A 7770 should be about 20% faster than a 7750 on the same machine with the same tasks. So theoretically if I ran 2* 7770 with 2* tasks each on my poor Athlon 64 x2, I would have completion times about 76 minutes. If you get 1 task in 60 minutes with a 7770 on your computer, then I am very happy with my current performance. Of course, I'm only running E@H on ATI because SETI is having server problems; I use my nVidia computers for E@H.
I do like the 7750's. They do not use a 6-pin connector!
These seem to be very good numbers. I was interested to see the CPU use so I had a look at your hosts list but you don't seem to have a 7770 currently in any of your hosts. If you've changed it for a different GPU, perhaps the former results are still in the online database and therefore still accessible. Can you advise if the results are still there thanks.
I have a 7770 in this host, also running 2x, but I can't get anywhere near 3600s. It's being driven by a 2.8GHz sandy bridge dual core and with 1 free core, it takes 4800s for 2x. Yesterday, I experimented running with no CPU tasks at all and the 2x time dropped to 4600s which is still way short of 3600s. It's now back running a CPU task again and the GPU task time has returned to 4800s.
Any clues on what you had to do to get to 3600s? Thanks.
Gary results show for the GPU apps a CPU time of around 800 secs while astrocab results show only 300 secs...
As astrocab said his CPU was running at 4.2Ghz (if it's not a typo) then is heavilly OCed and there is a big jump in CPU speed from Gary's 2.8Ghz...
Not sure if the speed is enough to make such a big difference in the CPU time, may be the version of the drivers and also the fact that mobos that support ivy bridge CPUs ussually have support to run the PCIe as version 3.0 while with a sandy the mobo will be running the PCIe at most in version 2.0... and who knows what other differences there are between both systems...
4.2 sure, it is not a typo.
pci-e 3.0 doesn't make weather too, even for 7970@pci-e 2.0 i noticed no performance loss.
7770 gpu load was about 98% during run. try 12.11beta, it's about 20% faster, than 12.10
and i'm astrocrab )
thank you.
Thank you very much for that! Did you swap that 7770 back in just to do 2 tasks for my benefit? I see you did those two tasks just a couple of hours ago so I'm embarrassed that you went to that much trouble just to provide some data for me. I also noticed what must have been the next task (this one) which was started on 'Cape Verde' (7770) and finished on 'Tahiti" (7970). At least this is what appears to have happened. What happens with the GPU utilisation factor when you change cards like this? You're running x5 on the 7970 aren't you?
Those results were quite helpful. I saw the CPU time was around 300s. On my 7770, the CPU time is around 800s. So I'm assuming that the reason for this is that I'm using PCIe2 (Sandy Bridge) whilst you're using PCIe3 (Ivy Bridge). Sure, the clock speed (2.8GHz to 4.2GHz) would certainly make some difference too, but I think the biggest factor may be PCIe3.
I have a 3570K Ivy Bridge host (this one) running a GTX650Ti. It is also overclocked to 4.2GHz. I tried that host initially with a GTX650, hoping to see a performance improvement when compared with the same card in a PCIe2 host. I found that there was almost zero difference (maybe a few seconds) between the two. I wanted to try out a GTX650TI so I replaced the 650 with a new 650Ti I'd just purchased. The performance improved in line with what you might expect for a 650TI over a 650 but I didn't think there was any particular benefit from PCIe3. So I made the (perhaps foolish) assumption that there was not much to be gained by buying PCIe3 capable motherboards and CPUs.
I'm now wondering if there might be a big benefit for AMD GPUs. If anyone knows for sure, please chime in. I'll probably end up building another 3570K based rig to put the 7770 in if it makes that much difference.
I haven't played with drivers. All my hosts run Linux and I just update from repositories as new stuff becomes available. I only have the one AMD card and that host is running openSUSE 12.2. I built the driver from the latest source at the time I installed the OS so it's probably 12.10. I'll have to check that and try the latest beta if it's likely to improve things.
Gary results show for the GPU apps a CPU time of around 800 secs while astrocab results show only 300 secs...
As astrocab said his CPU was running at 4.2Ghz (if it's not a typo) then is heavilly OCed and there is a big jump in CPU speed from Gary's 2.8Ghz...
