HD Add On VS Normal Mapping Brawl!
outrider42
Posts: 3,679
This is a question some have been wondering about. For most Daz Original main characters, they offer HD Add On products separately from their bundles. So what do they do?
First of all, they replace the normal maps that the standard model comes with. If you click the character preset for any DO HD add on, it will load the character with no normal maps and at subdivision 3. The regular character will load at subD 2.
So, what does it matter? What are the benefits of buying this HD shape? Does it look different? And what about performance?
I don't think anyone has really discussed performance of these HD add ons. I have always assumed they were more resource intensive, but I never tested that aspect. Many people will say they see little difference between the standard version and the HD version, or that the differences are very subtle. Well yeah, that is by design. The normal maps are doing the job that the HD morph does, so you are getting that detail by texture rather than mesh, that is all. But...the performance may surprise you.
I do not have many HD add ons. I have tended to avoid them. I have Zelara HD, so I tested her.
I created a basic scene at 5000 pixels high of a full body shot, hopefully enough to see the detail. One spotlight, one distant light, default HDRI. No hair, this is testing the character, not hair. I used the help file to get the render times, and I MSI Afterburner to report VRAM usage during the render. I used one of Zelara's bundled poses.
I have two GPUs, both are 1080tis. The 2nd GPU is my monitor GPU, so it always uses more VRAM. I believe the non display GPU is the most reliable one for VRAM use since its only task is rendering Iray. But I will report both VRAM values.
The first test was Zelara's standard default preset, the one that includes normal maps. No HD morph subD 2. No changes or alterations.
Total rendering time: 6 M 12 S, VRAM: 3300 & 4115 MB
So that is our baseline. Next up is Zelara HD. The normal maps are gone, and her subD is at 3.
Total Rendering time: 3 M 55 S, VRAM: 3261 & 4089 MB
And BOOM goes the dynamite! Not only does Zelara HD render much faster, but she even uses a little less VRAM in the process. The surprising thing here is just how wide a gap the difference is, over 2 minutes off of a 6 minute render. Just think how much time that saves over the course of a much longer render. Zelara uses TIF for her normals, but the file sizes are still modest, ranging from 3 to 4 MB each. That may be why the VRAM dropped by a small amount, even though we increased subD to 3. It turns out that trying to resolve topology through normal texture maps is quite expensive for Iray.
I did some additional tests, to see how far this goes.
Zelara HD level 4
Total Render time: 4 M 20 S, VRAM: 3959 & 4788 MB
Now the VRAM is starting to climb. This is due to the subdivision level more than anything. I found that increasing subD adds to VRAM even if you have no HD morph applied. After all, it is still subdividing the mesh. However, the render time is still much lower than the time it took for the normal maps.
I did not do subD level 5, though I can. I had already tested level 5 on a bluejaunt character, and I found VRAM shoot up quite a bit and the render time did as well. This time, the render did take longer than the normal mapped version. But it took cranking up to level 5 to do this!
You get a little more speed by dropping subD to level 2 and even 1, but the gains are surprisingly small. In particular, going from 2 to 1 had almost no changes to speed and VRAM. 2 or 3 seem to be the sweet spot.
Then I went back in time to Genesis 3. I wanted to see if HD rendered faster on older models. Back then we did not have dual lobe specularity. We also did not have chromatic sss, and boy is chromatic sss an expensive option. I chose Mei Lin 7 HD, so a fairly early G3 release back in the days when we were all just starting to grasp this thing called Iray. This was also when Daz was trying to figure out eye materials, remember the sparkly eyes that just refused to resolve?
So I did Mei Lin at her standard default, with the same setup as Zelara at 5000 pixels.
Total render time: 2 M 45 S, VRAM: 3490 & 4475 MB
Wow what a difference a generation makes. And not using chromatic sss. She renders in less than half the time as standard Zelara in the exact same setup. So perhaps this time, the HD morph cannot possibly beat this? So no normals, HD morph and subD 3.
