Desktop or Notebook?

[jokkebk]

A few weeks old thread but still my 5 cents...

Yes I think that 100 % vs. 167 % difference will come down mostly to personal taste, mobility vs. slightly faster analysis.

If you do analysis interactively, getting to review the first moves in 2 seconds vs. 3 seconds is not a big deal. If you want "fully processed and deep analyzed SGF" like some do, then it might be different whether you need to wait 10 or 15 minutes, but similar change would happen if you decide that "700 visits per move is not enough for me, I want 1000 visits".

I think the exact depth of analyzing each move and the time you get those in is a lot less important than doing dedicated, serious study. I think even 9p players would find most obstacles to progress in their heads instead of their laptops/desktops -- RTX 3060 would be plenty good for them. At 1k level, I'd probably do great with 1 playout if I took the time to think about my previous game.

But people like to obsess with "having the best possible" and "fastest possible", otherwise there would be no sports cars on the streets, as it's really irrelevant whether the top speed is 180 km/h or 380 km/h or acceleration to 100 km/h takes 3 seconds or 4 seconds. As far as GPUs go, it's relatively cheap way to feel that thrill of performance vs. a Ferrari.

[Gomoto]

I switched from my desktop with 3080 to laptop use only (Still using the same keyboard, mouse and desktop monitor.)

For my game reviews I use now exclusivly AI Sensei. The basic plan is good enough for my purpose.

I save a lot of energy using my laptop over the desktop and enjoy the zero noise silent computing.

Katago/Lizzie on the desktop provides faster browsing through the variations, but lacks the excellent spaced repetition training mode provided by AI Sensei.

I would be ready to start my desktop when I feel the need to explore an issue extra deep. But until now my laptop fulfills all my go, work and leasure needs (Are there other pleasures than exploring go?). The change from desktop to laptop is instant and effortless today with cloud computing.

Actually the slower AI Sensei speed helped me to think more and deeper about the explored variations and mistakes.

[xela]

Sorry to join the discussion so late...

My opinion is that if you're using the AI as a research tool, then the speed differences in your first post won't be significant. First of all you'll spend time choosing which games and positions to look at, and setting them up in your software. Then you'll want to "interrogate" the AI. This includes things such as observing whether or not the evaluations change with more playouts, comparing the policy values with the final evaluations, asking the AI to give a deeper evaluation of a second or third choice move (possibly including a "refutation" of those moves), exploring other variations, checking if two different AIs agree on the best move, ... You'll typically spend between two minutes and an hour exploring a position. Human judgement is the speed bottleneck, not GPU specs (as long as the GPUs you're comparing are within the same ballpark).

AI isn't yet good enough to consistently give superhuman results in all types of positions within a fraction of a second on consumer-grade hardware.

In many cases, the AI will choose a preferred move "at a glance" (within a fraction of a second), and the choice of best move won't change with deeper analysis; the only change is in the estimated winrate for that move. But sometimes the preferred move will change with more playouts. I've seen examples in pro games where the AI's "at a glance" suggestion is different from the human's move, but after a full minute of analysis the AI then agrees with the human choice. In other words, human judgement, at its best, in certain positions, can be better than the AI's "first instinct". But the opposite can also happen: the AI's "at a glance" move is the pro move, but a minute later the AI starts to suggest a different (presumably better) move.

Remember also that you can't tell just by looking at a move suggestion whether it's "superhuman" or on a par with pro judgement, or whether it represents a blind spot from the AI. You'll need more context to feel confident in the AI's suggestions, and you'll never be 100% sure. And there's not yet a consensus on how best to interrogate the AI.

So I would suggest giving only a small weight to the speed differences between notebook and desktop, and make the decision more on the basis of other factors (e.g. cost, convenience).

And I look forward to seeing how you approach the problem of using an AI to learn about go :-)

[RobertJasiek]

I expect to use it in different ways depending on study purposes (improving my play, teaching, finding new go theory, reevaluating earlier go theory, comparing AI play to mathematically proved perfect play) and AI mistakes I can identify.

