Life In 19x19

I just tried it. It seems decent enough, but it's incredibly slow, compared to CS or Igowin. Also, I'm pretty sure it made a rule mistake with a corner ko in a game I just tried (he won, I needed that ko for a liberty, he played at 1-2 and didn't let me take at 1-1, which was supposedly a legal move.)

RBerenguel wrote:
a)Also, a T61 of around 2007 is comparable (likely worse) in crunch power than an iPad Air.

b) << except I see no reason why the algorithm would necessarily be greedy for space
Huh!? Pattern libraries, fuseki libraries and transposition tables. With Pachi you can easily kill a device withh less than 1GB RAM, and the "recommended" setting uses 3GB for inner trees.

a -- really? What would a benchmark show for the CPU of an iPad Air? Keep in mind that for practical speed need to consider cache to ram transfers*. How much CPU cache does the CPU of your iPad air have?

b -- Space is of course relative to the history of computers. When I first was involved with computers (60's)even a few K was a big deal. When I again began working with computers in the late 70's a few M was a big deal. What is normal, our ideas of what "lots of ram" means vs "normal amount of RAM" changes over time.

So my "not greedy" meant "not needing more than could be expected as normal". Normal relative to what a "standard computer" could be expected to have. This really hasn't changed all that much since the days when those T61's were new and most new laptops don't have a lot more than the 4 gigs of my old T61's.

Devices like the iPad are powerful communication devices with reasonable computing power for simple tasks. As long as they are being used for their intended purposes they will seem just as strong as an ordinary laptop or desktop. Small, low power consumption devices, not intended for doing a "crunch". Which is fine, most users, even users of laptops and desktops aren't using them to do "crunches" either. But asking a machine to be a decently strong go opponent is asking it to do a "crunch". For any of the available programs to be playing at a few dan at real time speeds (realistic time controls) you'd want at least a machine with as much CPU as say an i7-3770 or i7-4770 in it. But a machine like that isn't out of the question for a home user.



* I learned that the hard way when after retirement I decided to learn C. After getting a decent grasp of the language I decided to tackle a "tuning" exercise but to my surprise even before trying to tune it was running faster than should have been possible (in terms of memory cycles). Didn't take me long to figure out that because I hadn't made the memory usage too large for this to be running entirely within cache.

Mike Novack wrote: a -- really? What would a benchmark show for the CPU of an iPad Air? Keep in mind that for practical speed need to consider cache to ram transfers*. How much CPU cache does the CPU of your iPad air have?

The A7 CPU has a speed of between 1.4-1.5 Ghz. As much as I can Google the details, so much can anyone else. Here are Anandtech's details. A T61 from around 2007 probably had an Intel Core 2 Duo (according to Lenovo's specs it says Core Duo, but I assume it's a mistake) with between 1.8 and 2.2 Ghz, more likely the former than the latter. According to Lenovo and Intel, it had a 4MB L2 cache and 667-800 MB front bus. An A7 has a very small (64kb instructions and 64kb data, which probably makes branching prediction incredibly fast compared to the Core 2 Duo) L1 cache (which the Core 2 Duo seems to lack,) a 1 MB L2 cache (shared by both computing units) and a 4 MB L3 cache. It may not sound impressive (well, it has L1, which is quite a boost to computing power already) but of course bus technology has changed since 2007: its memory bandwidth is at least 10 GB/s. That's more than 10 times faster than the T61 CPU, and it's not peak speed.

Also, although I guess they are not using it, the A7 has a dedicated GPU. If go software on iOS could take advantage of it for computing something (related to the MCTS algorithms, sure), it could pretty much be faster than non-optimised desktop code.

Mike Novack wrote: b -- Space is of course relative to the history of computers. When I first was involved with computers (60's)even a few K was a big deal. When I again began working with computers in the late 70's a few M was a big deal. What is normal, our ideas of what "lots of ram" means vs "normal amount of RAM" changes over time.

So my "not greedy" meant "not needing more than could be expected as normal". Normal relative to what a "standard computer" could be expected to have. This really hasn't changed all that much since the days when those T61's were new and most new laptops don't have a lot more than the 4 gigs of my old T61's.

Devices like the iPad are powerful communication devices with reasonable computing power for simple tasks. As long as they are being used for their intended purposes they will seem just as strong as an ordinary laptop or desktop. Small, low power consumption devices, not intended for doing a "crunch". Which is fine, most users, even users of laptops and desktops aren't using them to do "crunches" either. But asking a machine to be a decently strong go opponent is asking it to do a "crunch". For any of the available programs to be playing at a few dan at real time speeds (realistic time controls) you'd want at least a machine with as much CPU as say an i7-3770 or i7-4770 in it. But a machine like that isn't out of the question for a home user.

You keep talking about crunch of this and that CPU, where in no moment I estimated Crazy Stone or Igowin's strengths, AT ALL. So, stop nagging about this. If you are really interested in knowing more details about the speed and level of this programs, either get an iSomething (or an Android device, I think Crazy Stone is also available there) or contact the developers. I have exchanged some emails with the people at Unbalance and they are quite nice.

Also, when I started doing numerical analysis stuff at my degree (2001,) my current iPad Air would have blown well past what I was using. Speed/memory are all matters of the current times. As the adage goes, a current smartphone has much more crunch than the Apollo 11. And that all the Apollo 11 control room, for that matter.

haha... a least I won against crazy stones without handicap stones !
ok... I needed 379 score estimation and at least the same amount of undo so maybe that's more crazy stone beating itself... ...
But still... I never did that before what ever I tryed (even with infinite undo )


Usually I still need 5 or 6 stones, but here I was doing some experiment with sansan and I liked the start of the game so I tryed till the end.