Can I say that the benchmark which best reflects processors performance on rosetta@home is wheatstone (arithmetic floating point operations)?

If not, which one could lead to the best evaluation/guess of processors performance on rosetta@home?

Thanks!

---

AFAIK... BOINC benchmarks aren't really a good benchmark. AMD usually outperforms Intel on these benchmarks.

At least, that's how it was with older BOINC version. I don't know if they have fixed that.

---

I was not referring to the Boinc benchmarks. I was talking about sisoft sandra's benchmarks.

---

I was not referring to the Boinc benchmarks. I was talking about sisoft sandra's benchmarks.

Oh.
Well, CPU's with larger L2 cache memory usually generate more credits. A fast AMD with 1MB L2 Cache will create less credits to a relatively slower Intel with a 4MB L2 Cache.

---

Are you sure about that? Can you show me some facts?

Ok, as you say, cache memory should be an indicator of processor performance in rosetta, but what about benchmarks? Do you know about any which could be a good indicator to compare processors performance in rosetta without having to test them directly with rosetta?

---

Are you sure about that? Can you show me some facts?

Ok, as you say, cache memory should be an indicator of processor performance in rosetta, but what about benchmarks? Do you know about any which could be a good indicator to compare processors performance in rosetta without having to test them directly with rosetta?

You would know if you browse around this forum a bit longer...

Like I said, Rosetta heavily relies on L2 cache memory (not sure why exactly, but I think it's a good bet to say that it's because the models take up a lot of memory, thus storing parts of the models in L2 cache GREATLY reduces lag time that would happen if storing the model on RAM). So if you are looking for a PC-dedicated CPU, look for a CPU that has a large L2 cache size. (Right now, Intel has AMD in this department)

I'm not sure what you mean about "benchmarks"... when Rosetta doesn't use "benchmarks"... Rosetta gives you credit for the number of models you predict per workunit.

Hopefully that answers your questions.

---

I'm not sure what you mean about "benchmarks"... when Rosetta doesn't use "benchmarks"... Rosetta gives you credit for the number of models you predict per workunit.

...they are looking for a task to run, the results of which will (to the extent possible) accurately predict how efficient a given machine will be for running Rosetta tasks.

If a floating point intensive benchmark were available, that had a large enough application and data working set that it stresses the L2 cache... that would probably be about as close as you will get. The problem is that most benchmarks are designed to measure raw processor capability. And so they don't stress the L2, nor the bus.

Even if you were to create a benchmark of actual Rosetta work, I believe you would find that there is significant variation across the various types of tasks and proteins being studied. And so even a benchmark of actual Rosetta work will not be a perfect predictor.

---

Rosetta Moderator: Mod.Sense