Somebody with a GTX 550 out there? how much rac can we expect for this gpu?
Hi!
This is a quite new model, I couldn't find one among the top hosts.
But based on my own experienace with a somewhat slower predecessor model GTS 450 (also 192 CUDA cores), you shuould expect more than 8500 credits per day from the GPU alone (in 24/7 operation). This is without tweaking (e.g. running more than one task in parallel or overclocking).
It would be interesting to see just how much faster the GTX 550 is compared to the GTS 450, which is rather cheap now that it is kind of outdated.
This is a quite new model, I couldn't find one among the top hosts.
But based on my own experienace with a somewhat slower predecessor model GTS 450 (also 192 CUDA cores), you shuould expect more than 8500 credits per day from the GPU alone (in 24/7 operation). This is without tweaking (e.g. running more than one task in parallel or overclocking)
THAT would be really bad.
i can finish them in about 3800 secs on my ancient GTX260 with 192 cores.
this is linux-64 and the shaders are running @ 1295 MHz.
since a decent 550ti is running them above 2000 MHz - go figure...
Well, I wrote "more than ..." so I'm on the safe side :-)
Seriously, this adds an interesting comparision. While the GTX260 is "ancient" and has a slower clock rate, it has a wider memory interface. And while the Fermi GPUs technically have more registers, they can also execute more taks in parallel and if you compute the number of registers per task, Fermi has actually less registers per task! So it's not like Fermi GPUs are playing in another league compared to older GPUs with the same number of cores.
So all in all I guess it's safe to say that "greater than 10.000 credits/day" would be a better, yet still somewhat conservative guess.
It would be cool to have some diagrams mapping different models to runtime which would probably tell us what kind of influence shader clock, nr of cores , memory width etc have on runtime.
Seriously, this adds an interesting comparision. While the GTX260 is "ancient" and has a slower clock rate, it has a wider memory interface.
this does not help much - i also got a GTS250 which is only about 15% slower here.
Quote:
It would be cool to have some diagrams mapping different models to runtime which would probably tell us what kind of influence shader clock, nr of cores , memory width etc have on runtime.
as long as your app is doing a lot of CPU-work and GPU's are not getting saturated, it's hard to tell.
but one thing for sure: the linux-app is a lot faster than it's windows sister if you give it a free core.
It would be cool to have some diagrams mapping different models to runtime which would probably tell us what kind of influence shader clock, nr of cores , memory width etc have on runtime.
took the time and pushed the clocks on my GTX260 - single BRPcuda23nv270 tasks with one CPU-core spared:
620 MHz ~ 3.800 secs
700 MHz ~ 3.600 secs
that's disapointing...
but i did expect that.
oh, btw. CPU load is not about 20% but above 30% on a Q9450!
~7000secs (2 WU at the same time) GPU usage 80%, memory usage 662MB
So, for increasing the output by 100%, 2 WU rather than one, it increases the runtime by 40%. Which is pretty good i believe. The GTX 550 cost me 159€.
Could make a 12000RAC/day only with the GPU. Producing 1000credits every 7000secs.
Not bad. Although a GTX560 should be a lot faster with twice as many shaders.
And when comparing Fermi-based cards to older ones one has to keep in mind that compute capability 2.1 chips (anything except the high end GF100 and GF110 chips) struggle to use all of their shaders. Worst case is just 2/3. Not sure how well it works with the current Einstein app, haven't been following the numbers recently.
GTX 550
)
Hi!
This is a quite new model, I couldn't find one among the top hosts.
But based on my own experienace with a somewhat slower predecessor model GTS 450 (also 192 CUDA cores), you shuould expect more than 8500 credits per day from the GPU alone (in 24/7 operation). This is without tweaking (e.g. running more than one task in parallel or overclocking).
It would be interesting to see just how much faster the GTX 550 is compared to the GTS 450, which is rather cheap now that it is kind of outdated.
HB
RE: This is a quite new
)
THAT would be really bad.
i can finish them in about 3800 secs on my ancient GTX260 with 192 cores.
http://einsteinathome.org/host/4098263/tasks&offset=0&show_names=0&state=3
this is linux-64 and the shaders are running @ 1295 MHz.
since a decent 550ti is running them above 2000 MHz - go figure...
RE: ] THAT would be really
)
Well, I wrote "more than ..." so I'm on the safe side :-)
Seriously, this adds an interesting comparision. While the GTX260 is "ancient" and has a slower clock rate, it has a wider memory interface. And while the Fermi GPUs technically have more registers, they can also execute more taks in parallel and if you compute the number of registers per task, Fermi has actually less registers per task! So it's not like Fermi GPUs are playing in another league compared to older GPUs with the same number of cores.
So all in all I guess it's safe to say that "greater than 10.000 credits/day" would be a better, yet still somewhat conservative guess.
It would be cool to have some diagrams mapping different models to runtime which would probably tell us what kind of influence shader clock, nr of cores , memory width etc have on runtime.
CU
HB
Well, i'm expecting my msi
)
Well, i'm expecting my msi gtx550 cyclone 2 to arrive next friday, than we'll see.
Thanks all, for yours replys.
RE: Seriously, this adds an
)
this does not help much - i also got a GTS250 which is only about 15% slower here.
as long as your app is doing a lot of CPU-work and GPU's are not getting saturated, it's hard to tell.
but one thing for sure: the linux-app is a lot faster than it's windows sister if you give it a free core.
RE: It would be cool to
)
took the time and pushed the clocks on my GTX260 - single BRPcuda23nv270 tasks with one CPU-core spared:
620 MHz ~ 3.800 secs
700 MHz ~ 3.600 secs
that's disapointing...
but i did expect that.
oh, btw. CPU load is not about 20% but above 30% on a Q9450!
First Tasks done with the MSI
)
First Tasks done with the MSI GTX 550 TI cyclone 2 (192 cuda cores, memory width 192bits GDDR5, shader clock 1900Mhz, memory clock 2150MHz):
~4900secs (1 WU at time) GPU usage 56%, memory usage 330MB
~7000secs (2 WU at the same time) GPU usage 80%, memory usage 662MB
So, for increasing the output by 100%, 2 WU rather than one, it increases the runtime by 40%. Which is pretty good i believe. The GTX 550 cost me 159€.
Could make a 12000RAC/day only with the GPU. Producing 1000credits every 7000secs.
RE: The GTX 550 cost me
)
Not bad. Although a GTX560 should be a lot faster with twice as many shaders.
And when comparing Fermi-based cards to older ones one has to keep in mind that compute capability 2.1 chips (anything except the high end GF100 and GF110 chips) struggle to use all of their shaders. Worst case is just 2/3. Not sure how well it works with the current Einstein app, haven't been following the numbers recently.
MrS
Scanning for our furry friends since Jan 2002
RE: First Tasks done with
)
you'd probably be able to increase your output when running linux-x64.