Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / LeelaChess crashing in UCI go nodes
- - By turbojuice1122 (Gold) Date 2019-06-09 14:44
I am having trouble getting Leela Chess to work on my computer.  I used DriverBooster to update all of the drivers and restarted.  I have downloaded the cuda version and the weights from one of the networks into the same folder on my SSD, and when running lc0 and typing go nodes 100, it says,

Found pb network file: C:\LC0/weights.pb
Creating backend [cudnn]...

Then the window closes after a few seconds.  If I restart it and do the same thing, it closes after only about 1 second.

My computer has a 6-core Core i7-8700K at 3.7 GHz, 64 GB RAM, and an 8 GB NVIDIA GeForce GTX 1080

Any ideas?
Parent - - By Master Om (Bronze) Date 2019-06-09 16:21
try to update CUDA files. Something like that Nvidia SDK files. See what happens.
Parent - - By turbojuice1122 (Gold) Date 2019-06-09 18:04
Thanks, I downloaded and installed the CUDA toolkit from the NVIDIA website, but it still doesn't work.  (It takes longer to crash this time, though, I think.)

Should I download and install NVIDIA's SDK manager?  It seems that I need a special account in order to do this.  Is this typical?
Parent - - By MrKris (***) Date 2019-06-09 21:29
I do not have Lc0, but I noticed this page:

https://github.com/LeelaChessZero/lc0/wiki/Getting-Started

Towards the bottom:
" Common problems
lc0.exe exits without printing anything
Double check you have cublas64_100.dll, cudart64_100.dll, and cudnn64_7.dll in the same directory as lc0.exe
"
(maybe the toolkit did not put them in the right place)

Try, from the link above:
" Debug
For easier debug, verify install by running from command line (cmd.exe)
cd to directory with lc0.exe.
run lc0.exe --logfile=log.txt
Most recent weights file in that directory will be used automatically.
Add --weights=c:\path\to\weights_476.txt.gz to specify manually.
"
and see what the file log.txt says.
Parent - By turbojuice1122 (Gold) Date 2019-06-09 23:25
Thanks for the link.

I do all this, and in the log file, I get the following:

CUDA error: CUDA driver version is insufficient for CUDA runtime version

But...I just updated my CUDA drivers, and I made sure that my GPU is operating in maximum performance mode.
Parent - - By Master Om (Bronze) Date 2019-06-10 18:44
Try The CPU only version and see. I think will work.
Parent - By turbojuice1122 (Gold) Date 2019-06-10 22:38
Yes, this works, but it's very slow: 500 nodes per second.
Parent - By Peter Grayson (****) Date 2019-06-10 10:02

> I am having trouble getting Leela Chess to work on my computer.  I used DriverBooster to update all of the drivers and restarted.  I have downloaded the cuda version and the weights from one of the networks into the same folder on my SSD, and when running lc0 and typing go nodes 100, it says,
>
> Found pb network file: C:\LC0/weights.pb
> Creating backend [cudnn]...
>
> Then the window closes after a few seconds.  If I restart it and do the same thing, it closes after only about 1 second.
>
> My computer has a 6-core Core i7-8700K at 3.7 GHz, 64 GB RAM, and an 8 GB NVIDIA GeForce GTX 1080


I have the same problem with my i7-8700 + 6Gb GTX 1060. Latest nVidia CUDA drivers installed plus other nVidia stuff. I also installed the recommended Microsoft development package. Wasted a lot of time on something that should be so simple to do but concluded it just was not woth the effort and wasting more time. In comparison to settting up an alpha-beta uci engine, for LC0 the user friendliness gets a zero rating!
Parent - - By Uly (Gold) Date 2019-06-10 16:57
Just checking, can you run the CPU-only version? I also wasted a lot of time trying to get Leela to work on my GPU, and it seems it was impossible, but the CPU-only version did work (I posted my thoughts about it in Talkchess, but the TLDR is that Leela at Depth 8, which takes 10 minutes to reach looks worse than TSCP at Depth 8, which takes a second, such a disappointment but I might buy a capable GPU just out of curiosity...)
Parent - - By turbojuice1122 (Gold) Date 2019-06-10 22:40
Yes, that works.  But the thing is, I very definitely have a capable GPU--there just seems to be something screwy with the program itself.
Parent - - By Master Om (Bronze) Date 2019-06-11 16:42
now try the open GL version. See if it works.
Parent - - By turbojuice1122 (Gold) Date 2019-06-12 00:11
You mean the opencl version?  I tried this, and this works.

