mersenne.ca

[chris2be8]

Originally Posted by James Heinrich

However, if we go on the assumption that computing power doubles every 2 years then my rough estimate is that by 2050 we would have a yearly throughput of 6 trillion GHz-days and the project would be complete.

But that would need transistors much smaller than a silicon atom. In practice that won't happen.

In any case the ROT that computing power doubles every 2 years was only valid before about 2005. Then CPUs reached the thermal limit on how fast they could run without overheating. Clock rates havn't increased much since then. The number of cores on a chip has gone up, but at a slower rate. And that can't go on for ever.

[storm5510]

Originally Posted by chris2be8

...In any case the ROT that computing power doubles every 2 years was only valid before about 2005. Then CPUs reached the thermal limit on how fast they could run without overheating. Clock rates havn't increased much since then. The number of cores on a chip has gone up, but at a slower rate. And that can't go on for ever.

I have noticed that Prime95 responds more to speed than anything else. With speed comes heat. So, they put more cores on a chip and drop the clock speed. In my eye, this is sort of self-defeating. Prime95 will spread the load across all threads in one way or another. Using the affinity options can change this behavior, but only so much. More modern processors will throttle down if there is a light load on each thread. Something rated for 3 GHz, for example, may drop back to half of that.

In my case, I will only run something on a CPU which I cannot run on a GPU properly.

[James Heinrich]

Originally Posted by storm5510

More modern processors will throttle down if there is a light load on each thread. Something rated for 3 GHz, for example, may drop back to half of that.

Then your cooling is poor. My 7950X is rated at 4.5GHz base, 5.7GHz boost; if I run a full 16 threads across 16 cores it's rare that it would drop below 5.0GHz, I've not seen any core drop below 4.9 no matter the workload.

[Andrew Usher]

The mfactor upload worked this time; I believe the file didn't upload last time, having now seen the acknowledgement obtained when it does. I can hardly be blamed for assuming that it was discarded, though, as that's what your site says should happen. I guess that has also changed.

With regard to computing power, of course I agree that rates of improvement will continue slowing; but it's worth repeating that GIMPS throughput depends more on the number of users than anything else - if every PC in the world ran Prime95, we could have about that throughput now, from CPUs only, even.

[storm5510]

Originally Posted by James Heinrich

Then your cooling is poor. My 7950X is rated at 4.5GHz base, 5.7GHz boost; if I run a full 16 threads across 16 cores it's rare that it would drop below 5.0GHz, I've not seen any core drop below 4.9 no matter the workload.

My CPU cooling is not the best, IMO. Just an air/fan heat-sink.

I am not sure which OS you are referring to. I use dialogs in Windows 10 Control Panel to regulate mine, when needed. I can't run Prime95 with it set to 100%. I have seen the CPU approach 90°C. I drop it back to 99% the temperature will fall to 70°C. This is a clock speed change from 4GHz to 3.6Ghz. The default minimum, in my case, is 5%. This is probably and idle speed setting, no processes running. 99% is the default setting. 100% is considered to be a "turbo" mode.

Everything I have is long-in-the-tooth as some here like to say. I have to live with it and make do.

[kar_bon]

On your new Fermat page you give as source an older version of W.Kellers's page (last updated in 2001), a new (current) version can be found here.

[James Heinrich]

Originally Posted by kar_bon

On your new Fermat page you give as source an older version of W.Kellers's page (last updated in 2001), a new (current) version can be found here.

Thanks. That's actually the page I got the data from, but copied the wrong link from Google when I went searching for it again.

[cxc]

One small addition for your new Fermat page: G. B. Gostin discovered his 89th Fermat factor on 31 May 2023.

321941255325 · 2^1788 + 1 = 5,615,764,412 … 523,201 (550 digits) | F1784

Sod’s law: The moment you publish something, it’s either out of date or contains a typo.

[James Heinrich]

Originally Posted by cxc

One small addition for your new Fermat page: G. B. Gostin discovered his 89th Fermat factor on 31 May 2023.

Thanks, added.