Is Marmoset Toolbag Worth It

[Trinh Livingston]

Asa 2D/3D professional, I never use my graphics workstations for games.

I had migrated to Amd Ryzen (5990X) some time ago, not that I was disappointed with Intel, but the price/performance ratio was an argument.

Very happy with the processor... I was tempted by the AMD GPU because I was planning to buy the Nvidia 4090 RTX (currently on RTX 2080 Ti)

As soon as the new card was installed (7900 XTX 24 go) I installed the nice adrenalin interface... and did some personal bench test (rendering on 3D soft and Davinci resolve studio...). So far, nothing abnormal, even if I note a problem of reading under Davinci (4k without effect) a first baking test on substance (last update) : the machine froze immediately after the calculation of 'ambient occlusion' during the baking process...

I tried the same thing on Toolbag : same thing, but on Marmoset, it's just after the Baking process that the machine stopped... however, on small 3D objects not very complex, the baking is very fast. Also, the display and rendering is fluid and does the job.

Everything else seems to work well: Zbrush, the rendering of textures (ambient occlusion in this case) goes well, the display is fast... a speed gain overall... except that, when we ask for a precise task (here the Baking of a rather complex texture)... it seems that the card (or the drivers) are not up to the task....

A strange fact, I tried a "stress test" from the interface, this one, set to 60 sec, put the GPU in stress only for a few seconds... then, (while there was still 40 seconds of test, the "stress" was non-existent, no resumption, no alert, just a "rest" mode in the metrics)

So now I don't know what to do, I'm wasting time (I downgraded the drivers to version 22.12.2) I'm disappointed, tired and very angry that I don't have a hardware that should work and now I have to remove this little jewel to put back a card whose capabilities I already know (2080 Ti 11GB), but which never put me in such a professional mess

Hi bro, first want to tell, you too fast buy new videocard for work, need time for make new drivers and no reference card. I think temperature is your problem. you need check hotspot temperature in adrenalin you can make it. If temperature is ok, try proff driver 2022Q4 with blue interface.

About the 2022Q4 driver ? how to handle it : the installer does not recognize my card !

about the temperature, not really...

I didn't notice any increase in metrics on an ordinary benchmark (I even have excellent results) nor during rendering tests under UE5 or 3D max when this card pushed far away somes limits...

During the baking process... I think it's a phenomenon linked to the management of this kind of calculations (since it's a matter of throwing rays of an object in high resolution in order to create textures (normal map, ambient occlusion, curve, thickness, etc for the same object in low resolution...)

Then, concerning the purchase of hardware... yes, I already went through the "pro" hardware (at Nvidia... in fact, it's the first time I go through Amd for the GPU), I had several Quadro, at the time... and I was really disappointed by the "gain" brought compared to the financial cost.

From experience (I have been building my own workstations for at least 30 years... at this level, I would have killed someone the day I switched from win 3.1 to Win95 -> well, it's better now ;)) : I know that good graphics cards designed for "gaming", behave very well in graphics applications and especially for "indies" like me... and I'm not the only one in this case.

A lot of professionals will tell you... they prefer to invest in a good card for gaming, rather than having a "pro" card, worth 4 to 5 times the price and for not so optimal performances.

No, I think you can get by with cards and stations optimized for gaming, but I still hope to believe that it's a configuration problem, a poorly managed driver layer... (because I've seen some pretty good testimonials about the 7900 XTX under marmoset... but maybe not for the same reasons)

So to come back to the subject, no it's not about temperature : the fans are running correctly and when I do ordinary tests on benchmarks, I systematically check the metrics (moreover, I'm in a rather "cold" place and my workstation is amply optimized to manage high temperatures : water cooler, several chassis fans, optimization of hot/cold air flows)

In fact, I maintain what I said above, I think that the driver (or the card?) is not adapted to handle Ray-Tracing at very high load, as in the process of Baking (creation of texture from a high resolution model and projected on the UV of the map to improve the visualization of a model in low resolution)

While Nvidia (old model) takes care of it without much effort... it's a pity, while I was very happy with AMD CPUs, I'm still disappointed (until maybe an update)... especially since the benchmark (furmark in particular) shows a quite bluffing score...