I'm a bit illiterate with this stuff, in spite of a little Googling.  What exactly is the opencl version?  I see it stated as Open Cloud, but other than that, I really don't know.
Parent - By Labyrinth (*****) Date 2019-06-12 16:55
Open CL (Open Computing Language), is basically CUDA but open source since CUDA is proprietary Nvidia and is made to only work on Nvidia hardware (there are ways around this though).

Both are software frameworks that enable the use of a GPU's processing power for unusual tasks.
Parent - By Master Om (Bronze) Date 2019-06-12 18:46
Me illiterate too. But see i found a work around. :smile:
Parent - - By Lukas Cimiotti (Bronze) Date 2019-06-12 13:06 Upvotes 2
I'm running lc0 on a bunch of computers using CUDA on RTX 2080 and 2080Ti.
You don't have to install any NVidia stuff but the drivers.
Update Windows to version 1809, get all updates. Install GeForce Experience. Update the driver using this tool.
For me this works - I hope it will work for you, too.
Parent - - By turbojuice1122 (Gold) Date 2019-06-12 23:36
Thanks--though I can't get version 1809.  It makes me get version 1903, and upon attempted installation, says that I have hardware on the machine that isn't ready for this version.  So, I seem to be stuck with version 1803.
Parent - - By Peter Grayson (****) Date 2019-06-13 14:08 Upvotes 1

> Thanks--though I can't get version 1809.  It makes me get version 1903, and upon attempted installation, says that I have hardware on the machine that isn't ready for this version.  So, I seem to be stuck with version 1803.


I had the same message for my two, 10 year old Intel Q9550 machines when I ran the W10 compatibility tester. However, on the basis I had two machines running Windows 64 7 Pro that would become redundant in the not too distant future I decided to run the automatic W10 update on one of them to see what happened. Windows 10 64 Pro installed and is running fine and currently on version 1809. During installation there was a message regarding an incompatible graphics card but after installation I updated the Graphics card drivers and it switched to using the graphics card from Windows generic graphics drivers. The only comment I would make is to disconnect from the Internet once the installer has downloaded W10 because on my Dell dual Xeon machine it tried to run W10 updates during the installation from W7 Pro with the consequence the installation failed and it dropped back to W7. Just wasted time but it does seem to give that fall back safeguard if it fails to install properly.

I've just updated the dual Xeon to W10 Pro 1903 and no issues so waiting for W10 1903 update on the two Q9550 machines now.

Peter
Parent - - By Peter Grayson (****) Date 2019-06-14 21:54 Upvotes 1
With Windows 10 1903 update completed on the i7-8700 machine, I installed the nVidia 10.1 development pack and noted the graphics drivers were much later than the W10 installed drivers that reported they were up to date. Since installing the drivers from this package the LC0 CUDA engine is now working. Presently the performance on the nVidia GTX 1060 + 6Gb card looks to be similar to Brainfish 100619 64 BMI2 running on the i7-8700 CPU so a good match for comarison purposes.

Peter
Parent - - By MrKris (***) Date 2019-08-12 06:19
Thanks, I did not notice my machine did not have 1903 (the web says its not automatic) I started the built-in 'Windows Update' manually and got it there.

Then I re-ran GeForce Experience, and 'clean' re-installed the Nvidia drivers.

RTX 2060: Lc0 Net 42850 , cudnn-fp-16 , threads 2 , NNCacheSize 2,000,000 , MinibatchSize 512 , MaxPrefetch 64

i7-4770K 8HTthreads: Stockfish_19080109 bmi 1GBhash

Lc0 knps mannually added as a comment:
[Event "Arena 3.5.1"]
[Site "Game 5mins + 2 secs"]
[Date "2019.08.08"]
[Round "?"]
[White "Lc0-v0.22.0"]
[Black "human"]
[Result "1/2-1/2"]
[BlackElo "2400"]
[ECO "C96"]
[Opening "Spanish"]
[Time "21:33:26"]
[Variation "Closed, Chigorin, 10.Bc2"]
[WhiteElo "2200"]
[TimeControl "300+2"]
[Termination "adjudication"]
[PlyCount "19"]
[WhiteType "program"]
[BlackType "human"]

