than the graphics card could store in its own RAM, almost 2 and a half times over. There was a good five minutes or more of unresponsiveness – got a low memory warning – but after everything loaded in, the viewport went back to running almost as smoothly as with the original 600K polygon scene.

With Display Optimization turned off and Texture Resources all the way down I was using 429 MB of the card’s RAM. With both of these fully enabled, 945 MBs of the card’s 1024 MBs of RAM was used with 1441 MBs pushed back onto system RAM. With optimization on best and texture quality turned down. the usage was 507 MBs of dedicated RAM and 98 MBs of system RAM.

We see that, while Display Optimization lumps very little on the graphics card, it can make a huge difference. Getting the best possible Texture Shaded preview, turning texture quality all the way up, adds a huge load to VRAM, even when using quite conservative textures – 213 of 290 were under 500 KB.

Turning Texture Shaded 2.0 on with full optimization and texture quality set to second highest caused DAZ Studio to grind to a halt as textures endlessly washed in and out of the VRAM and back into the system. With more system RAM and a graphics card with more memory bandwidth would probably give a quicker turn around, but the end result for most older or entry level cards, working on a scene this size and at this level of preview quality, would probably be beyond unworkable in terms of viewport interaction, posing etc. When texture quality was turned down to lowest settings viewport response was still painfully slow and unworkable.

When it comes to working at the highest possible preview settings in DAZ Studio it seems that newer mid/higher level consumer cards do definitely have a place. For those happy to work with lower quality texture settings on large scenes, older cards and upper entry level to mid-level cards are good enough. With further refinement of the Texture Shaded 2.0 preview it is likely that its usage, even on bigger scenes, will be less problematic.

Limitations

Not all cards, specifically ancient cards that are barely likely to meet the minimum requirements for DAZ Studio 4.6, will support Display Optimization. Some cards will offer varying degrees of compatibility, depending on drivers, VRAM capacity and level of geometry/texture detail. I encountered some graphical glitches when posing characters when VRAM and GPU were fully loaded.

Of course, RAM is not the only consideration with this feature, GPU loaded tended to be high when performing most viewport tasks such as posing, so bigger is better. Even my HD 5770 chomped along happily in instances with a good deal of geometry, but choked up under the weight of copious textures at high texture quality settings. Something like a Radeon HD 7850 with 2 GB GDDR5 or a GeForce GTX 660 would be more than sufficient. A HD 7770 would be similar to my current card in terms of a simple comparison, but has a newer GPU architecture and faster clock speed. The 7770 is compact (don’t need huge tower to fit the bugger in), works happily on a 450 w power supply (depending on the rest of your system) and draws everything it needs direct from the PCIe port, without need for additional cables.

GPU rendering AMD OpenCL Compile/Kernel Issues

The third item is not actually related to DAZ Studio, unless you also happen to be a LuxRender user. In that case there is an indirect link. This issue is something I had some awareness of, but due to the technical nature of the issue, I was not able to wrap my head around it and didn’t feel comfortable writing about. It turns out the issue can rather simply be put into laymen’s terms. To quote directly from my informant, again from the DS forum at D3D, “There is a bug in the OpenCL compiler of the catalyst driver that makes the kernels too big (using much more memory than nVidia compiler).”

What this means is that in certain situations OpenCL will chew up all your system memory and crash. As far as I can tell this will have little or no effect on current LuxRender users, but this problem could put the brakes on LuxRender GPU development, and has already seen Blender Cycle’s implementation of OpenCL put on the back burner. Thankfully AMD seems to be quite keen to fix the issue, but when a fix will be available is anyone’s guess.

Final Thoughts

poser ds preview comparison

DAZ Studio Texture Shaded 2.0 VS Poser Pro 2014 preview

With Texture Shaded 2.0 DAZ 3D seem to be making good advances in catching up to Poser in terms of real-time OpenGL viewport preview. At the moment it is slow, and unusable for real-time in all but the smallest scenes (at least with texture quality maxed out), but it will undoubtedly see a good deal of optimisation. With improved shadows in the preview, DAZ would be getting very close to the superb quality of the current Poser offering. At this point Poser’s much better quality to speed preview is undoubtedly the winner, and if this is anything to go by, then anything more than an entry or mid-range card would be over kill for most users.


