[QUOTE=wblipp;217249]This factor is composite. It is the product of the two smallest already known factors, 6643859687 * 3195696508967[/QUOTE]

Thank you William, We should be more careful while testing... :redface:

Luigi

Taking 3321928171 from 78 bits to 79 bits
Taking 3321928697 from 75 bits to 79 bits
Taking 3321932597 from 75 bits to 79 bits

Luigi

ET_, 3321932597 was factored, but you had a mis-match on your site where you listed the factor but did not show it as factored on this page:
[url]http://www.moregimps.it/billion/expo_t.php[/url]

(It also kinda broke my code, but I've made it a little more robust now.)

Also, way to go wblipp for catching that composite factor. It's not something I'd even think to check for with these numbers.

[QUOTE=lavalamp;217276]ET_, 3321932597 was factored, but you had a mis-match on your site where you listed the factor but did not show it as factored on this page:
[url]http://www.moregimps.it/billion/expo_t.php[/url]

(It also kinda broke my code, but I've made it a little more robust now.)

Also, way to go wblipp for catching that composite factor. It's not something I'd even think to check for with these numbers.[/QUOTE]

Good catch, thanks. Shats always hippen in a row. :smile:

[COLOR="Red"][SIZE="3"]Cancelling[/SIZE][/COLOR] 3321932597 from 75 bits to 79 bits
[COLOR="Lime"][SIZE="3"]Taking[/SIZE][/COLOR] 3321931363 from 75 bits to 79 bits

Luigi

Since the recent pace of updates has been rather brisk, I've often been quite slow in updating the offline versions of the graph and the comparison site.

Therefore I've changed things up a bit, from now on when anyone goes to either of these URLs:
[url]http://2721.hddkillers.com/graph/[/url]
[url]http://2721.hddkillers.com/compare/[/url]

They will automatically load the newest data from the Eleven Smooth and MoreGIMPS websites. That data will then be cached and will be used for all future queries in the following hour.

The /live/ URLs will still work, but they'll take you to the exact same page as if you hadn't included it. At most the data will only be an hour out of date so I figure it's not a problem.

Taking
3321928699
3321928777
3321928927
3321928963
3321928999
3321929041

from 2^75 to 2^79. :smile:

Oliver

[quote=TheJudger;217402]Taking
3321928699
3321928777
3321928927
3321928963
3321928999
3321929041

from 2^75 to 2^79. :smile:

Oliver[/quote]

Ohhh, the project will be at Level 13 at no time... :smile::smile:

I'll do 3321929053, 3321929059 and 3321929113 next.
I'll try to bring them up to 2^80! :smile:

And I will take 3321931363 to 80... :smile:

Luigi

I'll take 3321929173, 3321929179 and 3321929197 to 2^80.

Hi,

[QUOTE=lavalamp;214377]Expanding on this post I made a while back:
[url]http://www.mersenneforum.org/showpost.php?p=176635&postcount=21[/url]

Here are some performance figures for various high end graphics cards (all units are GFLOPs). I have coloured pairs of cards as there are comparable in terms of the role they have in each companies line-up:[code]

Card SP DP

GTX 275 673.92 84.24
[color=green]GTX 285 708.48 88.56[/color]
[color=blue]GTX 295 1192.32 149.04[/color]
[color=red]GTX 480 1344.96 168.12[/color]

[color=blue]HD 4870 X2 2400 480[/color]
[color=green]HD 4890 1360 272[/color]
[color=red]HD 5870 2720 544[/color]
HD 5970 4640 928

[color=green]Previous generations highest spec single GPU card.[/color]
[color=blue]Previous generations highest spec dual GPU card.[/color]
[color=red]Current generations highest spec single GPU card.[/color]
nVidia does not yet have a dual GPU 400 series card to compete with the 5970.[/code]To work out the performance of any recent nVidia or ATI graphics card, here is the formula:
SP performance = shaders (aka cores) * frequency / 500

To get the DP performance, divide by 8 for nVidia GPUs, and divide by 5 for ATI GPUs. Note that for dual GPU graphics cards such as the 5970, there are 1600 cores per GPU, so it is usually listed as 3200 cores.

Hands down the ATI cards take the performance crowns for single and dual precision performance.[/QUOTE]

[B]So I think I'm biased towards Nvidia so take the following with care![/B]

32bit integer addition with carry:
Nvidia: one instruction (one clock(?)), 3 registers (2 input, 1 output)
ATI: two instructions (one clock each?), 4 registers (2 input, 2 output)

96 bit "long" integer addition (using 3 32bit integers):
Nvidia:[CODE]
res.lo = __add_cc (a.lo, b.lo);
res.med = __addc_cc(a.med, b.med);
res.hi = __addc (a.hi, b.hi);
[/CODE]
[B]This actually works in my code.[/B]

ATI:[CODE]
carry.lo = ADDC_UINT(a.lo, b.lo);
res.lo = ADD_INT (a.lo, b.lo);

carry.med = ADDC_UINT(a.med, b.med);
res.med = ADD_INT (a.med, b.med);
carry.med += ADDC_UINT(res.med, carry.lo);
res.med = ADD_INT (res.med, carry.lo);

res.hi = ADD_INT (a.hi, b.hi);
res.hi = ADD_INT (res.hi, carry.med);
[/CODE]
[B]I haven't tested this, I've read 'AMD_Evergreen-Family_ISA_Instructions_and_Microcode.pdf'[/B]

I count 3 vs. 9 instructions!

Integer multiplication is similar to Nvidia, they have a fast 24bit integer multiply and a (slow?) 32bit integer multiply.
Above 2^71 (factor size) a GTX 480 a little bit more than 3 times faster compared to a GTX 275. This is related to the 32bit multiply which is fast on Fermi. :smile:

My Opinion: ATI has more but simpler compute cores!


Oliver