4. With your 580 and your CPU with 6 physical cores scheduled as 12 by hyperthreading, I strongly suspect you'll find processing of these GW tasks to be much more efficient at multiplicities higher than 1. I suspect you'll see 2 far better than one, 3 a good bit better still, and quite possibly more steps up. Try them, but remember to keep the queue request low enough to avoid gross over-fetch.
Does gravity now have smaller GPU RAM requirements? Because some of the old ones were rather large (3GB+), so doubling them up on 4GB cards is a bad idea - resulting in exceedingly slow runtimes as it accesses system RAM. The few new ones I checked were only using 1GB.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Does gravity now have smaller GPU RAM requirements? Because some of the old ones were rather large (3GB+), so doubling them up on 4GB cards is a bad idea - resulting in exceedingly slow runtimes as it accesses system RAM. The few new ones I checked were only using 1GB.
All we have is our own observations and official statements. Yesterday Bernd advised that the current set had memory requirements from 0.7 to 2.7 GB.
Taken at face value that suggests that the person to whom I was giving advice, with an 8Gb card of a type that I believe slows down rather gracefully on GW tasks when the memory used just barely exceeds available, might well do best at 3X. Running 3X on a 4GB card would be much, much braver.
As to observation, the not very diverse sample I've seen here has been reported as using 1.2Gb of dynamic memory. I'm currently running them at 4X on an 8GB 5700 card.
Nah, LGA1366 forever... ! I'm happy with these for running like a budget multicore solution. Budget... because these don't have much any other than usage value left. Currently this cpu plus motherboard costs like less than 1/10 of what a i9 setup would cost. CPU-Z benchmark claims that at 3.8 GHz this still has 62 % single core power compared with a stock i9-9900KF and multithreading with 12 threads gives 47 %. In reality this can't be quite that good as this lacks the newer processor instructions and doesn't even support higher than about 1800MHz speeds of DDR3. But anyway... considering the second hand market prices now, still very usable semi vintage stuff ! I won't let go of these.... no I won't :'(
Okay, I checked runtimes on my system with stock settings on the RX 580. About 1450 sec for 1X and 1650 sec per task when running 2X. My test tasks were from 200Hz area. I guess your 2X runtimes are pretty normal. As I'm running my Xeon overclocked that probably makes the difference (W3680 is 3333MHz ... runtimes 2100 sec).
Nah, LGA1366 forever... ! I'm happy with these for running like a budget multicore solution. Budget... because these don't have much any other than usage value left. Currently this cpu plus motherboard costs like less than 1/10 of what a i9 setup would cost. CPU-Z benchmark claims that at 3.8 GHz this still has 62 % single core power compared with a stock i9-9900KF and multithreading with 12 threads gives 47 %. In reality this can't be quite that good as this lacks the newer processor instructions and doesn't even support higher than about 1800MHz speeds of DDR3. But anyway... considering the second hand market prices now, still very usable semi vintage stuff ! I won't let go of these.... no I won't :'(
Okay, I checked runtimes on my system with stock settings on the RX 580. About 1450 sec for 1X and 1650 sec per task when running 2X. My test tasks were from 200Hz area. I guess your 2X runtimes are pretty normal. As I'm running my Xeon overclocked that probably makes the difference (W3680 is 3333MHz ... runtimes 2100 sec).
Pah! My Xeon X5650s are better. Both are 32nm and came out Jan 2010, but I can (and do) have two CPUs per motherboard. And mine can take 288GB RAM per CPU, yours is only 24GB! I bought mine (all four of them) 6 months ago for £7=$9 each. Probably rubbish at gravity, both are slower per core than my newer i5 8600K, but I don't have any decent GPUs to connect to them. I'm probably using up the world's oil reserves powering this old junk, but I like collecting free and virtually free stuff on Ebay and Gumtree and making them do something useful.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Can the server tell what cards we have or what instruction set they have? If so it shouldn't send gravity to those cards. Also the program itself should abort with an error, not just run indefinitely. Surely if the instruction required is not on the chip, it should give up!
I agree. I believe the server doesn't have that kind of database of the different gpus. But maybe that kind of function would require massive changes in the scheduler and elsewhere... and other things (more urgent) are under developement. Only guessing.
I agree. I believe the server doesn't have that kind of database of the different gpus. But maybe that kind of function would require massive changes in the scheduler and elsewhere... and other things (more urgent) are under developement. Only guessing.
At least the work unit's program should notice the instruction it's trying to carry out isn't available, and stop with a "computation error". I'm surprised it can work at all - surely an illegal function call should at least crash the program.
That's an X5660, one up from mine. I was referring to your w3680, which Intel claims cannot access more than 24GB.
And please don't beep. Believe it or not, I have a Yellow Fronted Amazon parrot which bleeps every 3 seconds, sometimes for half an hour. It's that annoying microwave or smoke alarm beep. Again and again and again.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
...Yesterday Bernd advised that the current set had memory requirements from 0.7 to 2.7 GB....
With an early set of tasks that had DFs (delta frequencies) of 0.1, I recorded VRAM memory usage of ~0.45 GB; so, lower than expected. Those of DF 0.15 & 0.20, however, did use ~0 .75 GB, while those of DF 0.25 & 0.30 use ~1.2 GB. I'm waiting to see whether this trend of regularly increasing memory requirements continues through the Spotlight data series.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
Recent average individual task times for the Spotlight workunits, when running 3X concurrent tasks, are ~6:20 for my RX 5600XT, and ~8:30 for my RX 570s. No invalids out of 450 completed valid tasks.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
Recent average individual task times for the Spotlight workunits, when running 3X concurrent tasks, are ~6:20 for my RX 5600XT, and ~8:30 for my RX 570s.