Total render time: 1 M 55 S, VRAM: 3449 & 4435
Holy smokes, Mei Lin blazed by Zelara! However her VRAM usage was actually higher. Mei Lin has normals that are in TIF which are 48 MB each. However, removing those had little impact on VRAM, so that is interesting. But the difference in render speed is pretty high for Mei Lin as well.
Pics of standard with normals vs HD level 3, you should be able to right click to see the full image if you wish to nitpick over them. There are differences, but they are quite minor.
I have replicated these results with other models, and Divamakeup has seen similar as well. So I think we can confirm this, HD Add Ons, when you do not use the normals, can actually save both VRAM and time over using normal maps on standard models. But this mainly holds true of the Daz Original main cast. Outside of them, the PAs may create normal maps to work together with HD shapes, meaning you cannot simply delete them like you can the DO line or you will lose detail.
What does this mean for the future? Well, maybe we don't need normal maps so much! Though including them for people who export is still a plus, I think it would be great if more characters could be created this way. Even though GPU technology is advancing fast, it still helps to be able to get an edge anyway you can. The gains in performance are huge, and many users would appreciate being able to shave off a large percentage of render time like this.
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
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
Interesting result, thanks for showing it.
As subD is effectively creating geometry (even if it's on the fly) it seems reasonably intuative that this would be easier for the rendering engine to do than to consider a section of surface, and modify its properties it in the light of one or more large referenced images. It's also probable the extra work from subD levels would go up in a squared manner as the subD level increases and eventually cross over the additional work level from the images. What you have done is show experimentally that the cross over point is actually higher up the scale than I was imagining. If asked I'd have said I'd be expecting it between subD 2 & 3.
It might be interesting to experiment to see if the normal map penalty is largely independent of normal map resolution or strongly dependent. on resolution. That might help people trying optimise rendering times.
Regards,
Richard.
This is interesting!
I build my stuff using mid poly resolutions and normal map baking assuming that the regular game dev workflow to save resources would apply. I'll do some tests with some of my environment props when i find the time
I miss the point of HD morphs for subtle features, where they shine are monsters/creatures where actual 3D geometry is needed and maybe popping veins, otherwise maps do a far better job IMO for pores, wrinkles etc
the only real use for wrinkles would be animated ones with expressions and that is much neglected
HD morphs do not replace normal maps. While both can add alot of detail, they are completely different in function.
Through morphs, there are better interactions with other objects such as clothing. HD Morphs also allow a more organic body shape than regular morphs. While on human figures the level of HD detail may not appear that great, it really becomes useful on extreme characters such as monsters.
Well, there's the option to use HD morphs from some more special figures - like Edie or Mrs Chow for females or Floyd for males - and apply them to more youngish looking figures to add some years to their look, without having to fiddle with the textures. It helps to get away from the photomodel look that so many figures - males and females alike - have.
Question, if I dial in HD body of one character to replace the non-HD body of another character, should I remvoe the normal map of the non-HD character now that he/she has a HD body morph?
A follow-up question: the product HD Body Shapes for G8 from Daz Originals does not come with textures while the HD character I owned has bump map (I assume it functiosn similarly as normal?). Does this mean I can remove normal/bump map if I replace the character body with the Daz Orginal HD Body Shapes dial but if I use body dial of the said HD character I still need to apply the bump/normal map of the source character?
I honestly use both normal maps and the HD add-on...most of the time. So for example with Babina 8, I would use her add-on and the normal. It's really down to your preference and what you want to see. Personally, the combination of HD add-on and normals adds just a bit more pop to me in terms of visual detail. But again, that's my preference.
I normally end up turning down bump maps...they're generally set entirely too high and that's what creates that odd clay or gumby-skin look.
I would like to use both HD add-on and normal but the difference in rendering time as shown by outrider42 is just too great...
Thanks for the tip on bump map. I hate the gumby-skin look but always afraid to mess with textures.
What is the difference, if any, between starting with the DO HD model or starting with the standard version and dialing in the HD?
