I don't think it uses PCIe 3.0 by default - pretty sure I saw a message by Jeroen (and here it is) saying he was on the case with nvidia to get a 'Force PCIe 3.0' option put in.
You can tell by firing up the control panel and looking to see what the current width and transaction rate is (5 GT/s for PCIe2.0 and 8GT/s for PCIe 3.0). Check while the GPU is working, as the rate can be scaled back for power-saving if the GPU is idle.
Another thing that may be acting here is the amount of free CPU cores you have on each host... as the 3770 has only 8 vs the 980 that have 12, if your settings are the same (use xx% of the processors) that percentage on the 980 will leave more free cores than on the 3770...
NVIDIA drivers should enable PCI-E 3.0 by default with an Ivy Bridge CPU (3770K), Z77 motherboard, and the slots set to GEN3 in BIOS. One option to check would be to download and run GPUZ. With a single card installed, GPUZ should report PCI-E 3.0 @ x16 3.0 under load. I would also suggest keeping the CPU load below 50% with HT enabled to get the best performance out of the BRP GPU application. Alternatively, disable HT and make sure one core is available for BRP.
I will set 38% (3 out of 8) of processors to crunch. I know that speeds up both CPU and GPU times, but will there be a net loss of Asteroid points? I know I am using much more of the CPU (100%) compared to about 50 or 60% for all processors being used which I asssumed would result in more points being crunched. Is that true? I care about my Asteroid points not being overly compromised. I can accept a reasonable trade off. Asteroids crunch about 55 minutes / WU faster on my 3770K @ 100% (all 8) processors active and 100% CPU usage compared to all 12 processors at 100% usage on i7 980.
What are the processing times you are getting out your 680's on PCIe 3.0... My 670 is getting 45 min/WU @ 3 WU per GPU @ 5 WCG WU's CPU crunching. Does that sound like it is maximized to you, Jeroen? It is in a 16X 3.0 slot according to GPU-Z.
with 3 cpu tasks & 1 core free to feed the GPU. 2 wu are ~10% lower per unit in both elapsed & cpu whereas the 3 wu were just marginally ahead of the 1 wu costing more in cpu. I'm currently running 2 wu and getting 50000 credits/day total from this host.
Your 3 wu elapsed shows about 10% lower elapsed than my data and probably is a result of your 3770K MB.
What are the processing times you are getting out your 680's on PCIe 3.0... My 670 is getting 45 min/WU @ 3 WU per GPU @ 5 WCG WU's CPU crunching. Does that sound like it is maximized to you, Jeroen? It is in a 16X 3.0 slot according to GPU-Z.
The lowest processing times that I have seen with my 680s were in Linux and with the cards in x16 3.0 slots. Two tasks ran at 1220 seconds. With the cards in x16 2.0 slots, two tasks run at 1358 seconds. If I remember right, I was able to run three tasks at around 1800 seconds each but that was quite a ways back. There was little gain running three tasks at once which is why I ran two instead. My 680 cards for some reason run significantly slower in Windows with this project. I have a 680 running in Windows in an x16 2.0 slot and two tasks run at approximately 1670 seconds.
Something I would suggest trying is to disable HT in BIOS and suspend CPU projects temporarily. Check to see what your 670 processing time looks like with these changes.
GTX 670 crunching slower on 3770 K PCIe 3 slot than on i7 980 PC
)
I don't think it uses PCIe 3.0 by default - pretty sure I saw a message by Jeroen (and here it is) saying he was on the case with nvidia to get a 'Force PCIe 3.0' option put in.
You can tell by firing up the control panel and looking to see what the current width and transaction rate is (5 GT/s for PCIe2.0 and 8GT/s for PCIe 3.0). Check while the GPU is working, as the rate can be scaled back for power-saving if the GPU is idle.
Hi Mel, You haven't said
)
Hi Mel,
You haven't said which MB you are using. Perhaps not all the PCIe slots are 3.0 x16.
Gord
Another thing that may be
)
Another thing that may be acting here is the amount of free CPU cores you have on each host... as the 3770 has only 8 vs the 980 that have 12, if your settings are the same (use xx% of the processors) that percentage on the 980 will leave more free cores than on the 3770...
NVIDIA drivers should enable
)
NVIDIA drivers should enable PCI-E 3.0 by default with an Ivy Bridge CPU (3770K), Z77 motherboard, and the slots set to GEN3 in BIOS. One option to check would be to download and run GPUZ. With a single card installed, GPUZ should report PCI-E 3.0 @ x16 3.0 under load. I would also suggest keeping the CPU load below 50% with HT enabled to get the best performance out of the BRP GPU application. Alternatively, disable HT and make sure one core is available for BRP.
I will set 38% (3 out of 8)
)
I will set 38% (3 out of 8) of processors to crunch. I know that speeds up both CPU and GPU times, but will there be a net loss of Asteroid points? I know I am using much more of the CPU (100%) compared to about 50 or 60% for all processors being used which I asssumed would result in more points being crunched. Is that true? I care about my Asteroid points not being overly compromised. I can accept a reasonable trade off. Asteroids crunch about 55 minutes / WU faster on my 3770K @ 100% (all 8) processors active and 100% CPU usage compared to all 12 processors at 100% usage on i7 980.
What are the processing times
)
What are the processing times you are getting out your 680's on PCIe 3.0... My 670 is getting 45 min/WU @ 3 WU per GPU @ 5 WCG WU's CPU crunching. Does that sound like it is maximized to you, Jeroen? It is in a 16X 3.0 slot according to GPU-Z.
You can try to run
)
You can try to run http://www.techpowerup.com/gpuz/ to check if it's running in x16 3.0 mode.
Hi Mel, I ran some E@H tests
)
Hi Mel,
I ran some E@H tests a few months back with the GTX 670, 31090 driver, on Win7 x64, I5-3570K, PCIe3 and found that:
1 wu, 1013 sec elapsed, 173 cpu, 42500 credits/day
2 wu, 1843 sec elapsed, 313 cpu, 46000 credits/day
3 wu, 2975 sec elapsed, 523 cpu, 43500 credits/day
with 3 cpu tasks & 1 core free to feed the GPU. 2 wu are ~10% lower per unit in both elapsed & cpu whereas the 3 wu were just marginally ahead of the 1 wu costing more in cpu. I'm currently running 2 wu and getting 50000 credits/day total from this host.
Your 3 wu elapsed shows about 10% lower elapsed than my data and probably is a result of your 3770K MB.
Gord
RE: What are the processing
)
The lowest processing times that I have seen with my 680s were in Linux and with the cards in x16 3.0 slots. Two tasks ran at 1220 seconds. With the cards in x16 2.0 slots, two tasks run at 1358 seconds. If I remember right, I was able to run three tasks at around 1800 seconds each but that was quite a ways back. There was little gain running three tasks at once which is why I ran two instead. My 680 cards for some reason run significantly slower in Windows with this project. I have a 680 running in Windows in an x16 2.0 slot and two tasks run at approximately 1670 seconds.
Something I would suggest trying is to disable HT in BIOS and suspend CPU projects temporarily. Check to see what your 670 processing time looks like with these changes.