Those are good numbers, but how many CPU cores does it take to do that?
I have an RX 570 under Win7 64-bit, but could put it on Ubuntu. Would that help much?
Those are good numbers, but how many CPU cores does it take to do that?
I have an RX 570 under Win7 64-bit, but could put it on Ubuntu. Would that help much?
The two RX 570s are each running 3X concurrent tasks in an older machine with a 4c/4t AMD Phenom II X4 940 (this host). The RX 5600xt is running with a 2c/4t Pentium G5600 (this host). A few years back Linux/Ubuntu ran gamma-ray pulsar tasks slightly faster than Windows, but I don't know how the systems compare with these GW tasks. I expect that as the current GW data series progresses, GPU memory requirements will increase and task times will lengthen, but we will see!
Ideas are not fixed, nor should they be; we live in model-dependent reality.
archae86 wrote: 4. With your
)
Does gravity now have smaller GPU RAM requirements? Because some of the old ones were rather large (3GB+), so doubling them up on 4GB cards is a bad idea - resulting in exceedingly slow runtimes as it accesses system RAM. The few new ones I checked were only using 1GB.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Peter Hucker wrote:Does
)
All we have is our own observations and official statements. Yesterday Bernd advised that the current set had memory requirements from 0.7 to 2.7 GB.
Taken at face value that suggests that the person to whom I was giving advice, with an 8Gb card of a type that I believe slows down rather gracefully on GW tasks when the memory used just barely exceeds available, might well do best at 3X. Running 3X on a 4GB card would be much, much braver.
As to observation, the not very diverse sample I've seen here has been reported as using 1.2Gb of dynamic memory. I'm currently running them at 4X on an 8GB 5700 card.
vsral wrote:Didn't know a
)
Nah, LGA1366 forever... ! I'm happy with these for running like a budget multicore solution. Budget... because these don't have much any other than usage value left. Currently this cpu plus motherboard costs like less than 1/10 of what a i9 setup would cost. CPU-Z benchmark claims that at 3.8 GHz this still has 62 % single core power compared with a stock i9-9900KF and multithreading with 12 threads gives 47 %. In reality this can't be quite that good as this lacks the newer processor instructions and doesn't even support higher than about 1800MHz speeds of DDR3. But anyway... considering the second hand market prices now, still very usable semi vintage stuff ! I won't let go of these.... no I won't :'(
Okay, I checked runtimes on my system with stock settings on the RX 580. About 1450 sec for 1X and 1650 sec per task when running 2X. My test tasks were from 200Hz area. I guess your 2X runtimes are pretty normal. As I'm running my Xeon overclocked that probably makes the difference (W3680 is 3333MHz ... runtimes 2100 sec).
Richie wrote: vsral
)
Pah! My Xeon X5650s are better. Both are 32nm and came out Jan 2010, but I can (and do) have two CPUs per motherboard. And mine can take 288GB RAM per CPU, yours is only 24GB! I bought mine (all four of them) 6 months ago for £7=$9 each. Probably rubbish at gravity, both are slower per core than my newer i5 8600K, but I don't have any decent GPUs to connect to them. I'm probably using up the world's oil reserves powering this old junk, but I like collecting free and virtually free stuff on Ebay and Gumtree and making them do something useful.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Quote:Quote:Can the server
)
I agree. I believe the server doesn't have that kind of database of the different gpus. But maybe that kind of function would require massive changes in the scheduler and elsewhere... and other things (more urgent) are under developement. Only guessing.
BEEEEP *alarm sound to indicate a fail*
This one has 48 GB ; )
Richie wrote:I agree. I
)
At least the work unit's program should notice the instruction it's trying to carry out isn't available, and stop with a "computation error". I'm surprised it can work at all - surely an illegal function call should at least crash the program.
That's an X5660, one up from mine. I was referring to your w3680, which Intel claims cannot access more than 24GB.
And please don't beep. Believe it or not, I have a Yellow Fronted Amazon parrot which bleeps every 3 seconds, sometimes for half an hour. It's that annoying microwave or smoke alarm beep. Again and again and again.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
archae86 wrote:...Yesterday
)
With an early set of tasks that had DFs (delta frequencies) of 0.1, I recorded VRAM memory usage of ~0.45 GB; so, lower than expected. Those of DF 0.15 & 0.20, however, did use ~0 .75 GB, while those of DF 0.25 & 0.30 use ~1.2 GB. I'm waiting to see whether this trend of regularly increasing memory requirements continues through the Spotlight data series.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
Recent average individual
)
Recent average individual task times for the Spotlight workunits, when running 3X concurrent tasks, are ~6:20 for my RX 5600XT, and ~8:30 for my RX 570s. No invalids out of 450 completed valid tasks.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
cecht wrote:Recent average
)
Those are good numbers, but how many CPU cores does it take to do that?
I have an RX 570 under Win7 64-bit, but could put it on Ubuntu. Would that help much?
Jim1348 wrote:Those are good
)
The two RX 570s are each running 3X concurrent tasks in an older machine with a 4c/4t AMD Phenom II X4 940 (this host). The RX 5600xt is running with a 2c/4t Pentium G5600 (this host). A few years back Linux/Ubuntu ran gamma-ray pulsar tasks slightly faster than Windows, but I don't know how the systems compare with these GW tasks. I expect that as the current GW data series progresses, GPU memory requirements will increase and task times will lengthen, but we will see!
Ideas are not fixed, nor should they be; we live in model-dependent reality.