I have done some optimization on the CPU apps that increase the speed by about 50% in my tests. Credit calculation will stay the same, so this should equate to about 50% more credit per unit time on CPUs since you will now process them faster.

If this app runs without trouble in beta, then it will be promoted to be the default application for everyone.

all CPU apps have the new beta version available. If you would like to try out the new application, enable "Run test applications" from your Project preferences

edit:
v1.06 now released as beta for all platforms.
expect ~50-60% faster on CPU
expect ~25-100% faster depending on WU and GPU.

Ian

Up to now the 1.05 version seems ok on my Win11 pc...

thanks for the report. I'll check the statics from all systems and all platforms in a day or two before calling it good. I might have another small update to test in beta before releasing the default version.

i will have new GPU versions available soon also. 25-75% faster depending on GPU.

Yeah here too. Thanks for your effort, this speedincreases are phenomenal ^^

---

---------- 24/7 Crunching since 2011 -----------
DSKAG Austria: http://www.dskag.at

i will have new GPU versions available soon also. 25-75% faster depending on GPU.

Great news.
Would it be technically possible to create bundled tasks, as I think E@H does with BRP7, especially for more powerful cards?
GPU tasks are already quite short. Even on a low-end card like RTX 3050, they take ~7 minutes on average with the current app. Judging by what you’ve achieved with apps on Einstein@Home, that 25-75% is not your (Claude’s) final word.

In reply to makracz's message of 10 Nov 2026:

i will have new GPU versions available soon also. 25-75% faster depending on GPU.

Great news.
Would it be technically possible to create bundled tasks, as I think E@H does with BRP7, especially for more powerful cards?
GPU tasks are already quite short. Even on a low-end card like RTX 3050, they take ~7 minutes on average with the current app. Judging by what you’ve achieved with apps on Einstein@Home, that 25-75% is not your (Claude’s) final word.

just a small correction first. I do not believe Einstein BRP7 is batched like this. at least not in a way that was changed since its initial release. you might be remembering that they did this with BRP4. the BRP4"G" tasks were the batched ones. i think like 5x or 8x bundled BRP4 tasks.

But for here, I have thought about it. I think it should be possible, but the results would not be comparable and wouldn't be able to directly validate against a non-batched task, which changes some of the "purity" of the project. and I wouldn't want to batch CPU jobs since some of them can run very long already. which puts me in a position where I have one sub-project running without CPU cross validation, which helps me prove that all devices are truly producing the same correct result. I have to think about if I trust the apps enough to spin off a sub-project that only GPUs contribute to, and then think about the WU generator logic and the validator logic. and then decompressing those results to fill in the database where they need to go (and prevent the single-WU tasks from being send in tandem)

If I can do it, I'd like to though.

FYI, the new v1.06 apps are now posted to beta for all platforms. I'll monitor the results and if they look good after a few days will just promote everything to the new app

1.06 looks good at first glance; no bugs so far.
Linux, NV 5080
https://kesslerathome.org/results.php?hostid=547

In reply to DF1DX's message of 10 Nov 2026:

1.06 looks good at first glance; no bugs so far.
Linux, NV 5080
https://kesslerathome.org/results.php?hostid=547

Looks good also on my WIndows machine, with an AMD gpu

I love short tasks. That motivates me to add slow machines and old GPUs as well.