{TC 5+2} 1. e4 {+0.28/15 6" 15k} e5 2. Nf3 {+0.26/17 4" 18k} Nc6 3. Bb5 {+0.25/18 5"
19k} a6 4. Ba4 {+0.32/18 7" 18k} Nf6 5. O-O {+0.32/19 4" 24k} Be7 6. Re1
{+0.48/16 8" 23k} b5 7. Bb3 {+0.47/16 3" 22k} d6 8. c3 {+0.47/17 6" 28k} O-O
9. h3 {+0.43/18 4" 27k} Na5 10. Bc2 {+0.60/13 11" 19k User Adjudication --- average 21.3k/move}
1/2-1/2

Stockfish_19080109-bmi nps (black move reset to above if diferent) G/5'+2"
1...e6 8M; 2... 8M; 3...Nf6 8.2M; 4... 8.1M; 5...Nxe4 8.2M; 6... 8.2M; 7... O-O 8.2M; 8... Na5 8.2M; 9... 8.2M; 10... c5 8.4M --- average 8.17M/move

8.17M-Sf / 21.3k-Lc0 = 384

850(standard LR 1) / 384 = 2.21 LR non-standard method above
Parent - - By Peter Grayson (****) Date 2019-08-12 11:13
Glad you got it working.

I made my comment on the LC0 performance being equal to the i7-8700 after 50 games of one my test sets playing the engine against Brainfish. The first 25 opening lines were mostly King's pawn openings when the score was W7-D36-L7. The second part were mostly non King's pawn openings and LC0 looked a different engine when it suffered something of a drubbing with a score of W1-D34-L15. I ran with the TCEC network used at the time of the test. I have not tried the latest release and maybe network 42850 will be better.

My other machines do not have the graphics card capability because until the LC0 engine came on the scene graphics card performance was never a consideration or necessary to obtain a high standard of chess. It still isn't and accommodating LC0 currently adds significant cost to a machine purchase. As I understand it, long time controls and big hardware are necessary to obtain a good performance from LC0.

The contrast in the wins between LC0 and Brainfish was that LC0 could get caught out by a sharp tactical shot and its wins tended to take 100+ moves whereas Brainfish looked tactically sharp and smart when delivering mate, finishing off LC0 in under 50 moves in 7/100 games!

My own observations of LC0 output suggested the kN/s and ply depth are meaningless compared to an alpha-beta engine and may be produced just for uci protocol compliance.

Peter

This game highlighted the tactical naivity of LCO ...
[Event "Brainfish 100619 vs LC0 CUDA"]
[Site "Newport, South Wales"]
[Date "2019.06.16"]
[Round "100.1"]
[White "Lc0 v0.21.2 CUDA"]
[Black "Brainfish 100619 64 BMI2"]
[Result "0-1"]
[ECO "D46"]
[Annotator "0.30;0.15"]
[PlyCount "90"]
[EventDate "2019.06.14"]
[EventType "tourn"]
[SourceTitle "Fritz Engine Tournament"]
[Source "Grayson"]