Not sure if the speed is enough to make such a big difference in the CPU time, may be the version of the drivers and also the fact that mobos that support ivy bridge CPUs ussually have support to run the PCIe as version 3.0 while with a sandy the mobo will be running the PCIe at most in version 2.0... and who knows what other differences there are between both systems...
Horacio,
Thanks for your reply!! I was too busy composing a reply to Astrocrab to even notice your reply until after I had posted mine.
My experience is very much with nvidia GPUs. I have two 550Tis, quite a few 650s now :-), and one 650Ti. All of the kepler series seem largely unaffected by the quality of the CPU and/or the number of free cores. A GTX650 seems to take pretty much the same time installed in a 3570K host at 4.2GHz as it does in a Celeron G540 running at 2.5GHz. Take a look at this host which is a dual core Celeron G540. The 650 in it is taking just over 61 mins (3680s) which is pretty much the same as when I had a 650 in the 3570k host linked in the previous message. They are all running GPU tasks x2.
I've also found that the GPU performance doesn't really change much with free CPU cores. If you look through the tasks list for the G540, you will see plenty of CPU tasks. I started out always having a free core on each host. I found that the GPU performance didn't change if I set CPU core usage to 100%. The CPU task times increased a bit but the GPU times didn't change. So all my nvidia endowed hosts are running CPU tasks on all cores.
It seems to be a different story with AMD. I only have the one host with the 7770 and with x2 you are forced to have a free CPU core because of the 0.5 CPUs built into the tasks for opencl-ati app. By freeing up a second core it does make an appreciable difference as noted earlier. What I hadn't really appreciated was the CPU time difference between a 650 and a 7770 - 200s to 800s. This makes me hopeful that the 7770 may do a lot better in a 3570K host at 4.2GHz with both CPU speed and PCIe3 playing a part.
GPU-Z is also a handy utility
)
GPU-Z is also a handy utility that should also tell you what system bus and bandwidth the slot is currently running.
http://www.techpowerup.com/gpuz/
Hi Astrocrab, A 7770
)
Hi Astrocrab,
A 7770 should be about 20% faster than a 7750 on the same machine with the same tasks. So theoretically if I ran 2* 7770 with 2* tasks each on my poor Athlon 64 x2, I would have completion times about 76 minutes. If you get 1 task in 60 minutes with a 7770 on your computer, then I am very happy with my current performance. Of course, I'm only running E@H on ATI because SETI is having server problems; I use my nVidia computers for E@H.
I do like the 7750's. They do not use a 6-pin connector!
i retest 7770
)
i retest 7770 again:
1-1960s
2-3600s.
RE: i retest 7770
)
These seem to be very good numbers. I was interested to see the CPU use so I had a look at your hosts list but you don't seem to have a 7770 currently in any of your hosts. If you've changed it for a different GPU, perhaps the former results are still in the online database and therefore still accessible. Can you advise if the results are still there thanks.
I have a 7770 in this host, also running 2x, but I can't get anywhere near 3600s. It's being driven by a 2.8GHz sandy bridge dual core and with 1 free core, it takes 4800s for 2x. Yesterday, I experimented running with no CPU tasks at all and the 2x time dropped to 4600s which is still way short of 3600s. It's now back running a CPU task again and the GPU task time has returned to 4800s.
Any clues on what you had to do to get to 3600s? Thanks.
Cheers,
Gary.
http://einstein.phys.uwm.edu/
)
http://einsteinathome.org/workunit/138129653
http://einsteinathome.org/workunit/138129689
was runned together, please, check.
card: http://www.sapphiretech.com/presentation/product/?cid=1&gid=3&sgid=1159&pid=1468&psn=&lid=1&leg=0
cpu: ivy bridge i5 @ 4200 MHz
overall cpu load during crunching this two WUs was about 3% with no other task running.
i have made some screenshots, for any case.
i've found, what with catalyst 12.11beta WUs run noticeably faster, than with 12.10
Gary results show for the GPU
)
Gary results show for the GPU apps a CPU time of around 800 secs while astrocab results show only 300 secs...
As astrocab said his CPU was running at 4.2Ghz (if it's not a typo) then is heavilly OCed and there is a big jump in CPU speed from Gary's 2.8Ghz...
Not sure if the speed is enough to make such a big difference in the CPU time, may be the version of the drivers and also the fact that mobos that support ivy bridge CPUs ussually have support to run the PCIe as version 3.0 while with a sandy the mobo will be running the PCIe at most in version 2.0... and who knows what other differences there are between both systems...
