The new system comes tomorrow

 At last! Nearly six weeks after putting in my order, the new system arrives. This will allow my current system to be sent to the "shop" to be fixed. They're not enthusiastic about putting in a new CPU, so we'll start with the fact that when I shut the thing down, it won't restart without much faff.

Once it returns, I can use it as a backup/rendering machine, so I hope to get a lot more animation done as I can leave it popping out frames while I work on the new machine.

Current benchmarks

I should be able to post the new machine's GPU benchmark results tomorrow. They should see a nice jump, although nothing like the one I saw when switching from the 2080Ti to the 4090.

Benchmarky Synthetica

A synthetic benchmark gives you the equivalent of a stopwatch with a readout in microseconds. Big numbers that give you an inflated sense of the difference between two time gaps that, in reality, you wouldn't be able to identify without a stopwatch. That said, rendering is so iterative that a 10-15% performance difference can significantly gain productivity in the long run. We should be getting a double that performance boost. 

Highs and lows

 Hip Jiggerty Jig! First sale!

First of many! Well, a sale is enough for now.

Burn-in completed on the new system. Horrah!

Edit: It's 8 p.m. as I write this blog article. Clearly, the machine didn't clear quality control. Maybe there was a snag, or maybe they're just swamped. They could have had a big backlog of RTX 5090 orders to get through. Either way, this will be done on Monday unless they need to exchange a component and are forced to retest.

Hitting that wall

It's happened again. The prospect of selling stuff makes me really critical about the quality of my work. This scout ship is all done. It looks okay. However, it's half a decade old and has lousy mesh issues. I did some nipping and tucking, possibly making things worse.

The hull plates are a mess. There are rough mesh elements. All the bad things.

Alternate product

 So, the Lightning Sword is going back for some changes. It won't be product #2 for Blender Market. Last night, I thought of a good candidate for a quick turnaround: My Traveller-related Type-S Scout Ship, although it will need a name change to ensure that I completely dodge the possibility of IP infringement.

Improved UV maps. Not perfect, but it's a lot better. I look forward to seeing how much better my textures end up. Before worrying about that, I'm cleaning up and rebuilding sections. The aft landing gears are new and the front gear will be reworked. Also, the vernier thrusters, some of the back-plate details, and those blocks on the end of the top surface look messy because they stick out too much.

I'm aiming for submission tomorrow or Monday at the latest.

New Machine

Well, it looks like it failed its 24-hour burn-in. I hope this isn't uncovering something caused by incompatibility. Or maybe they spotted the 5090 is missing ROPs (Render Output Units), which a small number of 50-series cards contain silicon that's missing these pipeline elements. No great rush, this is still a good week before they said they'd be building, so it's all good so long as they can sort it out. 

Blender Market Product #2 shǎndiàn Jiàn Review/Rejection

 Where are we with the shǎndiàn Jiàn/Sword of Lightning space vehicle?

Modelling completed. Lots of UV unwrapping, some of which is acceptable. I did learn a lot about UV unwrapping, but this was a big and tricky model to handle if you're inexperienced. 

So I created a bunch of materials, but most of the vehicle was to be baked from Sanctus Library metal shaders with its excellent edge wear and grime systems. However, this is a big lesson for me: a model of this size and complexity requires a lot of PBR maps. Sanctus is pretty easy to use and you can bake textures in a few minutes. However, the results of baked maps are very rough. The truth is, procedural materials are essentially grainless, or can be. When you put that into a map, you get pixelation unless you make the maps impossibly big. Even with restricted texture resolutions of 4k/2k/1k, the amount of VRAM swallowed up is too much. I could go through all the textures and half the resolution and use more compression, but they are already not looking great.

So I'm disappointed. I put a lot of work into the model, which I still like, but I've taken the development down a path that led to bad results. Bad results but good learning, so this isn't entirely negative.

It's not a disaster. There are some okay elements. It's not up to the quality I now demand from myself. I wrote about this not too long ago: The journey of improving your art should see you outgrow your previous standards. The trick is to increase your standards but not let them exceed your capabilities too far. Otherwise, you might walk away from the effort. I'm not walking away, but this model isn't ready to hit the market. I need to create something smaller and more easily delivered.

Other news

1) I got a job. A very modest job cleaning at my son's school. It's modest, but it might be perfect right now. It leaves my brainpower and most of my hours free to continue to pursue Blender as a career. I may not take to it, but I will do my best.

At the same time that I got a call confirming my provisional offer, other people had just jumped ahead in my expectations...

Oh, my giddy aunt. It's more than a week ahead of schedule. Go, SCAN.co.uk! I'm not ready! I was going to tweak the order, but it's too late now. I'll live with 64GB of system RAM; more would rarely be useful, and I specified faster RAM this time.

I need a name! This machine is Athena (named after my Orbital Elements game's historical AI character). The previous machine was ISAC (named after The Division game). It stands for Intelligent System Analytic Computer.

Keep the computing theme? Maybe. Blender-related? Nah, something cool again. I'm putting my thinking cap on. 

Blender School: UV Unwrapping Revision

 UV Unwrapping tips (Grant Abbot)

1. When adding seams, you can use the shortest path.

2. Multiple objects can share the same UV map. Useful for game assets/memory efficiency.

3. Remember to apply scale. Blender’s unwrap system doesn’t account for object scale. If objects are not consistently scaled, UV map scales will be wrong.