{Intel(R) Core(TM) i7-8700 CPU @ 3.20GHz 3192 MHz W=11.2 plies; 6kN/s B=44.0
plies; 21,728kN/s; 819,873 TBAs} 1. d4 d5 2. c4 c6 3. Nf3 Nf6 4. Nc3 e6 5. e3
Nbd7 6. Qc2 Bd6 7. Bd3 O-O 8. O-O dxc4 9. Bxc4 b5 10. Be2 Bb7 11. e4 e5 12.
dxe5 {0.30/9 12} Nxe5 {0.15/27 15} 13. Nd4 {0.23/11 21 (Nh4)} Neg4 {-0.23/26 9}
14. Nf5 {0.32/14 9 (g3)} Bxh2+ {-0.13/30 15} 15. Kh1 {0.27/12 0} Bb8 {-0.14/30
7 (b4)} 16. f3 {0.00/13 35 (Bg5)} Nh5 {-1.19/28 8 (Qc7)} 17. fxg4 {-0.30/13 12}
Qc7 {-0.85/32 24} 18. Kg1 {-0.45/13 11 (g3)} Qh2+ {-1.10/29 16} 19. Kf2 {
-0.51/15 0} Ng3 {-0.92/30 15} 20. Bd3 {-0.42/13 24 (Rg1)} g6 {-1.04/28 12
(Nxf1)} 21. Nd4 {-0.04/16 21 (Qd2)} Nxf1 {-1.26/29 7} 22. Nf3 {-0.04/11 2
(Kxf1)} Bg3+ {-3.21/28 8} 23. Kxf1 {-0.30/16 11} Qh1+ {-3.08/32 26} 24. Ng1 {
-0.32/17 1} f5 {-3.45/32 7 (Bc8)} 25. exf5 {-0.07/18 37} gxf5 {-3.40/33 11} 26.
gxf5 {-0.53/25 21 (Qb3+)} c5 {-5.05/30 9 (Rae8)} 27. Be3 {0.44/11 25} Rxf5+ {
-7.32/31 10 (Rae8)} 28. Bxf5 {-2.16/9 24} Rf8 {-7.65/31 10} 29. Bxc5 {-3.04/10
41 (Ke2)} Rxf5+ {-11.92/31 12} 30. Ke2 {-3.25/10 7} Rxc5 {-13.53/29 9} 31. Kd3
{-3.29/11 29} Qh6 {-14.71/29 10} 32. Qb3+ {-3.40/10 5} Kg7 {-16.16/29 10 (Kh8)}
33. Qb4 {-3.20/9 19} Qg6+ {-17.13/30 16} 34. Kd2 {-3.40/10 12} Qg5+ {-18.00/28
8 (Bd6)} 35. Kd1 {-3.57/10 26} Qf5 {-22.17/29 14 (Rc4)} 36. Qd4+ {-3.18/8 20}
Be5 {-25.66/29 14} 37. Qe3 {-3.63/9 13} b4 {-30.49/25 8} 38. Nce2 {-4.97/9 24
(g4)} Qf1+ {-#17/48 8} 39. Kd2 {-7.00/8 0} Rd5+ {-#15/56 7} 40. Kc2 {-10.45/7
31 (Qd3)} Qf5+ {-#11/55 7 (Qxa1)} 41. Kb3 {-9.60/6 15 (Kc1)} Rd3+ {-#8/63 8}
42. Kc2 {-15.25/5 18 (Nc3)} Rxe3+ {-#6/86 7} 43. Kd2 {-45.37/7 15} Rd3+ {
-#5/134 7 (Bxb2)} 44. Ke1 {-55.24/6 14 (Kc2)} Bxg2 {-#3/245 0} 45. Rc1 {
-128.00/2 14 (Nf4)} Qf1# {-#1/245 0} 0-1

Parent - By MrKris (***) Date 2019-08-13 05:21 Edited 2019-08-13 06:57
Interesting about relative strength, Lc0 nets, and long TC's!

I did not even have a GPU (Arena 'graphics') so I had saved on that in the past.
I could afford the RTX 2060 but did not want (some might) the 'Super' for only $50 more.
The new GTX 1650 (4600 nps on the web 1 place only*) /1660/1660 Ti with their fp-16 looked okay, but the Ti a little close in price to mine I thought.
If I ever get a new CPU (there are so many now) I would rather transfer my RTX to it even if its 'slow' by then I could reduce CPU threads or something for A/B's.

Lc0's nps and depths look strange to me also. In my little timing trial 9.h3 4 secs. for depth 18 @ 27k, then 9...Na5 was manually, then it continued with 10.Bc2 11 secs. for only depth 13 and only 19k.

I liked your game. " … 25. exf5 -0.07/18 37 …" near 0 evaluation 18 plies but in just a few plies and a few good moves it did not expect, 26...c5! (Rad8) and 27...Rxf5!! (Rad8), black gets an old fashioned king hunt.

--- As to cost I would say RAM looks also like a potential difficulty, but I am just now looking into it.

