Anyone with a cpu with a lot of L2 cache? maybe an i7? Who can confirm that the S6 Search is heavily dependent on the amount of the L2 cache?
I think you can go into Task Manager and then under the processes tab find the unit and how much memory it is taking. If it is more then your L2 cache then you are in trouble as the units is constantly swapping out to the harddrive instead of being able to keep the whole thing in memory.
Mikey, the L2 cache is a fast built in memory in the CPU it's got nothing to do with how much RAM one have installed in the computer. Usually CPUs have L1 (level 1) cache for both instructions and data, then they have L2 cache for data and sometimes also L3 cache for data. The L2 cache on my i7 3770K is only 4x256KB so 1 MB, the L3 on the other hand is 8 MB.
S6 tasks take ~9.5 hours, I run the CPU @ 4.2 GHz HT=on and load 5 tasks on it along with one on the built in HD 4000 and 2 on the GTX 660TIi.
I knew that, what I was trying to say obviously not clearly, was that IF a unit fits completely into the L2 cache then it will run faster then if it doesn't. I have an i7-3612QM 2.1ghz cpu[Family 6 Model 58 Stepping 9], this means:
Level 1 Cache 256 KB
Level 2 Cache 1024 KB
Level 3 Cache 6144 KB
As I said mine is a laptop and I do not use the cpu to crunch gpu units, so only cpu units are crunched and only 5 of them at a time. I JUST finished another S6 unit it took 7:46:40 to finish. I use the laptop about half the day doing this kind of stuff on the forums, playing Freecell type games or doing my Genealogy. It sits on a little table with wheels that I have next to my couch, I can swing it around in front of me like now or swing it away. It crunches 24/7.
what I was trying to say obviously not clearly, was that IF a unit fits completely into the L2 cache then it will run faster then if it doesn't.
While this is true it's not relevant to E@H as WUs take almost 30 MB in main memory each. But they don'T have to fit into CPU caches completely, it's enough if the current working set (a slice of the whole data at a time) fits into the caches.
Anyway, Einstein does love main memory bandwidth, and since you can never have enough of it the caches help to to avoid trips to main memory. If Filipe is seeing a "dramatic" difference between those two hosts I suspect there's a bit more than just the caches: e.g. DDR2 vs. DDR3 and different FSB speeds, maybe one is running single channel memory.
what I was trying to say obviously not clearly, was that IF a unit fits completely into the L2 cache then it will run faster then if it doesn't.
While this is true it's not relevant to E@H as WUs take almost 30 MB in main memory each. But they don'T have to fit into CPU caches completely, it's enough if the current working set (a slice of the whole data at a time) fits into the caches.
Anyway, Einstein does love main memory bandwidth, and since you can never have enough of it the caches help to to avoid trips to main memory. If Filipe is seeing a "dramatic" difference between those two hosts I suspect there's a bit more than just the caches: e.g. DDR2 vs. DDR3 and different FSB speeds, maybe one is running single channel memory.
MrS
You CAN'T be doing this too me...it is ONLY 7AM and I am ALREADY learning something new!!! THANKS!!!
BTW: I made a silly typo in that post. Windows task manager actually shows ~300 MB per Einstein CPU task, give or take 100 MB. While there are different ways to measure memory footprint it's certainly enough to get the order of magnitude right.
RE: RE: RE: Anyone with
)
I knew that, what I was trying to say obviously not clearly, was that IF a unit fits completely into the L2 cache then it will run faster then if it doesn't. I have an i7-3612QM 2.1ghz cpu[Family 6 Model 58 Stepping 9], this means:
Level 1 Cache 256 KB
Level 2 Cache 1024 KB
Level 3 Cache 6144 KB
As I said mine is a laptop and I do not use the cpu to crunch gpu units, so only cpu units are crunched and only 5 of them at a time. I JUST finished another S6 unit it took 7:46:40 to finish. I use the laptop about half the day doing this kind of stuff on the forums, playing Freecell type games or doing my Genealogy. It sits on a little table with wheels that I have next to my couch, I can swing it around in front of me like now or swing it away. It crunches 24/7.
RE: what I was trying to
)
While this is true it's not relevant to E@H as WUs take almost 30 MB in main memory each. But they don'T have to fit into CPU caches completely, it's enough if the current working set (a slice of the whole data at a time) fits into the caches.
Anyway, Einstein does love main memory bandwidth, and since you can never have enough of it the caches help to to avoid trips to main memory. If Filipe is seeing a "dramatic" difference between those two hosts I suspect there's a bit more than just the caches: e.g. DDR2 vs. DDR3 and different FSB speeds, maybe one is running single channel memory.
MrS
Scanning for our furry friends since Jan 2002
RE: RE: what I was trying
)
You CAN'T be doing this too me...it is ONLY 7AM and I am ALREADY learning something new!!! THANKS!!!
Haha.. you're welcome
)
Haha.. you're welcome :)
BTW: I made a silly typo in that post. Windows task manager actually shows ~300 MB per Einstein CPU task, give or take 100 MB. While there are different ways to measure memory footprint it's certainly enough to get the order of magnitude right.
MrS
Scanning for our furry friends since Jan 2002