4. Use a generated texture when creating UV maps. ZenUV has one built in.

5. Average Island scale to even out the size of object parts on the UV map.

6. Dealing with UV stretching. Use UV Syn Selection (Arrows point at diagonals). Stretching can be problematic, but there is always a tension between removing stretching by adding more seams and keeping textures on the same island, where it’s easier to paint.

7. Select overlap. Complex meshes can contain areas where the UV faces overlap each other. The Select Overlap option allows you to view these areas.

8. Seams from islands. You can place seams on the existing UV islands by using Seams from islands.

4-Step Process (Josh Gambrell)

Automate Seam Placement

Start by letting Blender do the first round of seams.
Select edges by sharpness. 30-degree threshold

Four situations arise:

• Chamfer seams: A one-segment bevel (chamfer) has a greater-than-30-degree edge. You can remove one of the seams on the chamfer.
• Continuous sets of faces You only need one seam, so select and remove all others.
• Rings. When an unwrap creates a ring, the form has been crushed flat instead of cut and flattened. Add a seam, just one.
• Sneaky edge-markings. Sometimes the  threshold catches edges that don’t need seams. For example, an edge that’s just over 30 degrees. Remove the surplus seam.

If you follow these steps, you’ll nearly always have an optimal unwrap.

UV Mapping in 4 Steps (On Mars 3D https://www.youtube.com/watch?v=XleO7DBm1Us)

1. Apply Scale

2.  Project From View

3. Mark Seams

4. Unwrap

In overlays you can view UV Editing “Display Stetch” which shows stretching as a heat map.

Nightmare on Materials Street

 True to my word, I've lifted the Sword of Lighting project back up, and I swear that I will drag myself through the UV and material creation challenges without completely giving up 3D and becoming a goat farmer. 

In flux

In flux from the other side and further away

After completing some UV training (Blender School!), I understand the process much better. This hasn't immediately revolutionised my efforts, because a lot of it comes down to practice, but it is better. I now realise that unwrapping is iterrative and requires you to add more seams. It also can benefit from highlighting parts of the mesh and unwrapping using different strategies. The first effort on the model is the mission pod. I got a half-decent unwrap by highlighting the faces on each side, then unwrapping from the corresponding viewport orientation, like capturing blueprint projections, which are then organised onto the UV map.

Normal map for raised and lowered panels

Metalic map (rough) using Materialize free app

Colour map

I will add some colour flourishes and signs here and there when the overall materials are further along.

Back in the real world

I had a job interview yesterday. Try not to faint; it was a very modest role cleaning at my son's school. If I get it, I can continue to develop my Blender efforts to make an income. I'll clean all the horrid toilets to be able to do that. I should find out today or tomorrow whether I was successful.

I bought an add-on. Extreme PBR

 I'd seen a lot of recommendations about Extreme PBR. Everyone says how good it is. It's expensive, though. And I have other tools that do some of the same things. There's a sale on and 25% is a pretty good discount, so I picked it up.

Minute 5

Complicated! Very complicated. I immediately started to think that I will end up not using this because it's not intuitive, and there are a hundred right-in-your-face buttons. Naturally, you go to the documentation and the YouTube videos. This is what happened to Fluent Materializer: great tool, but too much work to get started.

Minute 20

Okay, the Texture Paint video finally revealed what's going on, and it's pretty neat. 

In default Blender, you can mask between separate materials, although more commonly, you'll use the Mix Shader to mask between different sources that you project into a single material slot. This quickly makes for a lot of nodes and a lot of spaghetti.


Extreme does all this on steroids. It can support up to four different materials within the same slot. Instead of using a separate mask for each layer (expensive!), it creates a mask that contains a colour for each material channel, Red, Green, Blue and Black. You then paint or remove those colours from the lask layer to expose the corresponding material. 

Imagine building this node tree by hand. <Cue Twilight Zone Music>

The quick paint effort on the test object took me about three minutes to set up and paint. Looking in the shader tree, you can see how this tasty sausage is made. It demonstrates precisely why node trees are not very friendly. Nodes give you near unlimited power to mix and blend different things together, but node trees very quickly become complex to the point where they stop any chance of flow unless you are a seasoned expert.

Onwards

I've only just scratched the surface. Extreme is capable of many amazing material feats, but I need to keep up with the learning, or I will never go beyond the first set of functions.

Missed deadline...

 The latest True-VFX competition's deadline was yesterday. Like a big fool, I'd not considered February a short month. In my very silly head, I had the entire weekend to knock a piece together. It wasn't to be! Although I've committed to submitting something late. 

These screenshots were the first couple of hours of work. Take note that I'm not building any new objects. All the asset objects were photo scanned in about ten minutes.

Tighter shot (50mm)

A more open shot

A wide open shot with realistic terrain height

Exaggerated terrain height

I wasn't keen on the flow of the hills in the background, although I do like the distant mountain peak. I think there will be a black pyramid behind the jackal statues.

New terrain

Better flows and a bigger road.

In this last shot, I'd appended the statues and figures from the previous scene but accidentally brought in the mountain as well. It pushed VRAM usage to 23GB and caused the render to take about an hour and a half instead of under 10 minutes. Sheesh!