Best Graphics Card: DAZ Studio, Poser, LuxRender, Octane

*See this article for updated graphics card suggestions for Iray and Octane.

On The Graphics Card Upgrade Trail

NVIDIA GeForce GTX 780
My graphics card, a much neglected AMD/ATI Radeon HD 5770, recently alerted me that a replacement might be an imminent requirement. A total flush of over 4 years of accumulated driver leftovers and a fresh install of the latest batch seems to have calmed the issue somewhat, though not entirely. While I was digging up the info to expand the life of my ailing card I had a good look at what’s out there, and now I feel like I’ve opened something akin to Pandora’s Box.

Generally I’m happy with how my system handles DAZ Studio 4, Poser Pro 2014, and LightWave 11, but once you’ve been bitten by with the upgrade bug… Anyway, in this article we will focus on what sort of graphics card is best for recent versions of DAZ Studio and Poser. We’ll also take a look at GPU based rendering and what works best with Octane and LuxRender.

Getting The Full Story Is A Technical Business

I will make the disclaimer up front that ensuring the absolute best performance for your dollar is an incredibly time consuming task. Even determining the best specific variant of a particular model requires a good deal of technical knowledge, and goes far beyond what I am able to cover here, which is intended as a general guide.

It is enough to say that just reading details such as RAM and GPU clock speeds, memory bandwidth etc is not always the full story. In some cases card specs can be down-right misleading. If getting the absolute best value is a high priority than benchmarks are a very good tool.Throughout this article I will make reference to several benchmark comparisons performed by Tom’s Hardware.

If you are interested in getting into the nitty-gritty then this article at EnthusiastPC will be invaluable – clarifying many technical elements in a concise and easy to read manner. This will help you cut through the marketing and give you a much better understanding of just what you’re looking at when examining card specs.

Reference Card

AMD logo
I will be using my own current graphics card as a reference point throughout. While it handles these programs well enough It is getting long in the tooth and probably won’t be found on the shelves in many good computer stores (bricks and mortar or online). With this in mind it should be fairly comparable to many other cards out there under upgrade scrutiny. So, to make things as clear as possible it is helpful for me to provide some basic specs.

• The card: Radeon HD 5770
• Released: Oct 2009 (Before AMD bought ATI)
• PCIe 2×16
• 1GB GDDR5 RAM
• Core clock speed: 850 MHz
• Memory clock:1200 MHz

Additionally I should point out that my card can no longer reach full specs (at 100% load) without crashing the display driver. This could be caused by long term moderate overheating issues – good reason to keep an eye on things and perform regular maintenance checks.

It may also be of interest that I’m running an Intel i7 860 with 8 GB of DDR3 1333 MHz RAM.

Contents

DAZ Studio
Poser
Professional/workstation graphics cards</>
Octane
LuxRender
Benchmarking with LuxMark
Conclusions

So, Which Is The Best Graphics Card For DAZ Studio And Poser?

A lot of new DAZ Studio and Poser users assume that a top shelf graphics card will speed up their renders. This is a mistake, and an easy one to make if technical information is not your cup of tea. A good graphics card still has benefits for every 3D app I can think of. Some programs like 3DS Max will make more use of advanced OpenGL, CUDA, and OpenCL features. The biggest difference you will see in DS and Poser between cheaper and more expensive cards is in the OpenGL preview (that thing you do 99% of your 3D work in). More GPU RAM and processing power will result in faster viewport response, and an increase in the size and complexity of scenes one can reasonably work with.

