How are F@h's FLOPS calculated?
Posted: Mon Mar 05, 2012 6:04 pm
I apologize if this has been authoritatively answered before, I've done some searching but was unable to find a previous post. I'm wondering, how are Folding@home's FLOPS calculated? Here's the information I have so far:
1. On the PS3 FAQ, there's a section titled "How are the FLOPS calculated?" but while there's useful information provided, the question isn't directly answered.
2. In July of 2011, Bruce wrote here: "I'm not certain, but I think each WU completed reports that a specific number of FLOPs were completed." I think I've read this elsewhere, and that the server measures how long the WU takes to complete, but I haven't been able to find where I read it. Has this been definitively stated one way or another?
3. I understand that FLOPS aren't the most accurate way of measuring scientific throughput. I've read this a number of times now, including in Bruce's above statement. I'm just wondering how F@h arrives at their numbers is all.
4. The native FLOPS are calculated conservatively, while the x86 FLOPS are less so. The latter measurement provides a more realistic number when considering the standard x86 hardware.
5. The FLOPS FAQ has some good information, but doesn't explain how FLOPS are measured, or if they are determined in accordance to Bruce's post.
I'd appreciate any help in answering this question. If there's any relevant posts or if the question has been answered before, I'd sure appreciate a link. I've been trying to find one.
1. On the PS3 FAQ, there's a section titled "How are the FLOPS calculated?" but while there's useful information provided, the question isn't directly answered.
2. In July of 2011, Bruce wrote here: "I'm not certain, but I think each WU completed reports that a specific number of FLOPs were completed." I think I've read this elsewhere, and that the server measures how long the WU takes to complete, but I haven't been able to find where I read it. Has this been definitively stated one way or another?
3. I understand that FLOPS aren't the most accurate way of measuring scientific throughput. I've read this a number of times now, including in Bruce's above statement. I'm just wondering how F@h arrives at their numbers is all.
4. The native FLOPS are calculated conservatively, while the x86 FLOPS are less so. The latter measurement provides a more realistic number when considering the standard x86 hardware.
5. The FLOPS FAQ has some good information, but doesn't explain how FLOPS are measured, or if they are determined in accordance to Bruce's post.
I'd appreciate any help in answering this question. If there's any relevant posts or if the question has been answered before, I'd sure appreciate a link. I've been trying to find one.
