[QUOTE=science_man_88;410928]I saw 1.9 Mbps at the start not much after that[/QUOTE]

Care to support that with empirical data? How many bytes were involved? The metric you gave was bits per second, not the quantity of data passed. And you didn't include the sample period.

Perhaps you might consider including an attachment which contained the data you captured? Temporal period would be useful.

[QUOTE=chalsall;410929]Care to support that with empirical data? How many bytes were involved? The metric you gave was bits per second, not the quantity of data passed. And you didn't include the sample period.

Perhaps you might consider including an attachment which contained the data you captured? Temporal period would be useful.[/QUOTE]

I made a video but I can't upload it I don't even think the filetype of the video my tablet makes is supported. edit: I'd say maybe a couple seconds.

[QUOTE=science_man_88;410930]I made a video but I can't upload it I don't even think the filetype of the video my tablet makes is supported. edit: I'd say maybe a couple seconds.[/QUOTE]

Please forgive me for this, but I don't care about videos.

When I ask my men questions, I expect answers. Immediately.

Deal with it.

[QUOTE=chalsall;410933]Please forgive me for this, but I don't care about videos.

When I ask my men questions, I expect answers. Immediately.

Deal with it.[/QUOTE]

Well he's not one of your men, is he? This is merely a forum, everyone here is welcome to share the observations.

[QUOTE=Dubslow;410950]Well he's not one of your men, is he? This is merely a forum, everyone here is welcome to share the observations.[/QUOTE]

You are right; that was inappropriate. My apologies.

How interested are we in this? Where I come from source code just means C/C++/C# or some manner of code fiddled with to be easier for humans to comprehend, machine code translates to assembly pretty readily and with that I have all the source code I need to read what's happening right in his compiled program. I would have to read the license first to see if reading machine code is prohibited :) Similarly the SSH keys can be readily extracted from a memory snapshot of the program to decrypt the traffic....

If someone wants to PM me the download link I am happy to investigate further. I do also happen to have 5 GPUs...

I will make it no secret that I too am interested in the large Mersenne prizes. That would be winning for the sake of winning vs. for the profit - any of us will certainly be in the red when this is said and done even with clever accounting. I too have looked at special modifications for highly-parallel algorithms for LL and TF running on multiple GPUs in the same host system. There are areas for improvement but I would be very surprised to see several orders of magnitude as claimed here.

To those nitpicking OP for missing various optimizations to the TF code or so, would you forgive someone who missed a 20-30% speedup in the math but found improvements to cache or memory access friendly GPU utilization that resulted in an advantage several orders of magnitude in size? We all have different skills in different areas. I would guess the next big breakthrough we see will be either drastically new math rendering full LL tests unneeded, or someone with serious background in GPU computing looking at the algorithm implementation in a whole new way. The cache/memory architecture on GPUs is very different from CPUs and we may find vanilla FFT based algorithms are not as well suited as something that may appear less efficient in big O.

I do believe ambitious newcomers can often see angles and find interesting solutions and improvements that entrenched parties miss. An abrasive ego often comes alongside those creative personalities, however the poster here has done very little to encourage belief other than his feigned outrage at criticisms and skeptics. In my experience those who make the largest impact are those who listen to criticism and grow from it instead of being outraged by it. That said, many of the best doctors have poor bedside manner.

[QUOTE=airsquirrels;411436]How interested are we in this?[/QUOTE]
Don't waste your time. If you want to play, then take cudaLucas, to which you have sources and everything (and it is bloody faster) and try to make it use more GPUs for same iteration. Transferring data over... (what? SLI? grrr). Most probably you won't be able to make it faster, it already has lots of internal optimizations within its belly. But you can try.
[QUOTE=airsquirrels;411436]I do believe ambitious newcomers can often see angles and find interesting solutions and improvements that entrenched parties miss. An abrasive ego often comes alongside those creative personalities, however the poster here has done very little to encourage belief other than his feigned outrage at criticisms and skeptics. In my experience those who make the largest impact are those who listen to criticism and grow from it instead of being outraged by it. That said, many of the best doctors have poor bedside manner.[/QUOTE]
:goodposting: :tu: this is very well said!

[QUOTE=airsquirrels;411436]How interested are we in this?[/quote]
Not sure. There is insufficient information to determine if we should be interested.

[QUOTE=airsquirrels;411436]There are areas for improvement but I would be very surprised to see several orders of magnitude as claimed here.[/quote]
Well, see, that's the thing. No claim has been made regarding the performance of the program. We still don't how much time the program will take to complete a large LL test. The OP has given no indication. Perhaps you can run it on your 5 GPU setup and let us know the ETA? [Link to link (nvidia dev forum) available up-thread] As it is, we have seen TF performance and it is *terrible*


[QUOTE=airsquirrels;411436]To those nitpicking OP for missing various optimizations to the TF code or so[/quote]
Pardon me, but this goes directly to the issue of competence of the programmer. If such basic things are so off, we should be *highly* skeptical of any and all claims.

[QUOTE=airsquirrels;411436]I do believe ambitious newcomers can often see angles and find interesting solutions and improvements that entrenched parties miss. An abrasive ego often comes alongside those creative personalities, however the poster here has done very little to encourage belief other than his feigned outrage at criticisms and skeptics. In my experience those who make the largest impact are those who listen to criticism and grow from it instead of being outraged by it. That said, many of the best doctors have poor bedside manner.[/QUOTE]
We are very open-minded here - newcomers, oldcomers, everyone gets the benefit of doubt. But vague claims are not sufficient. There have been lot of reasonable questions in this thread that the OP has not at all attempted to answer, none of which, btw, requires him to reveal his secret sauce. :smile:

[QUOTE=airsquirrels;411436]

If someone wants to PM me the download link I am happy to investigate further. I do also happen to have 5 GPUs...

[/QUOTE]

The download link can be found in the NVidia forum linked from here in an earlier posting. We are all dying to hear what happens if you install the software on such a machine. I assume that someone with that much firepower is security aware enough not to install this software on a production platform though.

[QUOTE=tha;411474]I assume that someone with that much firepower is security aware enough not to install this software on a production platform though.[/QUOTE]

I keep some freshly imaged bootable flash drives around for just such an occasion. I'm traveling the next few days but when I get back I'll investigate. Who knows maybe I'll find the first 100M prime ;-)

If that is your goal, I really think that running CudaLucas and Prime95 gives you better chances to find them faster ;)
How long has the author not replied? Too long. He recognized that he is not getting the response he thought in this forum.