The next rule of thumb that inexperienced upgraders come up against is that Nvidia is the best and only real choice when it comes to performance. To an extent this is true. Nvidia cards tend to give better overall performance and have a history of better driver support, but there are instances where Nvidia find themselves relegated by their much smaller competitor, AMD. Chiefly AMD cards beat Nvidia when it comes to OpenCL performance, but I’m starting to get off track. Let’s focus on OpenGL.

OpenGL Performance

DAZ Studio

So, what cards are best for OpenGL? Well, just about any off the shelf consumer grade card will give you very good performance (AMD or NVIDIA). For programs like DAZ Studio, which does not make use of more recent OpenGL features (2.2 being all the way back from the mid noughties), even many older graphics cards that have been off the market for years like the Radeon HD 2600 should still be enough to get running, but their legacy drivers might not be stable on newer version of Windows. If nothing else old cards like these meet DS’s minimum system specs. You can still find these old cards in unopened boxes on sites like eBay selling very cheaply, but they are often the same price or more expensive than contemporary entry level cards that have higher performance architecture and components.


Of course, systems with these older cards will not be able to preview larger scenes with textures turned on without considerable slowdown, if at all. As a general indication, my system experienced moderate viewport slowdown with a scene of 600 000 polygons. For further clarification that’s the same as loading in 28 Genesis 2 figures. In general this is more than enough for me and how I use DAZ Studio. Checking Catalyst Control Center (basic tool for monitoring AMD graphics card performance) showed that the scene was definitely putting my card to the test. Note that a static scene will add little or no load to your card. It is only when moving the camera, posing, and moving objects that the GPU comes into usage.

While I say the slowdown was moderate, fine posing at that level would not be much problem for someone with reasonable patience. For these situations turning subdivision to 0 and switching the view to smooth shaded or a wire frame will improve performance somewhat.

A bigger, better and newer card has very little to offer, other than viewport speed, but comparing that performance between an entry level card like the Diamond Multimedia HD 6570, MSI HD 7750, or EVGA GeForce GTX 650, and a mid-range card like PowerColour HD 7850, Sapphire Vapour-X HD 7950, or a GTX 660, a mid-range card like a will win out noticeably as more geometry is added to a scene. So, entry level for portraits is probably fine, but populated sprawling sci-fi city scapes or high detail natural landscapes will require lots of compositing different images and/or merging pre posed scenes (fingers crossed – no crashing).

Poser

Poser doesn’t give too much away about OpenGL version or requirements. What is clear is that Poser makes far more extensive use of OpenGL than DAZ Studio does. For a start DS doesn’t have the nice real-time shadows that Poser’s preview has. What the P 10/2014 system requirements suggest is a “recent NVIDIA GeForce OR ATI Radeon required for advanced real-time preview features”. I take this to mean just about anything on the market will do to get started (likes DAZ Studio), but performance will vary.

Even a 40 dollar GeForce GT 610 supports OpenGL 4.2 (current version is 4.4), which is probably a few releases above Poser’s requirements. Even cards that are released with support for older versions of OpenGL will usually continue to have their compatibility with newer versions updated as long as the manufacturer continues to support the card. I’ve heard of people using much older and/or less powerful cards, though performance can get slow with full preview options turned on.

So what it really comes down to with both DAZ Studio and Poser 9/10/2012/2014 is that almost any card you can go into a store and buy, or order online will be enough. A bigger and better card, all the way up to high-mid (HD 7970, GTX 780) to high end cards (GTX 690, HD 7990) will obviously give you better results, such as allowing for smother preview of larger, more complex scenes, but that extra performance comes at an increasing premium.

Professional Graphics Cards

I won’t spend too much time on professional or workstation GPUs. For most users of DAZ Studio and Poser, these are serious overkill. Those that spend a lot of time in programs like Maya, 3DS Max, LightWave, etc will receive much more benefit from better graphics cards, and considering a professional cards from AMD’s FirePro and NVIDIA’s Quadro line of GPUs should be on the agenda. These programs have many workarounds for dealing with large volumes of geometry, such as having geometry/objects past a certain distance display as bounding boxes, but these methods may not always be desirable.

