I had experimented with 13.2 beta before I went back to 12.8. I found better run times with r390&12.8 and the CPU usage was lower than 13.1. I might try that again.

And here I thought it'd be a relaxing night...

I had experimented with 13.2 beta before I went back to 12.8. I found better run times with r390&12.8 and the CPU usage was lower than 13.1. I might try that again.

And here I thought it'd be a relaxing night...

I`m sorry.

---

With each crime and every kindness we birth our future.

Feedback on 1764 (I'm still on the 12.8 driver):

Benchmark runs ran ok but when I put it into production I got 3/3 wu's errored out:

One threw an unhandled exception after I stopped BOINC and restarted it:

Unhandled Exception Detected...
- Unhandled Exception Record -
Reason: Breakpoint Encountered (0x80000003) at address 0x767D3219

Maybe a degug breakpoint left in?

The other two gave this message:

<message>
Maximum elapsed time exceeded
</message>

I noticed the estimated computation size was very very low, only about 1000 GFLOPS. Could that have anything to do with it?

Do you have a flop entry in your appinfo ?
If not include it please.

This app generates some binaries into your projects folder.
This takes some time.

---

With each crime and every kindness we birth our future.

Yes I do, it's 100e9.

Yes I do, it's 100e9.

Please free one core for a few units to finnish til it settles down.

---

With each crime and every kindness we birth our future.

Ok I'll do that. What do you think about the unhandled exception?

Ok I'll do that. What do you think about the unhandled exception?

I have no clue.
I can be different reasons.
If it doesn`t appear anymore it should be no problem.
Probably the app didn`t get enough CPU cycles to generate binaries.

---

With each crime and every kindness we birth our future.

By the way Mike, thanks a lot for doing this. I know it's a lot of work.

Here's my report for the four ATI GPUs that I can run the new MB HD5+ r1764 application on.

HD 6970 / HD 6950
HD5+ seems to perform just as well on the Cayman GPUs as the generic r1761 release, and that's fine - no noticeable performance difference so far. Running on Catalyst 11.7, Windows XP.

C-50 / E-450 APUs
The generic r1761 release crashed within minutes / few percentage points on both of these Fusion APUs. Very pleased to find that the HD5+ r1764 version seems to work well, and there even seems to be a substantial speed increase over the previous r426 I was using (completing in about half the estimated run-time). But have only processed one WU each so far, so too early to make a blanket judgement on the Ontario/Zacate APUs.

These are running on Catalyst 12.8 and 13.1, Windows 7 32-bit and 64-bit respectively. I'll try to see if there's any substantial difference between the Catalyst versions on the APUs.

---

Soli Deo Gloria

Dont forget to adjust -sbs switch.
It reduces run times even more.

Values between 156 and 256 seems the best choice on most devices.

Also keep in mind r1764 has much better accuracy than previous builds.

In conjunction with cuda x41zc number of inconclusives should decrease noticeable IMHO.

---

With each crime and every kindness we birth our future.

I had no further errors after the first three which could be due to some instability on my end. After a reboot it ran all night and churned out a stream of good results.

Am happy.

Oh, on the bench the 5 test wu's validated against v8b2 at q > 99%. For r390 q was as low as 95% for some so I do expect fewer inconclusives (assuming v8 is canonical).

Ok I'll do that. What do you think about the unhandled exception?

Exception reason is BOINC itself. IT thinks that app ran more time than it should and aborted it.
"Maximum allowed time exceeded" is the key note to that exception.

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

Oh, on the bench the 5 test wu's validated against v8b2 at q > 99%. For r390 q was as low as 95% for some so I do expect fewer inconclusives (assuming v8 is canonical).

Yeah, r390 is deprecated already as low-precision (relatively new builds) one.

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

Ok I'll do that. What do you think about the unhandled exception?

Exception reason is BOINC itself. IT thinks that app ran more time than it should and aborted it.
"Maximum allowed time exceeded" is the key note to that exception.

And to explain further: BOINC deliberately handles the "Maximum allowed time exceeded" situation by means of throwing that rather clunky exception: it's designed to invoke the debugger, and thus give the developer as much extra information as possible in case the excess run time was caused by some bug under the programmer's control.

In this case, Raistmer has plenty of samples, and I don't think he's calling for any more...

For the record, there doesn't appear to be a noticeable difference in performance between Catalyst 12.8 and 13.1 for low-end APUs on Windows 7 (clean binary caches with each installation). There may still be a substantial difference in high-end GPUs, I don't know.

---

Soli Deo Gloria

For the record, there doesn't appear to be a noticeable difference in performance between Catalyst 12.8 and 13.1 for low-end APUs on Windows 7 (clean binary caches with each installation). There may still be a substantial difference in high-end GPUs, I don't know.

Yes, indeed it is.

---

With each crime and every kindness we birth our future.

Observation: 1761 produced "WARNING: suboptimal workgroup size for PC_find_pulse_kernel_cl pass x" in stderr. From a quick look at the source I think that's just about wasteful memory alignment, shouldn't have much to do with speed.

Actually it can affect performace. Suboptibal workgroup size meand underloading GPU. Could you give link on result with that warning ?

I had tons of those Warnings too, but only on the VHAR's. Changing -sbs 256 to -sbs 128 was the solution for me. Changing the value in -period_iterations_num didn't make any difference whatsoever, the warnings for suboptimal workgroup size, still appeared.

No more of those warnings when changing to -sbs 128 though.

But the app is slower with this setting.
400 seconds running 3 instances on a 0.41 on my 5850.

---

With each crime and every kindness we birth our future.

Yes, i only have a 1GB card.
So reducing -sbs to 156 is required running 3 instances.
~900 MB in use with actual settings.

---

With each crime and every kindness we birth our future.

AMD recently released their 13.4 drivers which comes with a new OpenCL driver, 1124.2.

Has anyone tried them yet with r1760 MB app? I'm still on r390 due to my reluctance to mix and match my driver packages since I was told the OpenCL driver that came with 13.1, 1084.4 was buggy with r1760. Basically it crashed with 1084.4.

I don't do Astropulse so please don't talk about the Astropulse apps to me, this is only about the r1760 MB.

---

"Life is just nature's way of keeping meat fresh." - The Doctor