[goame]

As you know, I have tried to buy an RTX 3080 in vain and now await RTX 4070 to possibly become available at its MSRPs around January 2023. The alternative is a notebook. Desktops and notebooks have their advantages and disadvantages but a major difference is relative speed. Different benchmarks differ, cooling has a great impact etc. To simplify my decision-making, let me state these rough relative benchmarks for 2560x1440 Time Spy Graphics, which might not be the most meaningful benchmark for go AI but is also not the most meaningless:

Code: Select all

169%   RTX 3080 desktop
100%   RTX 3070 TI laptop

So a desktop build might give roughly 69% faster speed. What does this mean in practice when using, say, KataGo? Suppose 3080 desktop enables next move suggestions at above-human level within 0,25s, would 3070 TI laptop then simply mean move suggestions at above-human level within 0,4s and still be more than good enough? Or would move suggestions at above-human level require 2.5s versus 4s for a drastic difference in usability?

AI suggestions at 9p level are insufficient for me - I want clearly above-human level. You know, 9p can make mistakes but it must be hard to detect any AI mistakes.

What do you recommend me? Is a well cooled RTX 3070 TI laptop graphics card good enough for me 5d seeking clearly above-human level AI moves or is at least 3080 desktop mandatory?

For good notebook cooling, imagine MSI Titan GT77 (too loud for my taste), Alienware 17" (too loud and awkward design) or XMG Neo 17 M22 (to be available next month, noise to be tested). Can such notebooks maintain many hours of maximum go AI load without overstraining the cooling system? Or must it be a large desktop tower with exceptional airflow?

I have comtemplated these things for two years but still cannot decide whether notebooks make sense at all for go AI at super-human level. Please advise!

***

On a related question, not every notebook has all navigation keys as dedicated keys. When using go AI software, which navigation keys are needed frequently? Arrows keys? Pos1 and End? PageUp and PageDown? Is a mouse needed or will any touchpad do?

If you want to be happy then buy the Apple MacBook Pro M1 MAX, 32 GPU cores and 64 GB RAM:
https://www.apple.com/shop/buy-mac/macb ... re-gpu-1tb#
It's stronger than every super-human.
It's quiet.
It's not hot.
It need only 1/40 of watt compared to other GPUs.
You can use macOS, Windows 11 ARM and Linux, you can even run iPadOS apps.
Parallels and CrossOver works finde too.
It has by far the best screen to look at.
It has by far the best sound speakers.
It runs always at full speed and is not throttling down the performance after 1 minute to only 30% like other not Apple devices.

[RobertJasiek]

If I choose such, it would be the MacBook Pro 16 M1 Pro for its best battery life among the MBPs.

While it is said to be quiet, some testers have found up to 45dB under load. I cannot know what noise level KataGo would produce. Given that I would accept up to 43dB in a notebook, maybe this could be achieved. So for some, it might be a suitable choice.

However, the notch is absolutely unacceptable for me. Never will I buy any device with a notch / hole / drop in the display. Such equals a destroyed display on manufacturing. The following considers MBP M1 if it did not have a notch.

Apple has betrayed trust by its iCloud terms, which it tries to impose on my iPad at least daily and it punishes me for denying by prohibiting updates via WLAN. These terms claim to grant Apple the right to use all my local files and give them to anybody in the world. Although illegal and criminal, this is what the terms say. I do not buy a new device from such a company even if MBPs should be affected less thus far.

Apple's repair prices and customer disrespect in some warranty cases are beyond words. Apple's anti-repair legislation lobbying is untolerable. Again, Apple does not deserve to earn any more money from me.

[RobertJasiek]

Nvidia contributes greatly to the decision-making;) Price increases:

3080 10G -> 4080 16G = +72%

3070 -> 4080 12G = +80%

I do not know if AMD is any good.

[kvasir]

There could be months before you can buy 40x0 cards. It was announced this week, the release date that I found is 12th October but in the past it has taken the OEM card manufactures months to make new cards available. With the chip shortage it could potentially be 6-12 months but we will see.

The way that Nvidia operates is that they make chips and a very small number of cards. The bulk of the cards are made by OEMs that often have different idea, from that of NVIDA, about costs and availability. If the 30x0 market is generating revenue for the OEMs then they can't be expected to introduce the next generation cards in large numbers, instead they might make them available for very large amount of money

My 3080 laptop version is doing great things for me, every single day

[RobertJasiek]

My 3080 laptop

If I may ask, what laptop model, which CPU, what speed / noise mode, what noise level, and how would you subjectively characterise the noise?

[kvasir]

It is a Lenovo Legion R9000K which I think was only available in China but I think it pretty much the same as Lenovo Legion 7 which was released at least half a year later.

I am already at the stage that I don't know the specs anymore, so I am just reading what I can find

• AMD Ryzen 9 5900HX with Radeon Graphics 3.30 GHz
• 64 GB ram
• NVIDIA GeForce RTX 3080 Laptop GPU
• Other stuff

It is built like an oversized vacuum cleaner head but it looks cool. One side of the base is open with two fans that such air in, the air is then blown out the sides. I am impressed that it is often silent but frequently it will sound like a vacuum cleaner but a lot quieter, really. It is an annoying noise because it is high pitch that doesn't always blend in with other sounds. That is how it sounds, as for how loud then I'd say it is like a vacuum cleaner thorough a door at the worst but there are three settings (that you cycle through with shift-Q): quiet, auto and performance. The quiet setting mostly kills the noise, even when working the GPU, but it is not completely quiet. I live in a quiet place now but before I lived in a place that needed a constant AC running for me to stay alive and that device is louder than the laptop. The noise from the laptop will also disappear in a large enough room or a coffee shop. Basically when it comes to sound it is complicated.

I have no idea what the dB would be, I think it is a useless measure unless you actually have device to measure it. In my experience the dB level in a room is often dominated by one device like an AC or a refrigerator, but even moderate rain can outweigh anything that is in the room and rain makes pleasant sound while many artificial devices make unpleasant sounds.

Basically, if the main concern is noise and performance then I'd really recommend a desktop tower instead and you can put it under the table. If the main concerns include the possibility of taking the computer somewhere, then that is what laptops are for.

[RobertJasiek]

Very interesting, thank you!

Suppose performance mode has 100% speed. Can you specify how much of that speed remains in a) Auto and b) Quiet modes?

[RobertJasiek]

Notebookcheck tested the Legion 7 variant of your notebook and wrote:

AMD Ryzen 9 5900HX 87.8 W PL2, 87.8 W PL1
NVIDIA RTX 3080 Laptop 165 W TDP (incl. 15 W Dynamic Boost)

Idle 26 / 30 / 35 dB(A)
Load 40 / 50 dB(A)

Witcher-3 gaming has 46 dB(A), maximum load of both CPU and GPU has 50 dB(A). High fan speeds involve a whistling. A headset is recommended for gaming.

So their measurements agree to your subjective description. While it is not a silent notebook, for 88 W PL1 and 165 W GPU it is reasonable compared to similarly equipped fast notebooks.

[kvasir]

I don't know exactly what the different modes change but I suspect it has mostly to do with the fans. The silent mode seems to simply run the fans at a lower speed, the auto mode seems to change the fan speeds aggressively and the performance mode seems to run the fans at a higher speeds but actually if there is very little running then the computer will be silent in this mode too. Probably there is also some change in how aggressively the CPU and other components step up their clock rate or turn off unused things, but components will do this automatically anyway. Components like the CPU have very advanced power consumption control and heat throttling, while components like RAM and GPUs don't. Possibly, simply controlling the fan speed is the greater part of the effect but I just don't have a way to tell. I mean the fan speeds often change noticeably but the CPU clock speed might not when I cycle the modes and I don't know if I should describe that as the usual behavior or not, and I don't thing the "performance" mode disables throttling.

It is nice to be able to almost silence the laptop with Fn+Q (I said Shif+Q before but it is the the Fn key instead) and not know exactly what it does.

[RobertJasiek]

Quite a few newer "gaming" notebooks allow manual setting of fan curves as temperature over fan speed or such in some Balanced or Custom mode.

[kvasir]

I think the laptop came with some branded Lenovo programs to change different settings. I installed a fresh OS when I got the laptop and don't have most of the pre-installed programs, a few were installed anyway by Windows Update but the upgrade to Windows 11 removed a almost all of this. Either they are not ready and might be installed with a later update or the system is not recognized by Windows Update as "needing" whatever branded programs are available. Preinstalled programs annoy me and I am happy to not have them, but then I don't know what additional settings they might have.