Quadro and FirePro cards have been shown to deliver vastly improved frame rate improvements over consumer cards when working with large volumes of geometry in production applications. For many users these cards are above and beyond budget. Mid-range work station cards can easily approach the price of upper end consumer cards. Another drawback for the many users is that they will not perform anywhere near as well as similarly spec consumer cards when it comes to gaming. So, if your work or hobby render/animation/modelling computer is also your gaming rig, then pro cards might not be a good idea.

The often touted difference in performance between consumer and workstation cards is in driver development, such as OpenGL optimisation and specialised integration with certain production applications like 3DS Max, Maya, AutoCAD, etc etc. There are other stated reasons, such as firmware design and the use of more precise error-correcting code RAM (ECC).

In short, in production level OpenGL environments pro cards are king. In this regard there are benefits for DAZ Studio and Poser users to have a FirePro or Quadro, but there is such a thing as overkill.

So with this interesting bit of information in the bag I began making eyes at AMD’s FirePro W5000 which performed admirably in LightWave. Even this relatively humble pro card was pushing what I was willing to pay, especially as till very recently it was a component I was quite happy to ignore. Then there’s a new development that turns this completely on its head.

GPU Assisted Rendering

Outside occasional gaming binges, my only other heavy use of GPU technologies is when I dabble with LuxRender. As I looked at more benchmarks I discovered a disturbing (though fortuitous) quirk, pro cards don’t perform too well in GPU rendering applications, both OpenCL and CUDA based. There are some CUDA based render engines out there specifically optimised for Quadro cards and some others where pro cards sit neck-and-neck with consumer cards.

If you have any interest in GPU based or assisted rendering then getting the best possible consumer card you can afford is a very good investment. But it’s still not that easy.

CUDA or OpenCL – Not Both Ways

Octane

If you want to work with CUDA based renderers like Octane, then getting yourself a higher end NVIDIA GTX card or two is the way to go. Go Titan if you’ve got the money. The official NVIDIA site has a list of cards that are CUDA compatible. Conveniently they are all ranked on CUDA compute compatibility.

Anything above a GTX 650 is rated as 3 + compute compatibility, while Octane requires at least 2 for full compatibility (GTX 465 +). More CUDA cores and GPU RAM are high on the Octane list of priorities. In this regard, getting the best out of Octane requires a big investment. A EVGA GTX 780 3GB will set you back around $660, while a EVGA GTX650Ti Boost 2GB
is about $170.

While smaller cards like the GTX 580 will chug along reasonably well for most uses it is seriously limited by its RAM. A simple enough fix, one might assume, would be to buy two GTX 580s. Seems reasonable, but with Octane the scene needs to be loaded into both GPUs, so while you will get quicker renders you are still faced with limitations on scene geometry and max possible textures. If multiple cards is under consideration keep in mind that having cards of equal RAM is the most efficient choice, as the card with the larger amount of RAM will have to defer to the smaller card. The biggest possible scenes call for the Titan’s 6GBs of GDDR5 RAM. In terms of pure speed, you can’t beat multiple GTX 770s, or 780s.

LuxRender

Right from the start I should point out that getting the biggest and most expensive GPUs will not give you a massive advantage in hybrid (CPU+GPU), at least not in most cases. GPU only versions of LuxRender such as SLG might be very fast and benefit greatly from high-end graphics cards, but they are somewhat experimental and are a long way from including all of CPU LuxRender’s features.

Lux’s developers warn users that GPU based Lux is not ready for production, though some do use it. I’ve said it many times, but I’ll say it again LuxRender really is for hardcore render nerds. Plugins like Luxus and Reality do make Lux more accessible, but then getting the most out of SLG (in particular) still requires a lot of time on the LuxRender Wiki, forums and experimentation.

