I just started running HADSM3 Mid-Holocene, and noticed there are a total of 7 people running the exact same simulation, with the possibility of up to 9 \"initial replications\".

Frankly, I feel I am sort of wasting my time if so many others are running the exact same work unit. My time would be better spent perhaps on projects that require perhaps 2 replications, and there are some that require no replications. This seems to me to be so much more efficient.

The HADSM3 Mid-Holocene unit will take approximately 3 weeks. If I am but one of 9 identical results, that seems horribly inefficient for the use of my computer, electricity, and time.

Thank you.

---

Regards,
Bob P.

This has all been discussed before, and you are NOT wasting your time, because the models are NOT \"all the exact same\", only the initial data sets.

By the time the models have finished being created, they will all be slightly different because of differences in the cpu of each computer. This is most noticeable in the difference between Intel and AMD, but also between 32 bit cpus and 64 bit cpus.

And not all of the initial starters will finish the run.

Every single result is put on Milo\'s Results website and made available to climate researchers worldwide.

---

Cpdn news

This has all been discussed before, and you are NOT wasting your time, because the models are NOT \"all the exact same\", only the initial data sets.

By the time the models have finished being created, they will all be slightly different because of differences in the cpu of each computer. This is most noticeable in the difference between Intel and AMD, but also between 32 bit cpus and 64 bit cpus.

And not all of the initial starters will finish the run.

Thank you, that is very interesting! The individual computer adds an element of randomness, if you will, to come up with a slightly different probabilistic result. And interesting that that result is useful as well, because so many other projects require a deterministic answer that is able to be duplicated \"closely enough\" by another contributor.

---

Regards,
Bob P.

If you look at the graphs of completed models from the same workunit it\'s usually possible to detect slight differences just from a visual comparison. It\'s best to open two or more windows to be able to compare them side by side.

AMD and Intel processors handle the computations differently in any case so are unlikely to produce identical results.

We know from what the people in Oxford said quite a while ago that the only computers likely to produce invalid results are those that are unstably overclocked - even when their models manage to complete.

---

Cpdn news