Boinc 7.6.9 detected the GPU.
Of course i lost all my tasks.

I still dont like it.

Server reply failed.
I need to heat my room now.

---

With each crime and every kindness we birth our future.

Boinc 7.6.9 detected the GPU.
Of course i lost all my tasks.

I still dont like it.

Server reply failed.
I need to heat my room now.

Post your Boinc startup please.

Looking at http://einstein.phys.uwm.edu/top_hosts.php AMD GPU hosts that have CAL support report the CAL driver version, ie: driver: 1.4.1848,
hosts that don't have any CAL support, just report the device, ie: [4] CAL Hawaii (3072MB)

And here Wedge009's mixed host also reports no CAL driver version number for his AMD Hawaii (4096MB) OpenCL: 2.0:

http://setiathome.berkeley.edu/show_host_detail.php?hostid=7123613

Claggy

Boinc 7.6.9 detected the GPU.
Of course i lost all my tasks.

Unless 7.6.9 reported at start up that all tasks were faulty for some reason and it threw them away, it's very possible that they are still there, but just in a different directory. Have you checked that the new BOINC is going to the right data directory?

---

Boinc 7.6.9 detected the GPU.
Of course i lost all my tasks.

Unless 7.6.9 reported at start up that all tasks were faulty for some reason and it threw them away, it's very possible that they are still there, but just in a different directory. Have you checked that the new BOINC is going to the right data directory?

Yes, it is.

I told Richard what i think is the issue.

I got Einstein work and its reporting my GPU as CAL Tonga.
Crunching Einstein to heat my room a bit.

---

With each crime and every kindness we birth our future.

Mike, your problems might be because the "nocal" plan class needs BOINC 7.0.0 as a min requirement. It was tested this way during Beta test and probably the plan class setting has been taken over to main.

See Beta plan classes spec file for details (min_core_client_version).

---

_\|/_
U r s

That file hasn't been updated for over a year - April 2014.

The most recent GPU names, starting with Bonaire through Hawaii, were supplied by [P3D] crashtest in October 2013: names before that were mostly supplied by Jord.

I just sent a new AMD detection list to David.
Nvidia and Intel details are detected from drivers.

---

Boinc 7.6.9 detected the GPU.
Of course i lost all my tasks.

Unless 7.6.9 reported at start up that all tasks were faulty for some reason and it threw them away, it's very possible that they are still there, but just in a different directory. Have you checked that the new BOINC is going to the right data directory?

Yes, it is.

I told Richard what i think is the issue.

I haven't found that yet, but my web access here went all intermittent while I was taking a time out. Only just getting access again.

But I agree with Jord, there's no automatic reason for tasks to be thrown away just because of a version upgrade - you're running anonymous platform, and that hasn't changed since v6.12.34 (I run the same version myself, partly so I can test that our installer-produced app_info files work there too).

Because of the web site problems, I haven't looked to see what your error number was - but you did heed the warning in our sticky thread, about upgrading past the v7.0.28 barrier, I hope?

And I built a new client with it.
Windows 64bit boinc.exe (7.6.12) (MD5 hash eff0ccb1632cf14ccbb53ca97083da35) only.

If someone wants to try it, get it from here (the captcha uses lower case letters, I found out).

That'll download boinc.zip (443KB).
Save someplace you like it, you can scan it with your favorite AV if you want to (I would).
Exit BOINC completely (BOINC Manager and client).
Unzip, save contents.
In your BOINC programs directory (default C:\Program Files\BOINC\) make a copy of your boinc.exe (or write it to an archived zip/7z).
Overwrite with the copy from my zip file.
Restart BOINC.

---

Nvidia ... details are detected from drivers.

Except for that stupid "cuda cores per multiprocessor" that changes with each new architecture release, but isn't exposed in the API - I try to keep David updated with those, and fortunately they don't happen very often.

But v6.12.34 still reports Peak Flops for my Maxwells (GTX 750 Ti) at a quarter of the proper values. Perhaps it's a good thing that we don't actually use Peak Flops for anything important, like estimating runtime.

I just sent a new AMD detection list to David.

Does that list make its way into server code as well? If so, it will be welcome at MilkyWay, where people keep reporting server messages like ati 380 -390 and furyx:

keep getting messages like "ATI GPU R600 (R38xx) does not support OpenCL

https://github.com/BOINC/boinc/commit/5c0ec94544f7ab26748f69b682df39d1d76dc742

---

https://github.com/BOINC/boinc/commit/5c0ec94544f7ab26748f69b682df39d1d76dc742

Just the client, then.

Yes, what else do you expect? It's only the client doing the detecting, nothing else. Unless I am getting very forgetful? :)

---

Yes, what else do you expect? It's only the client doing the detecting, nothing else. Unless I am getting very forgetful? :)

Well, the MilkyWay server (which I suspect has a half-completed server code update, dating from sometime last year), seems to spit out these messages mis-interpreting a 380 as a 3800. But that's for them to sort out. If they care. Which I suspect they don't.

Yes, what else do you expect? It's only the client doing the detecting, nothing else. Unless I am getting very forgetful? :)

Well, the MilkyWay server (which I suspect has a half-completed server code update, dating from sometime last year), seems to spit out these messages mis-interpreting a 380 as a 3800. But that's for them to sort out. If they care. Which I suspect they don't.

I suspect they just have a CAL driver requirement, and since that isn't reported, the server reports that OpenCL isn't available.

Claggy

It's also possible that since these weren't defined by name before, that the client detected them wrong, and then the server will give the warning.

+ case 34:
+ gpu_name="AMD Radeon R9 285/R9 380 (Tonga)";
+ break;

+ case 38:
+ gpu_name="AMD Radeon R9 390/R9 390X (Grenada)";
+ break;

That should now be fixed, but else ask someone there to test with my client (link elsewhere in this thread).

---

Interestingly, the user who complained about the message is reporting "[3] AMD Fiji (4096MB) OpenCL: 2.0" from v7.6.9 - but the word 'Fiji' didn't appear in the client code (apart from geographical references, of course) until tonight.

It's all a mystery, and I'm out of my depth with AMD kit anyway. Which makes it a good moment to sign off and head for bed. G'night all.

Interestingly, the user who complained about the message is reporting "[3] AMD Fiji (4096MB) OpenCL: 2.0" from v7.6.9 - but the word 'Fiji' didn't appear in the client code (apart from geographical references, of course) until tonight.

That's because of this bit of code in gpu_detect.cpp:

 for (i=0; i<ati_opencls.size(); i++) {
ati_opencls[i].description(buf, sizeof(buf), proc_type_name(PROC_TYPE_AMD_GPU));
descs.push_back(string(buf));
}

This reads the processor type name directly from the GPU's OpenCL drivers.
So that's why some are shown as just Hawaii, others as just Fiji, it's the codename for the GPUs. But the specific model number is something that needs to be defined separately.

I think it could be read from the CAL driver part, but without CAL support in the newer GPUs, that's no longer possible.

---

Finally got some work.
The GPU is running at 56°C.

Lets see what i get when i`m finnished optimizing.
This will take some time.

---

With each crime and every kindness we birth our future.

All is running fine so far.

Richard, when i use the installer does it change the plan class ?
Don`t wnat to loose work atm.

Is it possible to use the 6.10 manager alongside with Boinc 7.6. ?
I did it running Boinc 6.12.34.

---

With each crime and every kindness we birth our future.