Only one problem is there. But this could be because of the new app or the longer Deadlines. I have on my single host over 6800 Credits in pending. This is much more than normal (2500 Credits)...
That's probably not due to the app, but the new line of WUs that are rolling out. Each WU needs to be run by at least two computers for anyone to get credit. Since the work is distributed amongst many computers, some results haven't got a chance to be finished and compared with yours. I've got about 11K worth of pending credit at the moment, but it'll all get verified by someone else eventually.
ATI 9600 Pro 128MB external AGP card, running at 8x, with Omega 3.8.421 drivers (which are way more up-to-date than any Catalysts for Windows 2000) and DirectX 9.0c November 2007 update.
Quote:
Does it work on 5.x.x builds?
Does anyone still test 5.10s then? ;-)
Kidding... I don't know. I'm not going to switch back either. I am testing BOINC 6 compliancy. As I said, 4.26 worked as it should, both the beta version and the stock version (I removed the app_info.xml and beta app for a couple of tasks).
Dos anyone still test 5.10s then? ;-)
Kidding... I don't know. I'm not going to switch back either. I am testing BOINC 6 compliancy. As I said, 4.26 worked as it should, both the beta version and the stock version (I removed the app_info.xml and beta app for a couple of tasks).
Hmmm... We gots a pickle then, as you know my feeling about stuff past 5.8.16... :-)
I would guess Bernd will try to duplicate it though...
That seems to be 4000 s slower than a Q6600 with the same clockspeed, but faster than a 200 MHz higher clocked old Opteron Dual-Core 285.
Quote:
The doubled path width is an internal performance optimization, this is nothing that a programmer has to care about "actively" by writing code in a certain way. Any SIMD SSE code will benefit from this.
Well, are your sure? I've read the AMD K10 software optimization guide recently and I can remember that only specific MOV-instructions can be processed within one clockcycle in the K10 architecture.
That seems to be 4000 s slower than a Q6600 with the same clockspeed, but faster than a 200 MHz higher clocked old Opteron Dual-Core 285.
Running Mike Hewson's Ready Reckoner (v7C) over it, I get a predicted average runtime over the cycle of 32611 seconds, compared with my Q6600 at the same speed (it's running at stock 2.4GHz), with an estimate of 23924 seconds averaged over the cycle - details in the dedicated thread below. That's over 8000 seconds difference.
has anyone tested to see if the compiler is putting an AMD penalty in this version?
I haven't found a "Round Tuit" yet... It is on my "to do list"... I was wanting to ride the wave down to the trough, and then make the change to the binary on the way back up. I'm thinking that if there is a penalty, it's not much of one. I could be wrong though...
Alternatively, someone with a performance viewer/tuner/whatever could take a look and see...
RE: Only one problem is
)
That's probably not due to the app, but the new line of WUs that are rolling out. Each WU needs to be run by at least two computers for anyone to get credit. Since the work is distributed amongst many computers, some results haven't got a chance to be finished and compared with yours. I've got about 11K worth of pending credit at the moment, but it'll all get verified by someone else eventually.
RE: Reboot
)
No.
ATI 9600 Pro 128MB external AGP card, running at 8x, with Omega 3.8.421 drivers (which are way more up-to-date than any Catalysts for Windows 2000) and DirectX 9.0c November 2007 update.
Does anyone still test 5.10s then? ;-)
Kidding... I don't know. I'm not going to switch back either. I am testing BOINC 6 compliancy. As I said, 4.26 worked as it should, both the beta version and the stock version (I removed the app_info.xml and beta app for a couple of tasks).
RE: RE: Does it work on
)
Hmmm... We gots a pickle then, as you know my feeling about stuff past 5.8.16... :-)
I would guess Bernd will try to duplicate it though...
RE: We have to wait and see
)
Ok, three of four WUs on a Phenom 9600 @2.4 GHz have finished:
h1_0897.65_S5R3__436_S5R3b = 28,393.84 s
h1_0897.65_S5R3__435_S5R3b = 28,598.41 s
h1_0897.65_S5R3__434_S5R3b = 28,186.84 s
That seems to be 4000 s slower than a Q6600 with the same clockspeed, but faster than a 200 MHz higher clocked old Opteron Dual-Core 285.
Well, are your sure? I've read the AMD K10 software optimization guide recently and I can remember that only specific MOV-instructions can be processed within one clockcycle in the K10 architecture.
RE: RE: We have to wait
)
Running Mike Hewson's Ready Reckoner (v7C) over it, I get a predicted average runtime over the cycle of 32611 seconds, compared with my Q6600 at the same speed (it's running at stock 2.4GHz), with an estimate of 23924 seconds averaged over the cycle - details in the dedicated thread below. That's over 8000 seconds difference.
has anyone tested to see if
)
has anyone tested to see if the compiler is putting an AMD penalty in this version?
RE: has anyone tested to
)
My 3500+ has record fast crunch times with this app. So, no, there's no AMD penalty with this one.
RE: has anyone tested to
)
I haven't found a "Round Tuit" yet... It is on my "to do list"... I was wanting to ride the wave down to the trough, and then make the change to the binary on the way back up. I'm thinking that if there is a penalty, it's not much of one. I could be wrong though...
Alternatively, someone with a performance viewer/tuner/whatever could take a look and see...
I just finished 3 workunits
)
I just finished 3 workunits on my 3 different computers with 4.26 vs. 4.32.
AMD X2 4800+, 1 gb. ram, Win. XP
44411 sec. with 4.26
29551 sec. with 4.32
Intel Quad 6600, 4 gb. ram, Win. XP 64 bit.
30502 sec. with ver. 4.26
22749 sec. with 4.32
AMD Athlon 2800, 512 mb. ram, Win. 2000
56253 sec. with ver. 4.26
40352 sec. with ver. 4.32
Workunits were worth the same credits with each machine.
Speedup is nice it would seem.
While I'm not on a super fast
)
While I'm not on a super fast host it's the fastest I have at the moment :) P4 2.8GHz Hyper-threaded. I run it with SETI at an equal share.
so far here's what I have:
wuid=37046895
h1_0910.35_S5R3__499_S5R3b
88,916.90 Seconds
This one was mixed 4.26 and 4.32
wuid=37047019
h1_0910.35_S5R3__491_S5R3b
76,516.78 Seconds
First time since I started that I have seen a time under 24 hours.