The problem I always have with normal and bump maps, and so does everyone else, is that looking at a character's silhouette is going to be to uniform and not reflect the variations in the maps; some will be too small to see unless close up, others should be visible but aren't due to their inherent weekness.
I seem to recall that the presentation by Nvidia about Iray stated it doesn't mind geometry, it has little render cost; this is from a couple of years ago though.
If you start from the DO HD model you don't have to dial the morph or change the subdivision level, it does it automatically. That's the only difference.
Not necessarily. If it looks bad or doesn't give you the result you want, you can remove it, but there's no rule saying you have to do it.
Here's a comparison shot on one of my custom characters...she uses a skin mashup, but her textures are mostly Babina 8. I didn't let the renders finish, so there would probably be even more definition if I had. Left is no normals with Babina 8 HD add-on @ SubD 4. Right is with normals (Babina 8's normals). The left is definitely lacking something...so I'll for sure continue to use normals with an HD add-on. It may add to the render time, but for portraits it's totally worth it to me.
I did a simple NOT apple-to-apple test by dialing just the body of non-HD commerical character to a HD body dial of another character and also dial down the source normal to 0. I chose torso because the advantage of using HD over normal should be the greatest here. Obviously the shapes are not the same but if I have to make a choice the HD body maybe has a bit more details. The difference in rendering time is big:
HD dial @ subD 4, normal 0 = 3:47
non-HD @ subD 2, normal 1 = 5:20
Wow, subD4 without normal = 70% of rendering time of subD 2 with normal Until now I always thought going higher on subD will kill my rendering time.
Most normals that come with DO base characters are .tif files and are ginormous. It might be worth saving them off as .png to see if that saves on render time. You shouldn't lose much image fidelity going to .png, especially if you use a low compression. Whenever I modify a normal map (which is often) I save it off as a low-compressed .png and haven't had any issues. (Something else of note is that a lot of characters come with .png color/bump/specular maps. I always save these off as .jpg and haven't noticed a loss in visual fidelity...but it saves bigtime on file size.)
Saving them as png will loose information; they are enormous because of all the information they contain. How much it will matter (if at all) will vary between rendered images.
The normal files of the character I replaced with is already .png and is les than 5MB each (I dialed arms, torso and legs to 0). I would never have thought such a small file could have so much impact on rendering time. Such a diference from modding games. I own more characters from Renderosity and none of them are HD. I think I could get more details while reducing rendering time by dialing their bodies into HD and remove the normal.
Compared side by side at 100% in Photoshop? Yes. In an actual render? Not likely. I'm super anal when it comes to image fidelity and I've never noticed a difference.
On a separate note, I took a look at Babina 8's normals and they seem much smaller in file size than other normal .tifs that I remember...I wonder if they are now compressing their .tifs. (Yeah, they have to be because Victoria 7's normal .tifs are 48mb compared to Babaina 8's at ~3mb...so what I said above would only apply to uncompressed .tifs.)
I don't think it is having that big of an impact on the actual rendering time. I suspect the extra time is from the pre-processing of the maps before it can even start to render.
I'm glad this thread came up as I've only just started looking at HD morphs for DOs now that I have a few. My initial tests on Josephene was that there is zero difference on a final render SubD 2 normal vs 4 HD close up but I forgot about normals providing the details on non-HD. Now I want to go back and test rendering times and resource usage.
Yeah, portraits are one thing, and you could want a combination of HD, Normals, and Bump for close-ups. However, for mid-range to distant shots, you really don't see that little extra detail. So it may not be worth the added render times to use all three for mid-range or farther renders. (Beautiful character, btw! Your customization looks great!)
Don't you think that's cherry picking? It's still time before you get a finished render.