With this qualification out of the way let’s move on. If you’re using LuxRender or another OpenCL based render engine then give NVIDIA a wide berth. CUDA is NVIDIA’s proprietary baby and they are terribly biased in its favour to the extent of having underdeveloped their OpenCL technology.

The absolute LuxRender/OpenCL card of choice right now is the Radeon HD 7990 (on special at time of writing), which is basically two 7970s squished together with precision. The next best option is to get two HD 7970s CrossFired. Benchmarks show that the 7990 slightly outperforms the two HD 7970, but the latter choice can save you a few pennies (depending on varriant – linked one is currently cheapest). Here’s some very interesting LuxMark results that include high-end consumer cards and a range of pro cards.

There is one rather large anomaly I noticed while perusing the LuxMark results page. GTX cards have been owning (that’s what the kids say, right?) the ATI cards. This is despite all the benchmarks from sites like Tom’s Hardware and AnAndTech constantly showing HD cards coming out on top, and often followed by ATI’s FirePro cards. So, what gives? Well I’ve heard rumour that some clever people out there are modifying their NVIDA drivers to perform much better with OpenCL. For the most part this is probably well beyond the scope of the average hobbyist, and might run into warranty voiding issues.

Cards from the HD 6XXX series begin to lose ground to higher tier GeForce GTX cards like the 770, 680, 690, 580, Titan (beefcake of the GeForce range), so if you like both CUDA and OpenCL based engines then a newer and higher grade GeFore GTX card is definitely worth considering.

Benchmarking with LuxMark

Out of curiosity I decided to benchmark my own somewhat degraded card. Not surprisingly the results were anaemic, at least when putting up my mere 271 (GPU only) against low to high 400s registered for the same card and test. My old i7 860 (entry level i7 at the time) put’s my card to shame in its own right with a score of 367, which sat smack in the middle of the other results for the same processor and test (Salsa). When their powers were combined the duo scored a mighty (hur, hur) 587. So, that upgrade is still looking quite attractive when a single HD 7990 can eat my results over seven times.

If you want to test your own system out, and if you’re a Lux user, I strongly suggest you do, you can pick up LuxMark (free of course), from the LuxMark section of LuxRender’s wiki.

But, WHICH card should I get?

Well, the resounding conclusion is the old hardware cliché, depends on what you plan to do with your system. A good graphics card will get you better performance with OpenGL viewports, but unless you’re working with expensive modelling and animation packages midrange consumer cards are more than enough.

If you live in programs like Maya, 3DS Max, Mudbox, LightWave etc, chances are you either already have a pro card, or considered making the investment. For these programs the difference in OpenGL performance between consumer and pro cards can be night and day. BUT pro cards are definitely not the best choice if you wish to have a general purpose machine capable of playing the latest games at appreciable speeds.

The big push for GPUs in the 3D/CGI, especially for hobbyists, is with CUDA and OpenCL based renderers, and here lies one of the most important and costly hardware and software decisions you will probably make. A good CUDA based card (GTX 680 or GTX 770) and an Octane licence with a plugin to Poser, DAZ etc will set you back anything from US $800ish up to $1300+ (Titan + Octane), or much, much more for one or more top shelf cards.

Alternatively you can get going with GPU and CPU/GPU LuxRender for as little as a AMD HD 5XXX (almost nothing – $200ish). Something beefier and more recent is highly suggested. HD 7990s are the way to go, but depending on where you buy from and what brand you select they can cost anywhere from $800 up to $1600. The 7970, an admirable performer, can cost between up to $500 or 600 for Sapphire’s delicious Vapor-X HD 7970. Remember that as you go down the hierarchy of HD cards, higher level GeForce cards begin to become more competitive for OpenCL, though tend to cost more.

There are so many factors to take into consideration when upgrading, such as board compatibility and power supply (PSU) requirements. Luckily these things are easy to nut out by looking at cards specs (check at manufacturer’s website to be certain). A follow up article is on the agenda. I hope I’ve managed to dish up some good info to help you select the graphics card that will work best for you.