[QUOTE=LaurV;356172]BTW, a supermod could mask the residue two posts above... Just in case...

@Robish: please do not post full residues of un-verified LL tests, some "credit hunters" will be tempted to "verify" them using your residue, and 20 years later we may find out we missed a prime, in case that residue is wrong.[/QUOTE]


Ok will (not) do ;-)

[QUOTE=nucleon;355457]Aaah, the power of 2 FFT issue.

Ok, next time I get a chance to experiment I'll play with something more suitable.

-- Craig[/QUOTE]

Ok, only a fortnight late. 7990@1GHz/1.5GHz mem clk

GPU0:
[CODE]
Iteration 10000 M( 38000009 )C, 0xfd9116e3760e4571, n = 2097152, clLucas v1.01 err = 0.1406 (0:39 real, 3.8668 ms/iter, ETA 40:48:19)
Iteration 20000 M( 38000009 )C, 0xd91fe21f272e5099, n = 2097152, clLucas v1.01 err = 0.1406 (0:39 real, 3.8636 ms/iter, ETA 40:45:38)
Iteration 30000 M( 38000009 )C, 0x7ec1302ed5173c26, n = 2097152, clLucas v1.01 err = 0.1406 (0:39 real, 3.8766 ms/iter, ETA 40:53:16)[/CODE]


GPU1:

[CODE]Iteration 10000 M( 38000009 )C, 0xfd9116e3760e4571, n = 2097152, clLucas v1.01 err = 0.1406 (0:38 real, 3.8664 ms/iter, ETA 40:48:05)
Iteration 20000 M( 38000009 )C, 0xd91fe21f272e5099, n = 2097152, clLucas v1.01 err = 0.1406 (0:39 real, 3.8693 ms/iter, ETA 40:49:15)
Iteration 30000 M( 38000009 )C, 0x7ec1302ed5173c26, n = 2097152, clLucas v1.01 err = 0.1406 (0:39 real, 3.8728 ms/iter, ETA 40:50:51)
[/CODE]

vs Titan:
[CODE]Iteration 10000 M( 38000009 )C, 0xfd9116e3760e4571, n = 2097152, CUDALucas v2.03 err = 0.1953 (0:18 real, 1.8289 ms/iter, ETA 19:17:59)
Iteration 20000 M( 38000009 )C, 0xd91fe21f272e5099, n = 2097152, CUDALucas v2.03 err = 0.1953 (0:18 real, 1.8124 ms/iter, ETA 19:07:16)
Iteration 30000 M( 38000009 )C, 0x7ec1302ed5173c26, n = 2097152, CUDALucas v2.03 err = 0.2031 (0:18 real, 1.8168 ms/iter, ETA 19:09:45)
[/CODE]
Titan is more efficient 250W vs 375W, but I guess th code for opencl is still early days.

-- Craig

[QUOTE=nucleon;356678]Ok, only a fortnight late. 7990@1GHz/1.5GHz mem clk

GPU0:
[CODE]
Iteration 10000 M( 38000009 )C, 0xfd9116e3760e4571, n = 2097152, clLucas v1.01 err = 0.1406 (0:39 real, 3.8668 ms/iter, ETA 40:48:19)
Iteration 20000 M( 38000009 )C, 0xd91fe21f272e5099, n = 2097152, clLucas v1.01 err = 0.1406 (0:39 real, 3.8636 ms/iter, ETA 40:45:38)
Iteration 30000 M( 38000009 )C, 0x7ec1302ed5173c26, n = 2097152, clLucas v1.01 err = 0.1406 (0:39 real, 3.8766 ms/iter, ETA 40:53:16)[/QUOTE]
Based on that data, R9 290X may very roughly yield from 3.514 to 2.811 ms/LL iter for that exp using that FFT (using default and boost clocks respectively).

Perhaps. In gaming, it is neck and neck with the titan based on early benchies from Tom's and Anandtech, But compute will be what matters... :smile: Also, it is the first gpu to have a [I]512[/I] [I]bit[/I] memory bus.

I get:

GTX 590 board:
[CODE]
Starting M38000009 fft length = 2097152
Iteration 10000 M( 38000009 )C, 0xfd9116e3760e4571, n = 2097152, CUDALucas v2.03
err = 0.1914 (0:40 real, 3.9348 ms/iter, ETA 41:31:21)
Iteration 20000 M( 38000009 )C, 0xd91fe21f272e5099, n = 2097152, CUDALucas v2.03
err = 0.1914 (0:38 real, 3.8383 ms/iter, ETA 40:29:37)
Iteration 30000 M( 38000009 )C, 0x7ec1302ed5173c26, n = 2097152, CUDALucas v2.03
err = 0.1914 (0:39 real, 3.8376 ms/iter, ETA 40:28:33)
[/CODE]

GTX Titan:
[CODE]
Iteration 10000 M( 38000009 )C, 0xfd9116e3760e4571, n = 2097152, CUDALucas v2.03
err = 0.2090 (0:19 real, 1.8580 ms/iter, ETA 19:36:26)
Iteration 20000 M( 38000009 )C, 0xd91fe21f272e5099, n = 2097152, CUDALucas v2.03
err = 0.2090 (0:17 real, 1.7758 ms/iter, ETA 18:44:04)
Iteration 30000 M( 38000009 )C, 0x7ec1302ed5173c26, n = 2097152, CUDALucas v2.03
err = 0.2090 (0:18 real, 1.7750 ms/iter, ETA 18:43:17)
Iteration 40000 M( 38000009 )C, 0x91312c6488821eac, n = 2097152, CUDALucas v2.03
err = 0.2090 (0:18 real, 1.7756 ms/iter, ETA 18:43:23)
[/CODE]

Interesting. So I guess 590 and 7990 almost tie.

Just for the sake of curiosity, I'm curious on titan timings [I]without [/I]the DP switch.

[QUOTE=kracker;356696]Interesting. So I guess 590 and 7990 almost tie.

Just for the sake of curiosity, I'm curious on titan timings [I]without [/I]the DP switch.[/QUOTE]

How do you get that kind of results without the DP switch ?
Mine switch is on. just logged into the node to check the setting.

sorry for hijacking cl thread.. but when I calculate, I get 58 ghz days LL testing with one titan per day with these timings on 73m exponents. It equals the statistics for that card, mabye a little better, but then I tweaked a little..

I updated the titan thread with some timings for the titan, we dont need to discuss that here in LL thread :)

Some timings for no reason really :razz:
2M FFT
HD-
7750: 15 ms
7770: 12 ms
7850: 8? ms
7870: 7 ms
7950: 4.2 ms
7970: 3.7 ms
7990: 3.9 ms(x2)

[QUOTE=kracker;356864]Some timings for no reason really :razz:
2M FFT
HD-
7750: 15 ms
7770: 12 ms
7850: 8? ms
7870: 7 ms
7950: 4.2 ms
7970: 3.7 ms
7990: 3.9 ms(x2)[/QUOTE]

Very interesting actually.
It kind of shows us where to exptect R9 290x.. I guess 2.5-2.8ms with the same FFT length test ?

[QUOTE=kracker;356864]Some timings for no reason really :razz:
2M FFT
HD-
7750: 15 ms
7770: 12 ms
7850: 8? ms
7870: 7 ms
7950: 4.2 ms
7970: 3.7 ms
7990: 3.9 ms(x2)[/QUOTE]
Confirm for 7990 and 7970, with about 5% longer for me (as I know, you push the memory clock, which I can't do, the cards seems not to be so stable, they are air cooled, and here is still hot during the day, when nobody home and no aircond running).