Unless it's simply the Toolbag designers who didn't optimize their development around this new card (for Substance, it's almost the same... and the problem occurs especially for the ambient occlusion or the "thichness", where the ray tracing load is at its maximum!)

I don't use those programs. I tested Blender once and noticed that AMD drivers did not work well when GPU acceleration was used, but this was years ago. I did try both the gaming driver and the other one, can't recall what it was called.

Hi MADZyren... oh yes, 1350 watts PSU... I think that's enough. As I mentioned above, benchmarks (and also Zbrush, Maya, etc) behave remarkably well when it comes to "normal" rendering even if in frame by frame rendering, generally the GPU is the forgotten hardware... But in real time, it's a nice advantage.

Here (in this case marmoset Toolbag 4) it is clearly a problem related to the hardware handling of raytracing... but I think I found a temporary solution and it works (see above), even if I don't have the "hardware speed". By the way, this is clearly a problem related to the fact that many of my colleagues usually work on Nvidia (I did for years...) but, given the crisis and the prohibitive prices of these days, I wanted to opt for another solution... and developers (Toolbag, Substance...) clearly don't want to bother with AMD compatibility: it's the least of their worries; I have the impression that for them, AMD's gaming line doesn't really interest them, while on the other hand (Nividia Game GC), they're quick to fix any problem because they know that most of the independent creators are on Nvidia (2D/3D, editing). Well, that's my point of view... and it's a pity, because I have a lot of respect for AMD, and their effort to offer us "accessible" and performing hardware.

Taking better images will certainly help, taking materials at a distance with a small zoom lens and allowing for cropping would probably help even out the lighting a bit. Using a diffuse reflector may also help even out lighting. Don't use flash if at all possible and if you have to - bounce it. Keep the camera as parallel to the surface as possible, if the lighting isn't good take the images on another day.

@SrPx next Wednesday will be on the location of a construction I want to map the materials, to develop a version of his simulation with maximum similarity (there will be also other versions more relaxed in term of realism).

Based on current PBR workflow a better way will be use the photos just as reference and use product such as Mixer to achieve the result "mixing" other available Megascan materials and decals, or better use the actual photos and products such as Materialize to create the final material, as from the beginning of the thread.

It seems to me that what you mostly need to stick to is to the "measured data". Certain sRGB values for one workflow, IOR values for the other workflow , depends on what you choose.

In that way, starting from that base you could work later more freely, but still I'd see a need of some tool to check the "validity" of the final resulting maps....

A tool like Substance, or, 3D Coat (though not sure how much of the shading part does it accomplish. I love its retopology options, and seems cool for normal maps) is almost a must for serious workflows. If not , use utilities here and there, and tables with real physical values.

For just prototyping, you could just do it the dirty way, take very hi res photos, then use Awesomebump, Crazybump, Xnormal, Marmoset tool bag, Substance or whatever, extracting from the photos and generating sth at least looking nice....

My take for being super pragmatic : just purchase some PBR materials from the most accurate source (careful, there's crappy sites) if is only some few materials (ie, some metals, some woods, glass, plastics, etc) and load them in your engine, focus in the prototype itself, as these materials, are not really expensive IF only using a few.

For a PBR based workflow, well... I have not -thankfully- needed to go deeply that way, as when I started working (back) as a designer, left games, no PBR around, yet. And as mentioned, for indy/casual/small company games, well... if you are the author and just you, I'd do what would give most fun and best selling point (and not necessarily that needs a PBR workflow) .But typically when doing gigs to other people, you are asked for PBR. The little I've done is just using pre-made materials : Even coming already with its diffuse map, its normal and/or height map, the roughness map, the metallic map, and depending on the case, the ambient occlusion map (AO). That in case of using one of the 2 workflows (metallic/roughness or specular/glossiness)), but for me has been the most common, rather than specular/glossiness. Anyway, I've done very little lately (though a lot of 3D, in general). Substance tools from Allegorithmic do support both workflows, and more importantly, do have tools to check the validity of what you are building. That is, besides starting from the needed measured data (ie, sRGB values for materials for the difuse map, depending of the type of "thing" of real life, and, if I'm not wrong, the IOR reflection/refraction values for the specular/glossiness workflow, it ALSO checks the validity of the material, while in the process. I don't have these tools, so, dunno.

3a8082e126