____
* https://groups.google.com/forum/m/#!topic/lczero/XCZroMMs1hI
Parent - - By turbojuice1122 (Gold) Date 2019-06-12 23:56 Upvotes 2
Update: the driver update with GeForce instead of Driver Booster was apparently the key.  The CUDA program now runs normally, and I'm getting 7 knps on the opening position.

Thanks, Lukas!
Parent - - By Uly (Gold) Date 2019-06-13 14:30
Hey, please keep up updated about what you think of Leela. I'm considering getting a decent GPU to run it (you were complaining of only getting 500 nps? I get 1 nps!) but the more data the merrier.
Parent - - By Lukas Cimiotti (Bronze) Date 2019-06-13 17:10 Upvotes 2
If you can afford it try to get a latest generation NVidia GPU - GTX 1650 and up or RTX. These GPUs support fp16 calculations which makes them much faster and much more efficient. A RTX 2080 Ti gets ~ 60 kn/s, a RTX 2080 ~ 45 kn/s.
I made a lot of tests: lc0 on a RTX 2080 is roughly as strong as Stockfish 10 running on 10 cores on a modern Intel CPU @ 4 GHz.
Parent - - By Uly (Gold) Date 2019-06-13 21:00 Edited 2019-06-13 21:03
Thanks for the suggestions.

I am looking for something more like a... GeForce 710? At least I could buy it soon, while with the GTX 1650 I'd need to wait like 8 months to buy it or something...

How "indecent" would it be? I'm "upgrading" from a 4CPU Q8200 @2.33Ghz, so the question is if a 710 would be worth it, or I'd rather wait (and hopefully in the future I don't wait again lest I just wait and wait for something better.)
Parent - - By Labyrinth (*****) Date 2019-06-14 02:29
So Geforce 710 is a low end card released January 2016, I'm surprised it's even a consideration. Even a geforce 1030 is like, 3x as fast, and I can't imagine that running Leela well. Even this is >4x as fast, although not sure about that seller.
Parent - By Uly (Gold) Date 2019-06-14 10:08
I have problems getting stuff by mail, so I'm limited to buying what I can directly go to a shop and get. The 710 is appealing because I can go to a shop and buy it, and it's cheap. The 1030 costs almost twice as much, and at that price range it might be better to just wait for the 1650 (the 1650 is just the equivalent of US$50 pricier than the 1030?).

But yeah, I'm ruling out the 710, seems it's not cost-efficient :smile:
Parent - - By turbojuice1122 (Gold) Date 2019-06-15 12:22
I actually have a GTX 1080, which is getting on the order of 5-10 knps, depending on the position, but according to the benchmarks, the RTX 2060 is slightly faster and a lot cheaper.
Parent - - By Vegan (****) Date 2019-06-23 01:53
The GTX 1080 is a powerful card but its designed for 3D graphics not AI

The RTX 2060 is the lowest cost card and it is not that good for games by comparison
Parent - By Labyrinth (*****) Date 2019-06-23 20:39

>The RTX 2060 is the lowest cost card and it is not that good for games by comparison


In comparison to what? 2080 Ti? An RTX 2060 is certainly 'good for games'.
Parent - - By turbojuice1122 (Gold) Date 2019-06-15 12:25
Have you done tests at longer time controls?  I have the impression that Leela benefits more from longer time controls/analysis times than do the mainstream engines--similar to how a human does.
Parent - - By Lukas Cimiotti (Bronze) Date 2019-06-16 08:20 Upvotes 1
I only did tests at a time control of 4 minutes + 2sec. increment.
There is one problem with longer games: leela keeps the whole search tree in memory. It so uses a huge amount of memory. I upgraded my computers to 64 GB RAM.
Parent - - By MrKris (***) Date 2019-08-13 07:06
What happens if one has too little memory? For example too long a TC?
I will watch Task Manager, maybe I've been swapping to my SSD and did not notice (which would wear it out eventually I think).

I tried Lc0's RamLimitMB UCI option. Its default 0 apparently means "no limit" and a higher value the limit.
With "1000" I got this:
Task Manager for Lc0 process stopped at rising at 2051 MB memory with the first warning and never went any higher.
So it seems to work with an implied Lc0 minimum of about 1051 MB first.

