VMware Folding Appliance
Posted: Sun Mar 08, 2009 8:48 am
On my new i7 machine, I was getting around 3K PPD running one SMP client. That's as many as I was earning in a MONTH on my old P4 machine so I wasn't dissatisfied. Even so...
I'm now running 4 instances of a "folding virtual appliance" using the free VMware Player and look to be getting something north of 8K PPD. FahMon is not wanting to cooperate so that's an estimate based on WUs worth 1920 points each and 12:30 (minutes:seconds) steps. At that rate each of the four VMware instances will compete a 1920 point WU in less than 24 hours.
How is this possible? Well, there is a new A2 core that's massively faster than the existing A1 core, but it only runs under Linux and OSX. So, even though the virtual machines are going to be a bit slower than FAH running 'native,' the faster A2 core way more than makes up for the difference.
It's crazy.
Here is where I got my inspiration:
http://forum.xcpus.com/xtreme-folding-h ... r-ppd.html
The A2 core is designed for 4 or more cores, but VMware only supports two cores per virtual machine and the i7 has 8 (4 physical cores, but Hyperthreading makes it looks like 8 to the OS), hence the need for 4 instances. With one SMP client under Windows I was seeing CPU utilization of only 50%. With the 4 VM machines I'm seeing utilization of 96% to 98%. There aren't a lot of cycles being wasted. One VM instance uses 25%, two use 50%, three 75% and four 100% so it's not rigging the system to run 4 instances.
I'm now running 4 instances of a "folding virtual appliance" using the free VMware Player and look to be getting something north of 8K PPD. FahMon is not wanting to cooperate so that's an estimate based on WUs worth 1920 points each and 12:30 (minutes:seconds) steps. At that rate each of the four VMware instances will compete a 1920 point WU in less than 24 hours.
How is this possible? Well, there is a new A2 core that's massively faster than the existing A1 core, but it only runs under Linux and OSX. So, even though the virtual machines are going to be a bit slower than FAH running 'native,' the faster A2 core way more than makes up for the difference.
It's crazy.
Here is where I got my inspiration:
http://forum.xcpus.com/xtreme-folding-h ... r-ppd.html
The A2 core is designed for 4 or more cores, but VMware only supports two cores per virtual machine and the i7 has 8 (4 physical cores, but Hyperthreading makes it looks like 8 to the OS), hence the need for 4 instances. With one SMP client under Windows I was seeing CPU utilization of only 50%. With the 4 VM machines I'm seeing utilization of 96% to 98%. There aren't a lot of cycles being wasted. One VM instance uses 25%, two use 50%, three 75% and four 100% so it's not rigging the system to run 4 instances.