Yep, looks like this application version doesn't help with the driver issue. Thanks for testing!
BM
OK, I assume you've got enough results for now? I'll switch drivers back again at the next convenient point - got one at nearly 90%, so it may as well finish.
While we're discussing the app for Intel, here's another idea: the hardware with compatible GPUs is guaranteed to be Ivy Bridge, Silvermont or newer. This means they're guaranteed to support all extensions up to SSE 4.2. I don't know if these can speed up the CPU part, or make it run more efficienctly, but you might be able to test this rather easily by just compiling with different flags.
If the code becomes faster, the GPU would idle for shorter "CPU processing intervals". This helps anyone not running multiple WUs.
Or the instruction fusion (1 vector operation instead of several scalar ones) might help power efficiency a little.
I ran a total of 8 tasks on the HD4600 with the new driver - so far five of them have been declared invalid. A good thing we're not allowing Beta apps to validate against each other:
2015-03-29 15:32:16.0436 [PID=32555] Only one Beta app version result per WU (#214692541, re#1)
2015-03-29 15:32:16.0450 [PID=32555] Only one Beta app version result per WU (#214700191, re#2)
2015-03-29 15:32:16.0468 [PID=32555] Only one Beta app version result per WU (#214692540, re#3)
2015-03-29 15:32:16.0494 [PID=32555] Only one Beta app version result per WU (#214367855, re#4)
2015-03-29 15:32:16.0515 [PID=32555] Only one Beta app version result per WU (#213930279, re#5)
2015-03-29 15:32:16.0527 [PID=32555] Only one Beta app version result per WU (#214700054, re#6)
2015-03-29 15:32:16.0538 [PID=32555] Only one Beta app version result per WU (#214699631, re#7)
2015-03-29 15:32:16.0551 [PID=32555] Only one Beta app version result per WU (#214700055, re#8)
On the newest driver it seemed to me that the tasks were completed about 18% faster. Not sure if that means anything. Did two tasks on newest driver before switching back to 3621. 1 invalid and 1 inconclusive at this point.
Higher fan speed with newest driver (Easy to spot on a gaming laptop from MSI :) ) But that is across the board. Also had higher fan speeds with BRP5 before and now BRP 6 Nvidia.
I have run 1.52 on a number of Intel i7-3770 CPU's equipped with Intel HD4000 graphics. Run time varies between 9 hours and 11+ hours. Using older 10.18.10.3621 driver.
Tasks don't seem to checkpoint until around 76% complete.
New driver 10.18.10.4176 from Intel. Did a bunch of BRP4's which seemed to be successful, now trying 1.52 beta app. First task is here
According the Mark's other thread, this one seems to be working for 3rd. Generation processors (Intel's preferred search term for Ivy Bridge). But don't use the matching 4th. Generation (Haswell) v10.18.14.4156 driver for Einstein, unless you find a way of elimination the 100% validation failure I saw when I tested it.
Yep, looks like this
)
Yep, looks like this application version doesn't help with the driver issue. Thanks for testing!
BM
BM
RE: Yep, looks like this
)
OK, I assume you've got enough results for now? I'll switch drivers back again at the next convenient point - got one at nearly 90%, so it may as well finish.
While we're discussing the
)
While we're discussing the app for Intel, here's another idea: the hardware with compatible GPUs is guaranteed to be Ivy Bridge, Silvermont or newer. This means they're guaranteed to support all extensions up to SSE 4.2. I don't know if these can speed up the CPU part, or make it run more efficienctly, but you might be able to test this rather easily by just compiling with different flags.
If the code becomes faster, the GPU would idle for shorter "CPU processing intervals". This helps anyone not running multiple WUs.
Or the instruction fusion (1 vector operation instead of several scalar ones) might help power efficiency a little.
MrS
Scanning for our furry friends since Jan 2002
I ran a total of 8 tasks on
)
I ran a total of 8 tasks on the HD4600 with the new driver - so far five of them have been declared invalid. A good thing we're not allowing Beta apps to validate against each other:
2015-03-29 15:32:16.0436 [PID=32555] Only one Beta app version result per WU (#214692541, re#1)
2015-03-29 15:32:16.0450 [PID=32555] Only one Beta app version result per WU (#214700191, re#2)
2015-03-29 15:32:16.0468 [PID=32555] Only one Beta app version result per WU (#214692540, re#3)
2015-03-29 15:32:16.0494 [PID=32555] Only one Beta app version result per WU (#214367855, re#4)
2015-03-29 15:32:16.0515 [PID=32555] Only one Beta app version result per WU (#213930279, re#5)
2015-03-29 15:32:16.0527 [PID=32555] Only one Beta app version result per WU (#214700054, re#6)
2015-03-29 15:32:16.0538 [PID=32555] Only one Beta app version result per WU (#214699631, re#7)
2015-03-29 15:32:16.0551 [PID=32555] Only one Beta app version result per WU (#214700055, re#8)
- there must be a lot of Beta testers already!
On the newest driver it
)
On the newest driver it seemed to me that the tasks were completed about 18% faster. Not sure if that means anything. Did two tasks on newest driver before switching back to 3621. 1 invalid and 1 inconclusive at this point.
Higher fan speed with newest driver (Easy to spot on a gaming laptop from MSI :) ) But that is across the board. Also had higher fan speeds with BRP5 before and now BRP 6 Nvidia.
@Bernd: is Intel aware of
)
@Bernd: is Intel aware of this problem?
MrS
Scanning for our furry friends since Jan 2002
I have run 1.52 on a number
)
I have run 1.52 on a number of Intel i7-3770 CPU's equipped with Intel HD4000 graphics. Run time varies between 9 hours and 11+ hours. Using older 10.18.10.3621 driver.
Tasks don't seem to checkpoint until around 76% complete.
Link to hosts:
Host A
Host B
Host C
BOINC blog
New driver 10.18.10.4176 from
)
New driver 10.18.10.4176 from Intel. Did a bunch of BRP4's which seemed to be successful, now trying 1.52 beta app. First task is here
BOINC blog
RE: New driver
)
According the Mark's other thread, this one seems to be working for 3rd. Generation processors (Intel's preferred search term for Ivy Bridge). But don't use the matching 4th. Generation (Haswell) v10.18.14.4156 driver for Einstein, unless you find a way of elimination the 100% validation failure I saw when I tested it.
RE: New driver
)
1st task completed, awaiting validation as I type this. Approx. 38.5k seconds, about 4k more than the older 3621 driver.
2nd task running. Link here.
BOINC blog