Iphone 4

[Swaminathan]

I'm buying Iphone4 - 16gb (I think), I wonder what the strength of Shredder/Fritz/Hiarcs/Glaurung for Iphone4 is on a PC scale?

[Razor]

at 3min/move around 2700 I think for Shredder

[Peter C]

I think Shredder is supposed to be 2700-2800 but I'd guess that's a bit high. HIARCS is by far the strongest of those, but Shredder (I hear) has the best user interface.

Peter

[Swaminathan]

Thanks for the replies, Guys!

[Tord]

I'm buying Iphone4 - 16gb (I think), I wonder what the strength of Shredder/Fritz/Hiarcs/Glaurung for Iphone4 is on a PC scale?

Stockfish 1.8 (the engine used in the current iPhone Glaurung; the new 1.9 hasn't been ported yet) is approximately 15 times faster on a single core of my 2.8 GHz iMac than on my iPhone 4. Assuming that doubling the speed gives about 60-70 Elo points, we have an estimated Elo difference of 234-273 between the iPhone 4 and the i7 iMac. Using the 3129 rating of SF 1.8 1 CPU on the CCRL list, we get an estimated CCRL rating of 2856-2895 for SF 1.8 on the iPhone 4.

Of course, whether the "double speed = 60-70 Elo points" rule of thumb remains valid over such a big speed differential is far from clear. On the previous iPhones -- which had much slower CPUs -- the difference in strength was certainly a lot bigger than you would guess just based on the speed. Stockfish's selective search simply falls apart tactically at such slow hardware, and it missed cheap tactical shots all the time. On the iPhone 4, it's a very different beast, and feels much more similar to Stockfish on a real computer. The 2856-2895 estimate give above is almost certainly too high, but I have no idea by how much.

I suppose I should run some tests against a few weaker PC engines some time soon.

[Swaminathan]

I'm buying Iphone4 - 16gb (I think), I wonder what the strength of Shredder/Fritz/Hiarcs/Glaurung for Iphone4 is on a PC scale?

Stockfish 1.8 (the engine used in the current iPhone Glaurung; the new 1.9 hasn't been ported yet) is approximately 15 times faster on a single core of my 2.8 GHz iMac than on my iPhone 4. Assuming that doubling the speed gives about 60-70 Elo points, we have an estimated Elo difference of 234-273 between the iPhone 4 and the i7 iMac. Using the 3129 rating of SF 1.8 1 CPU on the CCRL list, we get an estimated CCRL rating of 2856-2895 for SF 1.8 on the iPhone 4.

Of course, whether the "double speed = 60-70 Elo points" rule of thumb remains valid over such a big speed differential is far from clear. On the previous iPhones -- which had much slower CPUs -- the difference in strength was certainly a lot bigger than you would guess just based on the speed. Stockfish's selective search simply falls apart tactically at such slow hardware, and it missed cheap tactical shots all the time. On the iPhone 4, it's a very different beast, and feels much more similar to Stockfish on a real computer. The 2856-2895 estimate give above is almost certainly too high, but I have no idea by how much.

I suppose I should run some tests against a few weaker PC engines some time soon.

Thanks for the detailed reply, Tord!

You've educated me somewhat on this. I had no clue at all about the strength and am happy to read that engines could play at such high levels even on a mere 800-1000 mhz i-iphone 4! When I find time, I'd be glad to run a lot of tests of I-phone Glaurung against PC Engines of 2600 and above. I'd rather play a lot of games with chess program on i-phone than spend time playing online!

Btw, I'd hope that there will be automated EPD/PGN testing option in I-phone glaurung GUI! As of late, I have not done any serious tests on engine-engine matches. Mostly it's test suites -- I'm more delighted to see which puzzle(s) the engine missed/solved. This in turn improves my positional chess skills too.

Thanks a lot, Tord!!

Best wishes,
Swami

[Tord]

Thanks for the detailed reply, Tord!

You've educated me somewhat on this. I had no clue at all about the strength and am happy to read that engines could play at such high levels even on a mere 800-1000 mhz i-iphone 4!

Please be cautious in how to interpret my message: The rating numbers I mentioned should be considered an upper bound for the actual strength. The actual strength is almost certainly a lot lower, but I don't have any data whatsoever to help me give an estimate of how much lower. It feels very strong to me, but I'm not strong enough to tell the difference between a 2500 rated and a 2800 rated chess player.

When I find time, I'd be glad to run a lot of tests of I-phone Glaurung against PC Engines of 2600 and above.

Unfortunately, running such tests isn't very convenient for average users, because there is no straightforward way to run automated matches. The way I have done it in the past is to jailbreak my phone, install a command line UCI engine on the phone, and run this engine from a GUI running on my Mac. I can't even do this myself at the moment, because there still isn't a way to jailbreak iOS 4.1.

Btw, I'd hope that there will be automated EPD/PGN testing option in I-phone glaurung GUI!

I'm afraid that is extremely unlikely to ever happen. Trying to pack too many features into an iPhone program inevitably leads to a user interface disaster, so I have to prioritize very strictly and implement only the most popular and commonly requested features. Test suites are, I'm sorry to say, a feature that would only be of value to a very tiny sub-set of users.

Moreover, it's extremely easy to use a desktop computer to simulate running a testsuite on the iPhone. Just run a benchmark on both devices (the phone and the computer) first and measure the difference in speed. In my case, as I wrote in my previous message, the computer is 15 times faster. Then run the test suite on the computer with a single engine thread, 8 MB hash tables, and the thinking time per position divided by 15.

[Tord]

In order to get a first rough estimate, I've now started two time handicap matches on my iMac, one with SF 1.8 vs Kiwi 0.6d (2526 on the CCRL list), the other with SF 1.8 vs Glaurung 1.2.1 (2771 on the CCRL list). In both matches, SF has 20 seconds per game, and the opponent 5 minutes per game. All engines are running with a single thread. Because 20 seconds equals 5 minutes / 15 and one CPU core of the iMac is 15 times faster than the iPhone, SF in these matches should be approximately as strong as Glaurung on the iPhone 4.

Results tomorrow.

[Swaminathan]

Moreover, it's extremely easy to use a desktop computer to simulate running a testsuite on the iPhone. Just run a benchmark on both devices (the phone and the computer) first and measure the difference in speed. In my case, as I wrote in my previous message, the computer is 15 times faster. Then run the test suite on the computer with a single engine thread, 8 MB hash tables, and the thinking time per position divided by 15.

That's a good idea, Thanks Tord!

[Swaminathan]

In order to get a first rough estimate, I've now started two time handicap matches on my iMac, one with SF 1.8 vs Kiwi 0.6d (2526 on the CCRL list), the other with SF 1.8 vs Glaurung 1.2.1 (2771 on the CCRL list). In both matches, SF has 20 seconds per game, and the opponent 5 minutes per game. All engines are running with a single thread. Because 20 seconds equals 5 minutes / 15 and one CPU core of the iMac is 15 times faster than the iPhone, SF in these matches should be approximately as strong as Glaurung on the iPhone 4.

Results tomorrow.

That'd be an interesting match!

I'm looking forward to it. Am curious enough too.