Which one is better crunching that using a server with 2ghz 16 threads (can ECC Reg memory help on this?) or using regular desktop pc with 4ghz 8threads (can higher memory and bus help?) ?

---

I've moved Orgil's thread from the Windows board. More people will see it here on the Number Crunching board.

Threads is not as significant as CPU cores, or virtual CPUs cores available. And cores available is only helpful if there is enough memory to accommodate their needs without swapping. High memory/bus bandwidth can definitely help.

If clock rates and number of cores were the only difference, (i.e. low page faults on both, one thread per core on both, similar CPU architectures) I'd expect the performance results to be essentially identical.

[edit] Another key factor can be the amount of L2 cache available, so I'd have to assume L2 cache hit ratio being similar between the two as well. This may not be achievable in real life because the amount of L2 cache and how it ends up being used between a 8 core and 16 core machine will likely be rather different.

---

Rosetta Moderator: Mod.Sense

Orgil -

Since you are talking 16 threads I am going to jump to the conclusion that you are talking about a server CPU - maybe even a dual Xeon.

I have a single Xeon 2.67 quad core Mac Pro with 6 gig of memory - and HT, which gives you a total of 8 virtual cores backed by the 4 physical cores just hums right along.

I see a 30 to 35% increase with HT turned on (8 virtual cores) vs just running the 4 physical cores.

Hope that provides a little insight for your decision

CH

Well, if RAM isn't a limiting factor, I'd say the 16 threads will give a better performance overall in Rosetta, irregardless if it's 2Ghz or 4Ghz (with the same architecture).

---

Thanks! Now I'll crunch the provided information!! ;D

---

how do i set up virtual cores?

how do i set up virtual cores?

In the BIOS you can define whether you wish to enable hyperthreading, on some CPUs anyway. If it is enabled then it runs as though twice as many cores are available, and there is some extensions in the CPU that help it do so more efficiently then a CPU that is not hyperthread enabled.

Since most Rosetta calculations involve floating point arithmetic, one actual bottleneck becomes a shared resource that does the math rather then the CPUs. So some workloads run significantly better under hyperthreading, others, such as Rosetta that are so specific to arithmetic, don't get as much of a boost. In essence you still only have so many clock ticks (GHz) to spread around.

---

Rosetta Moderator: Mod.Sense

In the BIOS you can define whether you wish to enable hyperthreading, on some CPUs anyway. If it is enabled then it runs as though twice as many cores are available, and there is some extensions in the CPU that help it do so more efficiently then a CPU that is not hyperthread enabled.

Since most Rosetta calculations involve floating point arithmetic, one actual bottleneck becomes a shared resource that does the math rather then the CPUs. So some workloads run significantly better under hyperthreading, others, such as Rosetta that are so specific to arithmetic, don't get as much of a boost. In essence you still only have so many clock ticks (GHz) to spread around.

How old/new of a computer might have this feature?

In the bios it will be called hyperthreading right? Is it in the "advanced" area... or associated with anything that makes it easy to find?

My one "new" machine is an almost 2 year old mac. (I don't know how to access the bios on it)

As far as old machines goes. I got excited about the project and I have turned my last couple of PCs and converted them to just crunch rosetta using linux desktop as the OS. Over the years, I have also taken old discarded machines from friends and relatives over the years and cobbled together a couple more machines from parts. Most of these are machines that are 5-10 years old.

Thanks for the help.

In the BIOS you can define whether you wish to enable hyperthreading, on some CPUs anyway. If it is enabled then it runs as though twice as many cores are available, and there is some extensions in the CPU that help it do so more efficiently then a CPU that is not hyperthread enabled.

Since most Rosetta calculations involve floating point arithmetic, one actual bottleneck becomes a shared resource that does the math rather then the CPUs. So some workloads run significantly better under hyperthreading, others, such as Rosetta that are so specific to arithmetic, don't get as much of a boost. In essence you still only have so many clock ticks (GHz) to spread around.

How old/new of a computer might have this feature?

In the bios it will be called hyperthreading right? Is it in the "advanced" area... or associated with anything that makes it easy to find?

My one "new" machine is an almost 2 year old mac. (I don't know how to access the bios on it)

As far as old machines goes. I got excited about the project and I have turned my last couple of PCs and converted them to just crunch rosetta using linux desktop as the OS. Over the years, I have also taken old discarded machines from friends and relatives over the years and cobbled together a couple more machines from parts. Most of these are machines that are 5-10 years old.

Thanks for the help.

Hi

The only CPU families with HyperThreading are:
Pentium4
Atom
Core i3, i5 and i7
plus the xeon (server) versions of the above. Not all of the CPUs from any of these families have it enabled though.

Your Atom CPU can enable HT (if it isn't already - see how many cores show up in Task Manager), but as that CPU only has 512kB L2 cache, running 2 instances of Rosetta would almost certainly reduce throughput. I'd probably enable HT in the BIOS but set Rosetta to only use 1 core so other threads can run on the HT 'core' which should make it slightly more responsive than with HT off.

Your Mac won't have HT as it will almost certainly be Core2 based.

HTH
Danny

---

[
The only CPU families with HyperThreading are:
Pentium4
Atom
Core i3, i5 and i7
plus the xeon (server) versions of the above. Not all of the CPUs from any of these families have it enabled though.

Your Atom CPU can enable HT (if it isn't already - see how many cores show up in Task Manager), but as that CPU only has 512kB L2 cache, running 2 instances of Rosetta would almost certainly reduce throughput. I'd probably enable HT in the BIOS but set Rosetta to only use 1 core so other threads can run on the HT 'core' which should make it slightly more responsive than with HT off.

Your Mac won't have HT as it will almost certainly be Core2 based.

HTH
Danny

The Mac is an Intel 2.66 Core 2 Duo (I have not kept up with Chip sets for YEARS... but that sounds like two cores)

I found 2 machines that have the feature... but it was already turned on in those boxes.

If I am going to run them, I want to be efficient.

Thanks for the help.

-TK

[
The only CPU families with HyperThreading are:
Pentium4
Atom
Core i3, i5 and i7
plus the xeon (server) versions of the above. Not all of the CPUs from any of these families have it enabled though.

Your Atom CPU can enable HT (if it isn't already - see how many cores show up in Task Manager), but as that CPU only has 512kB L2 cache, running 2 instances of Rosetta would almost certainly reduce throughput. I'd probably enable HT in the BIOS but set Rosetta to only use 1 core so other threads can run on the HT 'core' which should make it slightly more responsive than with HT off.

Your Mac won't have HT as it will almost certainly be Core2 based.

HTH
Danny

The Mac is an Intel 2.66 Core 2 Duo (I have not kept up with Chip sets for YEARS... but that sounds like two cores)

I found 2 machines that have the feature... but it was already turned on in those boxes.

If I am going to run them, I want to be efficient.

Thanks for the help.

-TK

No HT then, I'm afraid. The older P4 and new i7 chips have HT only.

---