What kind of computer to buy?

[Jose]

hello everyone,

I’m newphie using odm.

I want use webodm for volumes measure in quarries.

What are your opinion about if it’s better a laptop with amd graphics (rx 5500m or rx 6600m) or nvidia graphics (rtx 3050ti, rtx 3060)?

The cpus between which I prefer, are: Ryzen 5 5600h and Ryzen 7 5800h.

My idea is install dual boot with win10 pro and manjaro and in the future use any cloud computing services like a Aws, azure, Google cloud, etc.

Thanks everybody.

[APOS80]

A Nvidia with >8gb of ram, a cpu with lots of cores and as much ram as you can get.

Buy a desktop!

[APOS80]

QGis has a great volume plug-in that can take two DSM.

[Jose]

Thanks for your answer.

My problem is that I travel to several quarries in Spain and I need to get work done when I stay in hotels, for this reason it is the one I am looking for a laptop.

I’m between hp 16-c0032ns (ryzen 7 5800h, 16gb ram, Rx 6600m, 512gb SSD m.2) and Lenovo Legion 5 15ACH6H (Ryzen 7 5800h, 16gb ram, rtx 3060, 1tb SSD m.2).

[Gordon]

You wont be able to cover very large areas with very high detail (ultra) with only 16GB of RAM, but it might be ok if they are small quarries.

[Jose]

Sorry for not answering, work is being hard and I forgot. The largest quarry that we have to fly is approximately around 2000 photos.

What your opinion about the recommended requirements, about ram an graphic card?

[shiva]

2000 photos is a lot for using a laptop.

GPU:
I am doing some benchmarking (still hope to soon publish it here) and small jobs (<500 pics) benefit greatly from GPU, cutting render time in half. But the bigger the job, the less GPU helps overall. When doing 2800 pics I need 828min with GPU and 870min without. Even with the fastest graphics in the world, that time would not be much shorter, since the GPU capable parts of the WebODM process are very few. See here:

The red bars in the beginning and towards 3/4 is the GPU usage. The rest is CPU.

RAM:
16Gbyte is very little especially for 2000 images to be processed.
A dataset with 1413 pics with 16Gbyte of RAM with feature- and pointcloud quality ultra takes 491min with 100Gybte of swap space in use. With 80Gbyte of RAM the same job is done in 316min and around 36Gbyte of swap space needed.
Also the speed of the SSD is important here. HDD won’t work, crashes with timeouts. NVMe is highly recommended.

SSD:
You mentioned a 500Gb or 1Tb SSD. Together with operating system and swapfile, it will be full after just a few projects.

With 2000 pics and a 8 core laptop processor and at least 32Gbyte of RAM it will take 10-24hrs to process one job. If it doesn’t get hung up on something. Smaller jobs should be done in one night.

I am running a dedicated, headless render machine. You could assemble a Ryzen 12-16core PC in an ITX case, add 64-128Gbyte of RAM and you can also add a CUDA capable graphics card (but that will increase the footprint).
Such a PC is very small (no screen, no keyboard etc.), can be carried around and you can use a slim laptop for the rest and your projects would be rendered in one night.

I like buying my IT equipment second hand and for the price of the HP Omen or the Lenovo Legion I could assemble the Ryzen miniPC and a nice laptop all second hand off course.

And about the HP laptop: it has a Radeon graphics card. That will not be helpful. At the moment WebODM supports CUDA cards only (NVIDIA Geforce, Quadro etc.).

All that said, if you would go with the Lenovo Legion laptop, get at least 32Gbyte or better 64Gbyte of RAM, add a 2Tb NVMe PCIe 4.0 SSD and it could work. I can not find details about the exact same model, but the Legion laptops from Lenovo often have a second M.2 2280 slot available. So you could just add another SSD.

[smathermather]

This is a super useful review! I’ve been wondering if this weren’t the case with GPU as datasets get larger based on responses on the forum, but it’s super useful to see it so concretely tested and described.