Fuel Economy, Hypermiling, EcoModding News and Forum - EcoModder.com

EcoModder.com (https://ecomodder.com/forum/)

- Aerodynamics (https://ecomodder.com/forum/aerodynamics.html)

- - The time has come... (https://ecomodder.com/forum/showthread.php/time-has-come-41409.html)

The time has come...

Some may have noticed that I've had high hopes for CFD in Blender; based on it's support for OpenVDB and node-based editiing.
Aerodynamic Streamlines Simulator | Blender Tutorial
It's no surprise to me that it's CGMatter* that stepped up. It is surprising the particle system isn't used, but he does use obscure nodes like Named Attribute. You can save following the tutorial for $1.50.

Obviously, this is a first pass. It's based on detecting whether a point is inside or outside the target, so there's no consideration of stagnation or laminar flow or low pressure in the wake.

The emitter is smaller than the target to keep the computation manageable, but Blender can be dispersed over a render farm so the troposphere's the limit.

*Default Cube - Blender Short Film

What exactly were you really expecting? Low core count with a lack of parallel processing in a limited space machine. IF You could run this in the original PC, the data still would be computing after 24 years with probably another 25 to go.
IIRC,the big boys that do this for a living have more resources.

Not to say what you accomplish isn't outstanding.

Quote:

Low core count with a lack of parallel processing in a limited space machine. IF You could run this in the original PC...

Which is the original?

Quote:

The Kenbak-1, released in early 1971, is considered by the Computer History Museum to be the world's first personal computer. It was designed and invented by John Blankenbaker of Kenbak Corporation in 1970, and was first sold in early 1971. Unlike a modern personal computer, the Kenbak-1 was built of small-scale integrated circuits, and did not use a microprocessor. The system first sold for US$750. Only around 40 machines were ever built and sold. In 1973, production of the Kenbak-1 stopped as Kenbak Corporation folded.

With only 256 bytes of memory, an 8-bit word size, input and output restricted to lights and switches, and no apparent way to extend its power, the Kenbak-1 was most useful for learning the principles of programming but not capable of running application programs. Interestingly, 256 bytes of memory, 8 bit word size, and I/O limited to switches and lights on the front panel are also characteristics of the 1975 Altair 8800, whose fate was diametrically opposed to that of the Kenbak. The differentiating factor might have been the extensibility of the Altair, without which it was practically useless.

The software architecture isn't limited to a stand-alone PC.

Quote:

https://renderpool.net › blog › cloud-render-farm-for-blender
Top 6 Blender Render Farm Services for 2022
May 28, 2022 Render Pool is a service that lets you render Blender scenes up to 20 times faster on high-performance servers. It supports Blender Cycles, Eevee, V-Ray and Arnold, and offers free trials, low prices and customer feedback.

Fairly recently, Wouter Remmerie, of AirShaper, 'told' us that, to achieve a single CFD road vehicle aerodynamic 'solution', of high certainty, will require an investment of about $ 4,000 ( US ).
This is what you'd pay for 1-hour in a full-scale, rolling-road wind tunnel.
-----------------------------------------------------------------------------------------
The question you might answer for members and guests is:
1) What will be the cost to create the 'model' that you're going to have 3D-scanned to create the data-cloud that will be imported into the CFD program ?
-----------------------------------------------------------------------------------------
2) Once you've 'paid' the $ 4,000 for your 'first' answer, what will it cost to 'modify' your model, and go through the whole process the second time ?

Render Pool offers 'free trials, low prices and customer feedback.' OTOH:

Quote:

https://www.crowd-render.com › single-post › pi-farm-on-arm
Pi Farm on ARM - crowdrender
Farm Hardware. 8x Raspberry Pi 4B 4GB @ 68.44 each. 8x 128GB MicroSDXC @ 12.49 each. 8x PoE HAT @ 22.99 each. 1x Cudy GS1010P 10 Port PoE Switch @ 79.90. 1x Cloudlet CASE: Cluster Case @ 69.90. Distro. I ended up going with Ubuntu for the pis. Prior to this I was using Raspbian as that was

That's about $355 total. Pi 5 would cost another $40 or so and have 2-3 times the performance.

The photogrammetry can be done with a smart phone or standalone scanner. IDK. Frankly, I'm more interested in the architecture than the costs.

The support for Open VDB is significant. Gaussian Splat rendering is in development:

Quote:

https://blenderartists.org › t › 3d-gaussian-splatting-plugin-for-blender › 1490357
3D Gaussian Splatting Plugin for Blender - Blender Development ...
Oct 31, 2023Hello Blender Community, I'm a Blender enthusiast with a keen interest in enhancing Blender's capabilities by incorporating a 3D Gaussian Splatting (3DGS) plugin. Specifically, I'm seeking to achieve the quality of 3D Gaussian Splatting rendering found in Unreal Engine and Unity. My attempt

The Pi 5s have a new PCIe connector so you can add graphics cards:

Quote:

https://www.tomshardware.com › news › raspberry-pi-5-successfully-uses-external-graphics-card
Raspberry Pi 5 successfully uses external graphics cards
Nov 29, 2023To bridge this connection, Geerling used a flat flex connector that lets it operate using the PCIe slot on the Pi 5. The external graphics card also needed quite a bit of juice — so much so that

Some potential humor in putting a $400-$1700 card on an $80 PI.

I'm not doing anything with the Tesla P100 at the moment. Lost interest in mining when it stopped paying as well, and required periodic inspections to recover from crashes.

Oh, that Tesla.

Quote:

https://www.nvidia.com › en-us › data-center › tesla-p100
Tesla P100 Data Center Accelerator | NVIDIA
NVIDIA Tesla P100 is the world's first AI supercomputing data center GPU based on Pascal architecture. It delivers up to 21 teraFLOPS of FP16 performance, 16 GB of HBM2 memory, and NVLink interconnect for scaling across multiple GPUs.

Interesting. Is it installed in a machine, or lying on a shelf somewhere?

Quote:

Originally Posted by freebeard (Post 693507)

Oh, that
Interesting. Is it installed in a machine, or lying on a shelf somewhere?

Laying in a server, that is laying on my garage floor.

The GPU requires a rack mounted server and will not function in a workstation since it has no active cooling and relies on the server to force air through it.

You're looking at around 500 watts to run the whole setup. A warning will say there isn't sufficient power running on 120v, but it still works. 240v is ideal.

Quote:

Originally Posted by freebeard (Post 693509)

duckduckgo.com/?q=submerge+gpu+in+oil+for+cooling&ia=web

Do you want to be a CFD guru?

Too hard. I'll stick to easier things like philosophy.