It does make some difference.it might mean the difference in render times dont scale in proportion - normals might not add x% to the render time, but x minutes disirregardless of everything else. it adds a minute to your 5 minute render - it also adds a minute to your 60 minute render. and so becomes negligable for complex renders. on the other hand, it could be the opposite. geometry takes more time to load in but once loaded in is faster (theres precedence for this one actually. Strand geometry hair frequently takes more time to load in the much more complex geometry, but during rendering having to compute transparency is more complex to calculate than plain geometry and cleans up slower)
That is what Daz has done for several years with their DO line. Pretty much ALL of the HD add on character presets will load without normal maps at all. I mean, that is literally replacing normal maps with the HD morph detail, how else can it be described? Of course these were not created as separate pieces, the normal is generally created from the HD in production. But from a customer stand point, the HD add on is sold as a separate product that requires the first...an upgrade...a replacement.
Normal maps were created as stop gap function in place of real geometry detail. When you can create geometry with enough detail, then you no longer need the normal map at all. This is not just a Daz Original thing, this topic has been discussed recently, like during the PS5 Unreal 5 Engine demo. The developers of the demo were quite explicit in pointing out that the environments did not use normal maps because their geometry was so great it was no longer needed. Normal maps would be redundant, and their presence would have been an unnecessary performance hit.
In this situation you basically doubling the intended effect. That is ok, but it is for your preference. Have you tried dialing the HD morph to 200% instead? If you can dial the morph to 200% while deleting the normals to get your desired effect, you should get a nice performance gain. I do not have Babina 8 so i cannot test it.
Some notes to make: Once you see any image in the render box, the precalculations are done. At that point it is all rendering time. And actually, high subD can take longer to load into VRAM...which makes the fact that the renders are faster even more impressive.
One thing is that I am using two 1080tis, not one, and I rendered my images at 5000 pixels in an effort to push times a bit longer (but not too long, I am not that patient). I do not believe many people in this thread have this amount of GPU power in their machines. Specifically, you would need at least a 2080ti to match this, or two good RTX GPUs running together. So when I say the render went from over 6 minutes to under 4 minutes, a savings of over 2 minutes, that is with over 7,000 CUDA cores. For nearly everybody else in this thread, both of those times would be much higher. If you are not seeing this, then there are plenty of easy answers why. There are plenty of other items you can have in a scene to shoot times up that would cause conflicting results. Also, I would wager most people are wearing clothes in their renders...not that I am judging anyone who perhaps is not...<.<. So obviously clothing covers most of the body except the face, and your choice of clothing and hair can wildly impact how fast or slow a render may be.
There are still more experiments to be done. I have Edie HD, and that model should be a great demonstration for Normal VS HD. Edie looks completely different without her normal maps, so this should be good. But I am exhausted after a long day and too tired to do anything, so this showdown will have to wait a bit. I might try my Zelara scene again with just one 1080ti to see how the results scale down. And I could also try a close up shot.
Just want to check if I am doing it right. Is dialing normal down from 1 to 0 in the Surface tab enough to remove normal map or there are more steps involved?
EDIT: Another question, if I dial in HD and then dial in other morphs, do I nulify the benefit of HD?
I have not tested that first one yet. I have been deleting the maps completely. My belief is you will get the speed boost, but the maps will still load in VRAM. But we'll see.
The HD morph will combine with any other morph just like all morphs do. You can even combine HD morphs if want. But keep in mind that the HD morph add ons are usually built specific to their characters, and may not quite look right with other characters. You can certainly try it and experiment with it, but your results may vary a lot depending on what they are. For example, adding Edie 8's HD add on to any young character will probably age them quite a bit, LOL.
As far as performance, the thing that effects it is the subdivision level. You can have any manner of morphs applied you want, HD, whatever, it the subD that determines the mesh resolution and effects the render times.
Why am I not asleep???? GAAAHHH
Sorry, another question, I have been hesistant to venture into HD fearing it will kill my GPU until now. If I use a lot of morphs that are not advertised as HD (EJ's body morphs seem to have a tons of details) on base G8F and dial up subD, do I get HD like details or I must use full body morphs that are labled as HD to forgo normal map?
I wonder if we will see HD clothings. As soon as I put cloths back onto my character the rendering time spiked up and I see the outfit pieces have normal and bump maps. Clothing in HD without normal probably could yield bigger gains but I wonder if it will kill dForce simulation.