But in game mode if it does not make the move with the first WARNING below, from analysis mode, then one would still have know what TC to use, but the limit will waste time without depleting system memory.
FEN: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1 initial position
Lc0-v0.22.0 , NNCasheSize 2,000,000
 1/2  00:00   4  190  +0.31  e2-e4 e7-e5
...
 17/53  01:12   1,828k  25k  +0.25  d2-d4 Ng8-f6 c2-c4 e7-e6 g2-g3 Bf8-b4+ Bc1-d2 Bb4-e7 Bf1-g2 d7-d5 Ng1-f3 O-O O-O c7-c6 Qd1-c2 Nb8-d7 Rf1-d1 b7-b6 Bd2-f4 Bc8-a6 Nb1-d2 Nf6-h5 Bf4-e3 Ra8-c8 Ra1-c1 h7-h6 Nf3-e5 Nh5-f6 Qc2-a4 Nd7-b8 Qa4-b3 Nb8-d7 h2-h3 Ba6-b7 Qb3-a4 a7-a6 c4xd5 c6xd5 Nd2-f3 b6-b5 Qa4-b3 Nd7-b6 Rc1xc8
 17/53  01:17   1,881k  24k  +0.25  d2-d4 Ng8-f6 c2-c4 e7-e6 g2-g3 Bf8-b4+ Bc1-d2 Bb4-e7 Ng1-f3 O-O Bf1-g2 d7-d5 O-O c7-c6 Qd1-c2 Nb8-d7 Rf1-d1 b7-b6 Bd2-f4 Bc8-a6 Nb1-d2 Nf6-h5 Bf4-e3 Ra8-c8 Ra1-c1 h7-h6 Nf3-e5 Nh5-f6 Qc2-a4 Nd7-b8 Qa4-b3 Nb8-d7 h2-h3 Ba6-b7 Qb3-a4 a7-a6 c4xd5 c6xd5 Nd2-f3 b6-b5 Qa4-b3 Nd7-b6
WARNING: Search has reached limit and does not make any progress.
 17/53  01:22   1,881k  23k  +0.25  d2-d4 Ng8-f6 c2-c4 e7-e6 g2-g3 Bf8-b4+ Bc1-d2 Bb4-e7 Ng1-f3 O-O Bf1-g2 d7-d5 O-O c7-c6 Qd1-c2 Nb8-d7 Rf1-d1 b7-b6 Bd2-f4 Bc8-a6 Nb1-d2 Nf6-h5 Bf4-e3 Ra8-c8 Ra1-c1 h7-h6 Nf3-e5 Nh5-f6 Qc2-a4 Nd7-b8 Qa4-b3 Nb8-d7 h2-h3 Ba6-b7 Qb3-a4 a7-a6 c4xd5 c6xd5 Nd2-f3 b6-b5 Qa4-b3 Nd7-b6
WARNING: Search has reached limit and does not make any progress.
 17/53  01:27   1,881k  22k  +0.25  d2-d4 Ng8-f6 c2-c4 e7-e6 g2-g3 Bf8-b4+ Bc1-d2 Bb4-e7 Ng1-f3 O-O Bf1-g2 d7-d5 O-O c7-c6 Qd1-c2 Nb8-d7 Rf1-d1 b7-b6 Bd2-f4 Bc8-a6 Nb1-d2 Nf6-h5 Bf4-e3 Ra8-c8 Ra1-c1 h7-h6 Nf3-e5 Nh5-f6 Qc2-a4 Nd7-b8 Qa4-b3 Nb8-d7 h2-h3 Ba6-b7 Qb3-a4 a7-a6 c4xd5 c6xd5 Nd2-f3 b6-b5 Qa4-b3 Nd7-b6
...
WARNING: Search has reached limit and does not make any progress.
 17/53  02:57   1,881k  11k  +0.25  d2-d4 Ng8-f6 c2-c4 e7-e6 g2-g3 Bf8-b4+ Bc1-d2 Bb4-e7 Ng1-f3 O-O Bf1-g2 d7-d5 O-O c7-c6 Qd1-c2 Nb8-d7 Rf1-d1 b7-b6 Bd2-f4 Bc8-a6 Nb1-d2 Nf6-h5 Bf4-e3 Ra8-c8 Ra1-c1 h7-h6 Nf3-e5 Nh5-f6 Qc2-a4 Nd7-b8 Qa4-b3 Nb8-d7 h2-h3 Ba6-b7 Qb3-a4 a7-a6 c4xd5 c6xd5 Nd2-f3 b6-b5 Qa4-b3 Nd7-b6
WARNING: Search has reached limit and does not make any progress.
{analysis then manually stopped in GUI}
Parent - - By Lukas Cimiotti (Bronze) Date 2019-08-14 10:25 Upvotes 1

>What happens if one has too little memory? For example too long a TC?


