Processing Power And Location

[towforce]

I have been pleasantly surprised by the amount of interest in 8-bit microprocessors - the devices that first made home computers and dedicated chess computers possible, so... why not have a quick survey of what we're doing today?

Please cast two votes - one for usage profile and one for whether you do this computing locally or in the cloud.

My opinion of how people should profile:

* ordinary users: Chromebook (low CPU power, low cost, long battery life, flip screen/keyboard, ton of great features for ordinary users, low maintenance (if it breaks, your new one will be up and running with all the setup you had on the old one within a few minutes of logging in), fine for occasional power requirements like infrequent video editing). I understand that many people don't want to be tied to Google, though

* ordinary users who occasionally require large amount of CPU (e.g. occasionally training an NN): Chromebook + cloud. If you buy the hardware and only use it occasionally, it will become obsolete before you've had value from it

* consistent requirement for power (e.g. professional video editing, correspondence chess): kick ass rig

My vote: my laptop is at the high end for "ordinary", but it's no kick ass rig, so I'm voting "ordinary". While we use the cloud servers at work, I don't use it myself for any high power processing needs.

[Ras]

My vote: medium power (mostly low, occasionally high). Linux desktop, 5700X octacore with 88W peak, RX 6800 GPU for gaming. Running everything locally.

[smatovic]

My Computer Timeline:
https://zeta-chess.app26.de/post/computer-timeline/

currently:

>2021 - MaschinaIV, used Core i5-6500 4x3.2GHz (Skylake 14nm with AVX2 from 2015),
32GB RAM, Intel HD530 iGPU, Nvidia GTX 750, AMD RX 550 GPU, UbuntuLinux+Win10

I tried in 2019 to switch to Cloud Computing and stop collecting GPU architectures, but it did not work out, hence I did set up Maschina #IV with some entry level hardware to tinker with computer chess.

I intended in 2008 to drop the desktop PC and use just a laptop for my studies, but then came the desire to tinker with GPGPU.

--
Srdja

[towforce]

My Computer Timeline:
https://zeta-chess.app26.de/post/computer-timeline/

currently:

>2021 - MaschinaIV, used Core i5-6500 4x3.2GHz (Skylake 14nm with AVX2 from 2015),
32GB RAM, Intel HD530 iGPU, Nvidia GTX 750, AMD RX 550 GPU, UbuntuLinux+Win10

I tried in 2019 to switch to Cloud Computing and stop collecting GPU architectures, but it did not work out, hence I did set up Maschina #IV with some entry level hardware to tinker with computer chess.

I intended in 2008 to drop the desktop PC and use just a laptop for my studies, but then came the desire to tinker with GPGPU.

--
Srdja

Interesting. How was your experience with using the cloud for intensive processing?

[smatovic]

[...]
Interesting. How was your experience with using the cloud for intensive processing?

In 2019, there were only Nvidia server brand GPUs to rent, setting up an virtual server via web-platform works like a charm, but billing was an issue, meanwhile GPUs are hot potatoes, you will need to talk through the sales department of the big cloud players with an kind of business plan before you get access. It does not suit my use case of hobby (gpu) computer chess programming.

--
Srdja

[towforce]

Interesting. How was your experience with using the cloud for intensive processing?

In 2019, there were only Nvidia server brand GPUs to rent, setting up an virtual server via web-platform works like a charm, but billing was an issue, meanwhile GPUs are hot potatoes, you will need to talk through the sales department of the big cloud players with an kind of business plan before you get access. It does not suit my use case of hobby (gpu) computer chess programming.

That seems counter-intuitive to me:

* business usage: going to need to run it all the time: indicated action - buy the GPU and count it as "investment" for tax deductions

* occasional user: don't buy the GPU - you won't get value from it before it becomes obsolete. Share the GPU with others via cloud rental

Also, what if you want to test how much a GPU will improve your algorithm, or how much more quickly you can train an NN? You're not a business user yet - but you might become one.

Targeting businesses but deterring private individuals from renting computer power with a high calculation throughput doesn't make sense. I can see why you'd do this with a mainframe or a supercomputer to which you'd have to dedicate more-or-less your entire resource to a single user for just a short computing run - but not one rack out of hundreds at a cloud centre.

[smatovic]

Why do companies/people lease/rent cars instead of buying them? Cloud is about outsourcing, scalability and availability. For gamers there are already a couple of game streaming services....how does the saying go? Real men do own own servers? Or alike...

***edit***
We had the GPU shortage during Corona, then a short time a GPU surplus, now everybody is into GPGPU again, the generative AIs, take a look at Nvidia market cap, or alike...

--
Srdja

[smatovic]

...just for the files:

Lc0 - GPU cloud grid: https://lczero.org/contribute/
SF - Fishtest: https://github.com/glinscott/fishtest/wiki/Fishtest-faq
OpenBench: http://chess.grantnet.us/

--
Srdja

[Dann Corbit]

A couple thousand watts around the clock.
An AMD 3790 threadripper with 128 GB RAM, 140TB disk farm, 2 x NVIDIA GeForce RTX 2080 SUPER is my main machine.
It runs 24x7x365.

I have other, old servers that I run sometimes that consume a lot more power. But only when my machine above is jugged and I have something else I want to do and I don't feel like waiting.

I often have electricity bills of more than $600.

[smatovic]

[...]

What happened to "Ezekiel", your 4x16 core server rig???

--
Srdja