4.2 sure, it is not a
)
4.2 sure, it is not a typo.
pci-e 3.0 doesn't make weather too, even for 7970@pci-e 2.0 i noticed no performance loss.
7770 gpu load was about 98% during run. try 12.11beta, it's about 20% faster, than 12.10
and i'm astrocrab )
thank you.
RE: http://einstein.phys.uw
)
Thank you very much for that! Did you swap that 7770 back in just to do 2 tasks for my benefit? I see you did those two tasks just a couple of hours ago so I'm embarrassed that you went to that much trouble just to provide some data for me. I also noticed what must have been the next task (this one) which was started on 'Cape Verde' (7770) and finished on 'Tahiti" (7970). At least this is what appears to have happened. What happens with the GPU utilisation factor when you change cards like this? You're running x5 on the 7970 aren't you?
Those results were quite helpful. I saw the CPU time was around 300s. On my 7770, the CPU time is around 800s. So I'm assuming that the reason for this is that I'm using PCIe2 (Sandy Bridge) whilst you're using PCIe3 (Ivy Bridge). Sure, the clock speed (2.8GHz to 4.2GHz) would certainly make some difference too, but I think the biggest factor may be PCIe3.
I have a 3570K Ivy Bridge host (this one) running a GTX650Ti. It is also overclocked to 4.2GHz. I tried that host initially with a GTX650, hoping to see a performance improvement when compared with the same card in a PCIe2 host. I found that there was almost zero difference (maybe a few seconds) between the two. I wanted to try out a GTX650TI so I replaced the 650 with a new 650Ti I'd just purchased. The performance improved in line with what you might expect for a 650TI over a 650 but I didn't think there was any particular benefit from PCIe3. So I made the (perhaps foolish) assumption that there was not much to be gained by buying PCIe3 capable motherboards and CPUs.
I'm now wondering if there might be a big benefit for AMD GPUs. If anyone knows for sure, please chime in. I'll probably end up building another 3570K based rig to put the 7770 in if it makes that much difference.
I haven't played with drivers. All my hosts run Linux and I just update from repositories as new stuff becomes available. I only have the one AMD card and that host is running openSUSE 12.2. I built the driver from the latest source at the time I installed the OS so it's probably 12.10. I'll have to check that and try the latest beta if it's likely to improve things.
Once again, thanks very much for your reply.
Cheers,
Gary.
RE: Gary results show for
)
Horacio,
Thanks for your reply!! I was too busy composing a reply to Astrocrab to even notice your reply until after I had posted mine.
My experience is very much with nvidia GPUs. I have two 550Tis, quite a few 650s now :-), and one 650Ti. All of the kepler series seem largely unaffected by the quality of the CPU and/or the number of free cores. A GTX650 seems to take pretty much the same time installed in a 3570K host at 4.2GHz as it does in a Celeron G540 running at 2.5GHz. Take a look at this host which is a dual core Celeron G540. The 650 in it is taking just over 61 mins (3680s) which is pretty much the same as when I had a 650 in the 3570k host linked in the previous message. They are all running GPU tasks x2.
I've also found that the GPU performance doesn't really change much with free CPU cores. If you look through the tasks list for the G540, you will see plenty of CPU tasks. I started out always having a free core on each host. I found that the GPU performance didn't change if I set CPU core usage to 100%. The CPU task times increased a bit but the GPU times didn't change. So all my nvidia endowed hosts are running CPU tasks on all cores.
It seems to be a different story with AMD. I only have the one host with the 7770 and with x2 you are forced to have a free CPU core because of the 0.5 CPUs built into the tasks for opencl-ati app. By freeing up a second core it does make an appreciable difference as noted earlier. What I hadn't really appreciated was the CPU time difference between a 650 and a 7770 - 200s to 800s. This makes me hopeful that the 7770 may do a lot better in a 3570K host at 4.2GHz with both CPU speed and PCIe3 playing a part.
Cheers,
Gary.
RE: Did you swap that 7770
)
yes )
yes
raised to 16-17 with 1 7970 running and about 27-28% with 2 7970 running. yes, x5
no, i have performed test with 7770@pci-e 2.0, same 3600 for x2.