It will start swapping which will hurt performance.

>I tried Lc0's RamLimitMB UCI option.


This option obviously hurts performance very much. One thing is very strange: I tried this option, but it didn't work for me. It simply had no effect.

The only thing that currently makes sense is upgrading RAM.
Parent - By MrKris (***) Date 2019-08-16 07:16
For RamLimitMB maybe your GUI does not print the 'info' lines?

> The only thing that currently makes sense is upgrading RAM.


Especially for analysis/infinite.

8GB System RAM , G/15'+5", Lc0 0.21.4 vs. Lc0 0.22.0 , initial position/no-book, different recent nets
(although that is not long by any means. Lc0 vs. Stockfish/A.B. at modest hash would help too, and I presume with a lower draw rate)
In the game the Lc0's reached a peak memory use fairly early, then declined sharply, I believe as material was exchanged.
I am guessing that NNCasheSize sets the number of entries and that they get smaller the lessor material.
The peak data is from Microsoft's Process Explorer https://docs.microsoft.com/en-us/sysinternals/downloads/process-explorer .
The game lasted 27 mins. The peaks total (although they probably occurred on different moves) was within my 8GB.
Final position when I adjudicated the draw, both a few centipawns:
8/1p3k2/p1b5/5p2/1P3K2/2B1RP1r/2P5/8 w - - 3 51
Parent - By MrKris (***) Date 2019-08-16 07:38
Lc0 0.22.0 , Infinite analysis (800000 NNCasheSize here and above)

The memory use started to slowly but steadily climb right away.
At 18 mins. the memory use  was undulating above and below 90%.
I stopped it a double that, 37 mins., when it had been swapping.
(I think, but am not sure, the undulation continued most of the rest of the time.)
Even though it was swapping, turning off analysis restored the RAM to a lower level and stopped swapping, in case one wanted to save the analysis or game or such.
Certainly something one would wish to avoid though. (Arena has a move in a fixed time level if one could estimate that.)
Parent - By MrKris (***) Date 2019-08-17 15:29 Edited 2019-08-17 15:37
i7-4770K 8GB total system RAM, Syzygy 5+6
Stockfish 1GB defaults
Lc0 0.22.0 Net 320x24.J13-swa-410000 defaults ex. PV=6, RTX 2060

TC: G/90 mins. + 10 secs. (Arena demo, no drawscore)

In game mode it is very difficult to anticipate Lc0's peak RAM use but there is one and the use declines rapidly with exchanges.

Lc0 RAM use:
4...Nf6 1508 MB
12...Be7 565 MB (queens exchanged, <2000 MB peak prior)
117...Kxf7 210 MB (K vs. K final position)

[Event "?"]
[Site "?"]
[Date "2019.08.15"]
[Round "?"]
[White "Stockfish_19081422c0"]
[Black "Lc0-v0.22.0"]
[Result "1/2-1/2"]
[BlackElo "2200"]
[ECO "B06"]
[Opening "Modern"]
[Time "01:23:45"]
[Variation "3.Nf3 d6"]
[WhiteElo "2000"]
[TimeControl "5400+10"]
[Termination "normal"]
[PlyCount "234"]
[WhiteType "program"]
[BlackType "program"]

1. e4 g6 2. d4 Bg7 {Book.} 3. Nf3 d6 4. Bd3 Nc6 5. c3 e5 6. h3 Nge7 7. dxe5 Nxe5 8.
Nxe5 dxe5 9. Be3 O-O 10. Bc4 b6 11. Qxd8 Rxd8 {Queens exchanged.} 12. Nd2 Bb7 13. f3 Nc8 14.
Bg5 Re8 15. Bb3 Nd6 16. O-O-O Bf8 17. Nc4 Kg7 18. Ne3 Be7 19. Bxe7 Rxe7 20.
Bd5 Rd8 21. g4 c6 22. Bb3 Bc8 23. h4 Red7 24. Kc2 Nb7 25. Rxd7 Rxd7 26. Ra1
Rd8 27. Bc4 Nd6 28. Rd1 Kf8 29. b4 Ke7 30. Bb3 f5 31. exf5 gxf5 32. Re1 Rf8
33. g5 f4 34. Nd1 Nf5 35. Kc1 Kd6 36. Nf2 Be6 37. Ne4+ Ke7 38. h5 Bxb3 39.
axb3 Ke6 40. g6 hxg6 41. Ng5+ Ke7 42. hxg6 Nh4 43. Rh1 Nxg6 44. Rh7+ Kf6
45. Ne4+ Ke6 46. Rh6 Kf5 47. Rh7 Ke6 48. Kc2 Rh8 49. Rxa7 Nh4 50. Nd2 Ng2
51. Rc7 Kd6 52. Rg7 Rh2 53. Ne4+ Ke6 54. Rg6+ Kd7 55. Kd3 Ne1+ 56. Kc4 Nxf3
57. b5 cxb5+ 58. Kd5 Nd2 59. Nxd2 Rxd2+ 60. Kxe5 f3 61. Ke4 f2 62. Rf6 Ke7
63. Rf3 b4 64. cxb4 Rb2 65. Kd4 Ke6 66. Kc3 Ra2 67. Kd3 Ke5 68. b5 Rb2 69.
Rf7 Kd5 70. b4 Ke6 71. Rf8 Ke5 72. Rf7 Ra2 73. Ke3 Kd5 74. Rf4 Rc2 75. Kd3
Ke5 76. Rf7 Rb2 77. Kc3 Re2 78. Kd3 Ra2 79. Ke3 Kd5 80. Rf4 Rc2 81. Rd4+
Ke5 82. Rd1 Rb2 83. Kf3 Rd2 84. Rc1 Rc2 85. Rd1 Rd2 86. Rc1 Kd4 87. Kg2 Ke3
88. Rc6 Rd5 89. Re6+ Kd3 90. Rxb6 Ke3 91. Re6+ Kd4 92. b6 f1=R 93. Kxf1 Rb5
94. Rh6 Kc3 95. Ke1 Kxb4 96. b7 Rxb7 97. Kd2 Kc5 98. Kc3 Rc7 99. Rh4 Ra7
100. Rc4+ Kd5 101. Rb4 Kc5 102. Rh4 Ra1 103. Rh5+ Kc6 104. Rh6+ Kb7 105.
Rh2 Ra4 106. Kd3 Ra3+ 107. Kc4 Ra6 108. Rh7+ Kc8 109. Rh1 Kb7 110. Rb1+ Kc7
111. Rh1 Rb6 112. Rh7+ Kd8 113. Rh8+ Kc7 114. Rh7+ Kb8 115. Rh8+ Ka7 116.
Rh7+ Rb7 117. Rxb7+ Kxb7 {Insufficient material} 1/2-1/2
Parent - By MrKris (***) Date 2019-08-13 08:28
Lc0 NNCashSize 2,000,000 2cpu net T40B.2-106
Allie defaults net 42850
i7-4770k 8GB
Arena demo games (256MB browser on also)
G/5'+2" max. memory 93%
G/15'+6" max. memory 96%
Parent - - By Master Om (Bronze) Date 2019-06-15 12:30
I made a lot of tests: lc0 on a RTX 2080 is roughly as strong as Stockfish 10 running on 10 cores on a modern Intel CPU @ 4 GHz.

Sir what about Lc0 running on GTX 1050 Ti ?
Parent - - By Lukas Cimiotti (Bronze) Date 2019-06-17 11:31
It will work, but performance won't be great.
Parent - By Master Om (Bronze) Date 2019-06-17 17:31
:cry::cry::cry::cry::cry::cry:
- By gsgs (***) Date 2019-08-16 07:27 Upvotes 1
is someone successfully running Lc0 in Linux ?
Up Topic The Rybka Lounge / Computer Chess / LeelaChess crashing in UCI go nodes

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill